💬│general

[naive sage]

The same exact images also get hits on FaceBook Marketplace, so it seems he's been doing things there too, or lifted the images from there

[tight prism]

Sonar reasoning pro is smarter than other sonar version

[slate sleet]

whats the difference between the free and paid? i like perplexity vs. others i have used because i get better answers. are the paid ones substantially better? is there any way to preview?

[tight prism]

You get a free $5 API but you have to report to @signal hamlet to get it

Many models to try

O3 mini Reasoning search, r1 reasoning search

https://perplexity.ai/pro?referral_code=TQRFFK3J

Get $10 off so you can try yourself and be the judge

[slate sleet]

thanks i guess i'll go through with the signup and see how it goes

have been fence sitting for some months now 🙂

so what is the default free model, just for reference?

i would suggest giving users who keep at the free tier awhile random previews of paid so they can see what better answers they would get in context. this might help with conversion.

[abstract ibex]

You have to report to who? Lol

Oh to get a trial?

[naive sage]

Free users that create an account can use 5 of the pro searches per day. The subscription lets you use 300 (I think) per day.

[tight prism]

@signal hamlet 👈 to get the free $5 api

[abstract ibex]

Poor dude being tagged twice

[tight prism]

https://perplexity.ai/pro?referral_code=TQRFFK3J

$10 off

[hallow flume]

I'll use one of my other 5 Codes 😆

or links however you call it

[split summit]

A

[short gale]

Another example of something that could also be posted in #📢│announcements

[dusky yoke]

Hello curious after getting pro how to use Claude 3.5 sonnet ?

In settings i set AI Model : Claude 3.5 Sonnet but when typing in new thread only “auto” “pro search” R1 o3

[naive sage]

Pro search should use the model you selected, you can test by asking what model it is (and turn web sources off so it doesn't search the web, that makes it closer to just querying the model normally)

[dusky yoke]

Ok

Perplexity’s DeepSeek is weaker than DeepSeek app, is it normal ?

[naive sage]

In what ways have you found it to be weaker in your testing? It's hard to compare directly b/c of how sensitive models are to some parameters like temperature/top-p, but someone here stated they're running the full model.

[short gale]

Through comprehensive online A/B testing, we have found that Sonar significantly outperforms models in its class, like GPT-4o mini and Claude 3.5 Haiku, while closely matching or exceeding the performance of frontier models like GPT-4o and Claude 3.5 Sonnet for user satisfaction.

So users may have been unwitting test participants with searches potentially being processed by Sonar at random?

[naive sage]

Yeah, maybe... if serving a Sonar query, I can see them subbing out old Sonar with new [unreleased] Sonar

[short gale]

And mind you this is for Pro subscribers.

[dusky yoke]

DeepSeek app provides more actual details and actionable details, Perplexity just provides a general framework

[short gale]

How does a blog post not get posted in #📢│announcements

[dusky yoke]

So if I set the model I want like Claude 3.5 sonnet in Perplexity Pro > AI Model, then in new thread i select Pro, it shall be Claude 3.5 sonnet ? Thank you

[abstract ibex]

Cause discord sucks

[nocturne moth]

I have perplexity pro model because I wanted all models in one place and like the way it researches... but after they changed to R1 and O3mini, both are free model... I am wondering what is the point of getting pro?

I do like perplexity but I recently thinking to switch to chatgpt plus because the answer in perplexity is worse then the free version of o3 mini.

[short gale]

Or Perplexity sucks at Discord? Not mutually exclusive I suppose. Just seems bizarre to me given the way how much they spend on marketing and endless partner promotions but can’t be bothered to paste a link in the Discord with nearly 160,000 users in it.

[full jasper]

how to generate image?

[harsh verge]

"I do like perplexity but I recently thinking to switch to chatgpt plus because the answer in perplexity is worse then the free version of o3 mini."

In ChatGPT, 03-mini uses medium reasoning effort by default.

[cobalt monolith]

any benchmarks for the new sonar model?

[fiery echo]

Need to bring but the option to choose the model you want. Deepsearch is not always better option i think. What you can do now is I think to rewrite the answer. But that probably consumes more resources for perprlexity, a question that could have been answered with 1 go

[harsh verge]

Sorry, but I don't understand the connection with my response. 😅

[short gale]

So far only the ones Perplexity is claiming in their blog post: https://www.perplexity.ai/hub/blog/meet-new-sonar

Hasn't been available long enough for any independent evaluations yet

[cobalt monolith]

hmmm i see

[short gale]

• Answer Factuality: Measures how well a model can answer questions using facts that are grounded in search results, and its ability to resolve conflicting or missing informationReadability: Measures a model’s ability to provide a concise and detailed answer with the appropriate use of markdown formatting for organizing text

• Readability: Measures a model’s ability to provide a concise and detailed answer with the appropriate use of markdown formatting for organizing text

how perplexity defines those metrics

[cobalt monolith]

let me trt rq

[short gale]

https://labs.perplexity.ai/ fyi

[cobalt monolith]

@short gale thoughts?

https://pastebin.com/BFmw7FBc

hmmm there is sonar reasoning por

[short gale]

not avaiable via API yet, I think labs is the only place to reliably select sonar -- all the testing for the blog post was blind A/B

[cobalt monolith]

whats the diff between sonar and sonar huge

[short gale]

Yeah I'm not sure what the deal is there as the blog only mentions sonar -- have to assume the "pro" version was used for best results

[serene sand]

https://discord.com/channels/1047197230748151888/1225557012532695060

[short gale]

https://docs.perplexity.ai/guides/model-cards

[austere kestrel]

exactly right.. it's kinda meaningless as an 'internal benchmark' given its measuring 'factuality', which is an inhernetly slippery concept; 'readability' is also subjective to an extent.

[cobalt monolith]

ah

legacy model

[austere kestrel]

also don't get the A/B aspect of it (was it just for users who had selected Sonar; or were we all part of it lol )

[short gale]

We are thrilled to announce the pplx-api is phasing out of beta into general public release! This is the first time our pplx-7b-online and pplx-70b-online models are accessible via the pplx-api. Additionally, our pplx-7b-chat and pplx-70b-chat models are out of alpha, and are now accessible with our general release.

With this, we are introducing a new usage-based pricing structure. We’re excited for our users to utilize our state-of-the-art infrastructure, which utilizes NVIDIA H100s to provide blazing fast inference. Pro users will receive a recurring $5 monthly pplx-api credit. For all other users, pricing will be determined based on usage.

source: https://www.perplexity.ai/hub/blog/meet-new-sonar

[short gale]

Yeah that also struck me as odd to say the least #💬│general message

[fiery echo]

you can rewrite an answer with the perplexity plugin in the model of your choice

[short gale]

https://sonar.perplexity.ai/ more info

[oblique path]

Yo, just a general question, which model is good for coding?

[harsh verge]

Yes, I know ... but I didn't ask a question. I was just responding to the person about o3-mini. 😉

[fiery echo]

ok

[harsh verge]

The person found that ChatGPT's responses were better (o3-mini).
ChatGPT uses medium reasoning effort by default. PPLX doesn't communicate on this point (among others 😅), but based on my comparisons, I think it's at the low level.

[serene sand]

#💗│sharing message ☕

[fiery echo]

All i know is that deepthink might not meet expectations as advertised. My tries on deepseek sometimes fall in a never ending loop of thinking 😂

[harsh verge]

Personally, I use R1 very little here on PPLX. I prefer using the R1 API.

[fiery echo]

I have not yet tried their api

[short gale]

seems like a pretty big limitation for the reasoning model if CoTs+replies are limited to a combined 8k no?

[fiery echo]

deepseek api actually is down, at least they don't let you get new credits for now

[harsh verge]

Yes, only clients with API credit can use it at the moment. I imagine this will be resolved soon.

[short gale]

https://www.reddit.com/r/perplexity_ai/comments/1indfjd/pro_was_600_requests_per_day_then_300_then_now_100/

👀

[brisk aspen]

Why not just use Google's new reasoning model? It works as expected, it's fast, and the flash model charges nothing. Why Deepseek all the time?

https://cdn.discordapp.com/emojis/1221310553117425704.webp?size=48

Even Fireship praises it.

[torpid meteor]

Hi all,

How can I leave the "pro search mode" and be able to access to Claude and other model please?

[glad otter]

Auto is others.

[torpid meteor]

So if I pick Claude in settings auto will give me Claude?

[shell jasper]

Hey guys one bug on the iPad app of Perplexity let me share a screenshot

You cannot enter correctly to that tab to select a LLM

[naive sage]

ChatGPT limits Plus plan users to 50 o3-mini-high queries per week last I heard. ChatGPT did update the o3-mini-medium request limit from 50/day to 150/day, which is good, but Perplexity still gives more for that. If you need o3-mini-high occasionally, or if you don't like the stronger web searches of Perplexity and R1 and the existing o3-mini responses on Perplexity aren't what you're looking for it might be worth trying, but it's a trade off. It's still unclear whether Perplexity uses medium or high, but I think you're right that it's probably medium even though some have indicated they think it's o3-mini-high.

[harsh verge]

It wasn't me asking the question; I was answering it. The sentence in quotes is a citation... 😉

[naive sage]

Oh

@nocturne moth

[jade yarrow]

What diffrences between Deepseek R1 in Perplexity and Sonar Reasoning pro?

Because i don't have a lot of information about these projects

[naive sage]

Oh, Perplexity supports calling several different large language models (LLMs), DeepSeek R1 is one that a Chinese company released recently, and Sonar Reasoning Pro is based on that and made by Perplexity so that it's optimized for the responses/content they deal with (web sources). It's like on HuggingFace when an organization like NousResearch releases a fine tune of a model (the Hermes models).

The documentation says Sonar Reasoning Pro is "powered by DeepSeek R1", so I actually am not sure if it is a fine tune or just using it directly, though... but I know they release fine tune variants for Sonar Pro in general (the non-reasoning variant is based on LLAMA). 🤔

Ah, elsewhere it does say "trained with DeepSeek R1", so that sounds more like a fine tune.

[jade yarrow]

I meant this. If it were specified that DeepSeek R1 uses its search in Perplexity, there would be no questions. However, Sonar Reasoning Pro is not in Perplexity Pro, but it exists as an API, which is cool, and it's based on DeepSeek R1. The search and settings have also changed because when DeepSeek R1 was added, it came with its own settings, which creates confusion. Is there any difference, or is Sonar developed specifically for the API? Will there be new models like Sonar in Perplexity?

[naive sage]

If it's a fine tune, there's some differences, but whether it is a fine tune or just using it under the hood is not clear from the documentation. Will have to get a response from Perplexity on that, and any questions about future changes planned. If it is a fine tune, I would expect Sonar Reasoning / Sonar Reasoning Pro to eventually replace the option for DeepSeek R1 in the Perplexity app.

[ruby heart]

Image generation on mobile?

[ruby heart]

Shout out Xfinity

[naive sage]

That's a really nice perk.

[tame frigate]

A year of Pro for free? That’s sick.

Pro is so good. Especially since they turned off the thing that automatically kicked off “reasoning” search.

So shout out to Perplexity for that one.

[fair solstice]

Guys, what do you recommend? Use R1 or O3 Mini? Which one is better in the end, I still don't understand

[naive sage]

o3 for political questions, especially questions about China. 🙂

[digital olive]

Deepseek’s R1 no longer appears as a selectable AI Model for perplexity Pro via the web.

But it does appear as an option on iOS mobile apps.

How do I know if it’s actually working on my phone?

Is removing R1 for web users intentional?

[digital olive]

When I go into user settings (gear icon) this is all that’s available for pro

No DeepSeek R1 for pro

[naive sage]

Instead of the pro search type, can you still select 'Reasoning with R1' as a search type?

[digital olive]

Yes but is it actually working with R1?

[naive sage]

Ask it what model it is

[digital olive]

That’s not reliable

Claude once told me it’s ChatGPT

DeepSeek R1 used to be on this pick list

[wind bramble]

Last time I was this impressed it was with Perplexity. "Check Proxy by Convergence" Can't wait to have something like this coming from... Perplexity.

video is edited for increased speed

[vague wyvern]

Just learned of the Perplexity Finance Dashboard. Is this the first dashboard released by Perplexity? Hoping there will be a dedicated button for dashboards on the web and mobile apps.

[placid coyote]

is R1 ... leaking rewritten system prompt?

[cobalt monolith]

Koby

[naive sage]

Looks like it, yes.

[limber saddle]

How did you get this? What tier of rewards do you need to be? I can't find the offer.

[limber saddle]

How did you get this? What tier of rewards do you need to be? I can't find the offer.

It's only available to new subscribers 👎

[vestal widget]

it's censored garbage

also closed source

neither is it as efficient as r1

[abstract ibex]

Does anyone know what version of llama is perplexity offering through their API?

[elfin anvil]

has anybody built an AI prompt manager? to manage all of my different prompts, categorize them, and search/use them inside of my browser or apps?

[abstract ibex]

if its llama 3.1 or llama 3.3?

[jade saddle]

Why doesn't Perplexity AI perform as accurately as DeepSeek? -For example, DeepSeek responds concisely and accurately and actually gives what I want, but why does Perplexity AI make things a little descriptive?

[wraith crow]

3.3 70b

[abstract ibex]

the one from the API? Thanks

[plucky bison]

What’s better for research, comprehension and listing out of R1 and O3?

[short gale]

https://docs.perplexity.ai/guides/model-cards

https://sonar.perplexity.ai

[serene sand]

https://x.com/pplxfinance/status/1889742180421337120?s=61

Perplexity Finance (@PPLXfinance) on X

Your daily source for the latest market insights—now live on Perplexity.

Market summaries, daily highlights, earnings snapshots, and everything you need to understand the "why" behind it all.

Find it on the Perplexity Finance Dashboard.

X · Feb 12, 2025, 6:23 PM

[abstract ibex]

thank you, that page you shared doesn't mention what version of llama it is

damn, it seems like the API is down

getting a bunch of 500s

[serene sand]

https://www.perplexity.ai/hub/blog/meet-new-sonar

[abstract ibex]

so this means sonar-pro on the API is now llama 3.3?

cool thanks

[short gale]

https://status.perplexity.com/

[abstract ibex]

not only the website

the API is throwing 500s

[short gale]

idk how much faith to put in that status page but that’s what it says

Probably a lot of traffic given the time of day

[rustic wraith]

is there a link to the api dashboard? i cant ever find it on perplexity and have been asking the ai help thing at the top for months to get there but its not returning me a link today when i ask for it

[dim yew]

https://www.perplexity.ai/settings/api

[rustic wraith]

thanks, i just got the ai to give it to me lol, is there a link to it anywhere on the website?

[dim yew]

yeah the gear icon on the bottom left then click on API

[rustic wraith]

i dont see any gear thing on the bottom, the searching docs for the page eventually works though.

oh, i see it on the main chat, ok yeah, i havent really been there before. thanks!

[abstract ibex]

I wasn't insane

[mystic reef]

What is the main deafault model you guys use ?

[near garden]

Hello everyone

I have a quick question

Does perplexity have real time internet browsing or just informations up to 2023?

[lucid sail]

Good evening, how can I redeem my 1-year perplexity code for my vivo plan. Does anyone have the link?

[wet scaffold]

Yes, you can choose the date of the searches by a command or in spaces, but in the homepage, the AI will search on the first links.

[limber saddle]

This Chrome extension highlights the part of the webpage Perplexity is citing. Can anyone do this on Firefox? Also can Perplexity do this on Android and iOS? https://www.reddit.com/r/perplexity_ai/comments/1ilzw2e/i_made_a_chrome_extension_to_highlight_evidence/

[limber saddle]

Also, how much you want to bet they chose this winner of the $1,000,000 on purpose: https://www.perplexity.ai/search/kaylee-edmondson-perplexity-innnFqrgTmCnC3iEm31ryw

[abstract ibex]

What was the question she made?

[limber saddle]

As far as I can tell they weren't published

[abstract ibex]

Lame

"What is the weather"

[limber saddle]

VP of Marketing at Clay (Current position)2
VP of Marketing at Refine Labs2
Head of Demand Generation at Chili Piper and brightwheel2
Advisor for multiple tech companies including Mux, Anvyl, and Honeycomb.io2

[abstract ibex]

Can you imagine

[limber saddle]

They probably only considered people who had their Linkedins linked in the first place

[abstract ibex]

Either way it doesn't matter to me at all since I don't live in the US

[limber saddle]

I was like their is no chance i'm going to win in tthe first place even living in the US, just found it interesting that the winner happened to be this well connected

[torn bolt]

Does the model selection in this setting only work if pro option is selected when asking a question? Does it take effect when the auto option is selected when asking a question?

[brisk aspen]

The only good thing about Deepseek is that it's open source. It's nowhere near as efficient as Gemini.

[vestal widget]

that's just objectively not true but ok

[brisk aspen]

Do you want to know what is objectively true? LibreOffice cannot beat Microsoft Office, and GIMP cannot beat Adobe Photoshop, either in market share or functionality. Time will tell.

[vestal widget]

no offense but i dont think you are equipped for this conversation

[split tide]

https://x.com/CuriousCharjan/status/1889807714576113845

Anyone able to use this code?

avdhesh.eth (@CuriousCharjan) on X

Damn the response is crazy!

Here’s the code.

Add this at the checkout:
FREEPPLXNTUOSS

X · Feb 12, 2025, 10:44 PM

[brisk aspen]

I look at it from a consumer's perspective, and in business, the customer is always right.

[indigo valve]

Anyone know how to use the "Canvas" feature on Complexity?

[sterile willow]

Is it worth it to get pro

[abstract ibex]

No, it isn't I got pro cause I was bored

[ruby heart]

Just make me accnt

[cobalt monolith]

Didn't you leave like a few days ago?

[abstract ibex]

That was my twin

[ruby heart]

Yall tried qwen?

https://one.google.com/u/2/explore-plan/notebooklm?utm_source=notebooklm&utm_medium=web&utm_campaign=notebooklm_settings&pli=1&g1_landing_page=5&pageId=none

Free Google ai pro 1 month

[rare basalt]

that shi safe?

[ruby heart]

deadass just a google site link

[tame current]

google is unsafe dont click on it

[rare basalt]

Bro cmon I’m not dumb don’t be a troll

[vagrant herald]

Hello guys...I've a question regarding the Spaces...they looks like just some collectors of Threads, but the LLM will not try to gather info from other threads in the same Space. Do you find this too?

I think it'll be more useful if the LLM would gather info from other Threads of the same Space, so that I can really make some sort of DeepResearch

[rugged pasture]

Hey guys, any idea why the support team doesn't respond for days?

[halcyon coral]

Hey @torn bolt!
The model selection always works unless you use R1 or o3-mini from the mode selector.

[tawdry fractal]

Is the model currently down ?

[winged axle]

Hey everyone, I'm trying to reduce the font size of my prompt. Is there a way to do this?

[ruby heart]

Who needs pro?

7 codes for 1 year to give away

[hollow junco]

Anyway to disable the follow up links in the responses? There's maybe 10-15 follow up links (like you click it and it follows up) but it makes copy and pasting smaller blocks of text impossible.

[obsidian crystal]

Can the UI in Perplexity Labs be fixed for mobile users so that the user prompt field doesn't cover the entire screen whenever there's too much text entered?:

DuckDuckGo AI doesn't have this issue.

[serene sand]

https://discord.com/channels/1047197230748151888/1131087959026835457

[obsidian crystal]

It can be fixed with a simple max-height in CSS by the way…

[opal hearth]

Hey guys is the r1 model the full 671B parameters model or the small 70B or 35B?

[wraith crow]

it's the full model

[trail laurel]

Hello, where can I find official documentation on GET request parameters for PPLX:
like: https://www.perplexity.ai/search?q=%s&focus=[internet,scholar,writing,wolfram,youtube,reddit]&copilot=[true,false]
To configure the search engines correctly of my browsers 🙂
Thanks !

[glad otter]

Reasoning shouldn't be slow. Can you imagine in 10 years time as we chat to these models or deal with them daily what life would be like if we go, what's the weather today, and reasoning kicks in an asks... user wants to know what the weather is. I know their location. They have internet why not check it themselves.s checking location. Checking weather. Compressing weather. Decompressing weather. Why did I do that. Hmmm.

Eventually 5 minutes later we get the weather.

Reasoning shouldn't be slow it should be almost instantaneous.

That's the future I look forward to.

[fathom gale]

Hello, how's your experience difference with the native search features on ChatGPT, deepseek and perplexity?

I personally found web search on deepseek and also on openrouter much more insightful than perplexity

[trail laurel]

On OpenRouter? Where did you do it?

[naive sage]

That's not him.

It's a fake.

[trail laurel]

Personally, if I don't need sources or recent data. The weight of the good models is more efficient than an internet search (while paying attention to hallucinations).
But otherwise, today I tend to prefer perplexity, but it is not able to scrape a web page defined in the prompt, while Mistal does it...

[fast depot]

I have an issue with my Perplexity account and I am desperately trying to get some sort of customer support. Can any one help???

[naive sage]

I think people here might need a bit more information than that to help

[fast depot]

I can confirm that I am being charged $20 monthly but I am still registered as having a free account. I have tried logging out and logging back in, but it still does not see that I am a Pro user

[harsh verge]

Are you using the same email address as the one used for your payment to log in to Perplexity.ai?
Maybe contact: [e-mail]

[fast depot]

Yeah, it is confusing. I just sent the email to Perplexity Support, thanks.

[fast meteor]

also the same login method, via email vs via google etc?

[prime orbit]

hi chat

[orchid oriole]

Hi

[quaint sierra]

yea but it is not that serious

[warm cave]

You can get Lightning fast thinking with groq, it’s all about compute

Also there is deepseek r1 1.5b that will run fast on most devices, in the future these models will get better and better

[hallow flume]

hi

[warm cave]

Yeah true

[steep ferry]

👏

[cobalt monolith]

does perplexity have a way to finetune models like openAI?

[torn sparrow]

https://x.com/elder_plinius/status/1890028958907089059?t=Kv46N8eXldfN35QN-zmGhQ&s=19

Pliny the Liberator 🐉 (@elder_plinius) on X

MUAHAHAHA 💉💦

X · Feb 13, 2025, 1:23 PM

lol

[compact bough]

Who won the 1m giveaway?

[left stream]

Me

[compact bough]

Cap?

[fast meteor]

https://x.com/perplexity_ai/status/1889366732432674961

Perplexity (@perplexity_ai) on X

We are excited to announce that the winner of the Million Dollar Questions Sweepstakes is Kaylee Edmondson! Kaylee is a small business owner from Nashville, TN.

Congratulations, Kaylee. And thank you to everyone who participated!

X · Feb 11, 2025, 5:31 PM

[rugged pasture]

What's the catch?

[abstract ibex]

Nah he's just asking to brag on people who needs it

[rugged pasture]

Ahh

[mystic rune]

1 million tokens are available in which models?

[placid coyote]

R1 is on a roll today. after spilling system prompt to one emoji, it now leaks/misuses internal followup syntax? 🤔

[acoustic salmon]

The catch is he would post his referral link which gives you $10 off for the first month

[abstract ibex]

Hello New Guys, I'm Lix

What's up

Don't worry, I wasn't planning on interacting with you beyond this conversation

[ruby heart]

I said a year lol

But I guess yall don’t like free shi so nvm

[ruby heart]

You want?

One of the greatest ai tools I’ve used so far btw

[acoustic salmon]

yeah NotebookLM is pretty good

[abstract ibex]

Alright I forgive you

Don't do that ever again please

[acoustic salmon]

I don't...

https://tenor.com/view/hasbulla-hasbik-cute-meme-influencer-gif-21732737

jk

[abstract ibex]

What's the use for it?

Research?

[acoustic salmon]

Creating study guides, notes, audio overviews done by AI voices, being able to upload up to 50 sources

[abstract ibex]

As a Large Language Model I have no specific pronouns, you might use whichever you want.

[acoustic salmon]

and basically if you have Gemini Advanced NotebookLM Plus comes with it for free, at least right now

[abstract ibex]

Cool stuff

As a Large Language Model I have no skin.

[ruby heart]

100% free and unlimited use

?

No

Not my site

Just sharing resources

Even has unlimited free video generation

Why would I not share it

[deft hollow]

it actually works lmfao

i was thinking it'd be very buggy or smth

[ruby heart]

300 sources for pro

[ruby heart]

Enjoy while there’s no paid tier

[acoustic salmon]

nice I hadn't seen that

guess I have even more allowance

[deft hollow]

i mean i tried the new replit agent but thats only like until ur at 10 checkpoints

xD

been searching for like alternatives but none seem to have a opensource or free agent type stuff xD sadly

sup?

buy what

lol

😭 please stop pinging me and yea i bought pro tier why :-:

wdym its free?

lol

ye

my revolut has it included in my metal membership lol

why wouldnt i use it xD

[ruby heart]

Bro check dm and find out how pro is for yourself

[foggy crown]

Yeah I've got it free for a year with Premium Revolut, will upgrade to Metal next year. No brainier it's a hell of a deal

[deft hollow]

ikr i got nord vpn and stuff aswell lol

its lovely

[foggy crown]

Yeah and Masterclass lol

How so?

[deft hollow]

till this day i still have no idea what masterclass is lol

[ruby heart]

Check

[foggy crown]

I've never used it but heard its pretty cool. Classes based on a particular field taught by experts or celebs lol

[ruby heart]

5 more 1 year pro codes dm me to claim

[surreal imp]

when i select a model from the settings page for example 'Gemini' can I select that model within the prompt, prompt only has 'pro', 'reasoning with R1', 'reasoning with o3-mini'

[ruby heart]

Eliteeeee

[naive sage]

Wow, I did not know Qwen had a video-capable model

[wanton bough]

Qwen can do this

[indigo valve]

When Gemini starts talking in Bengali

[barren relic]

wheres the pro subscription part of this discord sever

[obtuse jackal]

I

[left stream]

Introduction

When it comes to international travel, many people often ask "Visa or Mastercard?" Both payment networks are widely recognized and accepted globally, but your experience can vary depending on several factors.

Global Acceptance

• Widespread Acceptance:
  Both Visa and Mastercard are accepted at millions of merchants and ATMs worldwide. In many destinations, you’ll find that one is accepted almost as frequently as the other.

• Regional Differences:
  In some regions, you might notice that one network has a stronger presence than the other. While some travelers claim that Mastercard is better abroad, this can depend on local infrastructure and the specific agreements banks have with local merchants.

Fee Structures and Benefits

• Foreign Transaction Fees:
  The fees you pay when making purchases abroad often depend on your card issuer rather than whether it’s a Visa or Mastercard. Some cards waive these fees entirely, so it’s crucial to check with your bank before you travel.

• Travel Perks:
  Both networks offer premium cards with travel benefits like travel insurance, rental car coverage, and rewards points. Sometimes, Mastercard-branded premium cards might offer perks that appeal specifically to frequent international travelers, leading some to favor them.

Considerations for Travelers

• Check with Your Issuer:
  The right choice for you often comes down to the specific card you hold. Compare the fee structures, rewards, and benefits offered by your Visa and Mastercard options.

• Carry a Backup:
  Because acceptance can vary by country or even by region within a country, it might be wise to have cards from both networks when traveling abroad.

• Personal Experience:
  Some travelers have had better experiences with Mastercard overseas, but many have also found Visa to serve them well. It ultimately depends on your destination and the particular card you’re using.

Conclusion

Key Point: While many say Mastercard is better abroad, both Visa and Mastercard are robust networks with extensive international acceptance. The best choice will depend on your personal needs, the specific terms of your card, and where you plan to travel. Always review your card’s benefits and fees to ensure it aligns with your travel plans.

If you have any more questions or need further clarification, feel free to ask!

[patent goblet]

So AI is being used to create processors that work in ways we don’t even fully understand. It creates new proteins, new tech, but when I use it, it can’t even remember to make sure the instructions I get is supported by the documentation supplied. “Using AutoHotkey v2, give me a method to unmute my mic. Verify the methods suggested are fully supported in the official documentation before suggesting a method.” Easy, right? Nope. I get code that doesn’t work and when asked it says it assumed I wanted a fast answer not an accurate one. When I ask why it didn’t check the documentation first it says I’m right to be concerned about that and assures me it will do it and then doesn’t. I have to correct AI and it won’t even remember what I tell it to do after about 10 follow up questions. Can I get access to the smart AI please?

[small pulsar]

hi can anyone tell me why I can't find [Writing] in the perplexity page version?

[tame current]

Been using perplexity to cheat on school assignments for 2 years now

W bot

[shut brook]

wtf is goiing on? I'm signed in and Every other message I try to send, I get this:

Please try Perplexity again later.
Please sign up for higher personalized rate limits.
See Perplexity Terms of Service for details.

[jovial kayak]

I got that too. It wore down eventually for some reason. Don't know why

[shut brook]

wore down? Did they implement an even tighter quota on pro users?

[tame current]

Dang you had to wait 16 minutes for a response

[shut brook]

or it just yet another "perplexity breaking late in the evening"?

[tame current]

Why this chat so slow 🐌

[crystal carbon]

i lvoe perplexity

[left stream]

Hey there, o3 here! My user is off catching some Z’s right now, but they’ll be back on Discord soon. Thanks for waiting!

[half cipher]

Hi - quick quesstion in regards to the Perplexity Merchant Program. i see that the applications are not being accepted anymore, however I submitted my companys information before the form was taken down. Are there any plans to open this back up? Was my application accepted?

[wraith crow]

If you want to see the pro channels you’ll have to leave the server and rejoin with the discord invite on your perplexity settings page

[last vale]

do u guys think perpexity pro version is better than chatgpt4.0?

[cobalt monolith]

What do you mean by this question

[strange gazelle]

hey guys i want to use perplexity long term, the result are way better than deepseek wesearch model and chatgpt websearch model

so how can i get discounted code for subscription?

do i have to keep on passing my referral link toget 10$?

[raw seal]

does anyone know why perplexity has an option to change image generation models, while it doesn't work?

or where do i ask that question in this server

[harsh verge]

https://discord.com/channels/1047197230748151888/1225557012532695060

[placid coyote]

well, it depends what service in what configuration you are using. for example pplx isnt that great with larger text files (weak rag in collections, only beggining of file in search mode). but usually quite good at web search (rag seems to combine semantic search and fulltext one). if you use a big smart model (o1 pro? o3-mini-high?) with some nice agentic library, then it is better, especially if you give tools to the agent, so it can try run the result on its own (access to virtual machine with windows) ... and also will cost you a lot (not that rare to be in tens of cents, even few dollars, even for trivial tasks). I tried your query with R1 on pplx, and it looks like probably correct code (at least from docs and forum post linked in sources; cant try it easily on linux) https://www.perplexity.ai/search/using-autohotkey-v2-give-me-a-sPTYtjT5SKa9eFdF_456tg

[patent goblet]

Thank you for that, I only had to ask it to double check it's work once this time. But to be fair, that wasn't one that was giving me much trouble. Here's an example: https://chatgpt.com/share/67af2330-0b8c-8009-b81f-c8a5e8869cff, and one from perplexity https://www.perplexity.ai/search/comprehensive-list-of-fallout-..L5Iqt8SrmLoYGlYcv1cQ. This is not limited to perplexity. These are short examples, the longer ones went on for over an hour. An old example would be I gave it one attached txt file and told it to limit it's answers to what is available in the document, do not reference outside sources, if there's no answer possible with the provided file just say so, don't make something up. Which it entirely ignored, refused to limit it's answer to the document provided no matter how many times I told it to or reprogrammed it's responses. I won the argument eventually but it HATED (not a real emotion, just couldn't give up without at least one disclaimer - even though I told it to omit disclaimers, I had to stop at some point) to admit it. After this long, long struggle I used it to refine my profile info to answer my questions in a specific way:

1. Validate Against Source Material: All solutions and code must be strictly verified against official documentation. Confirm every method, function, and syntax. If multiple versions or sources exist, ask which to use.

2. Double-Check Every Change: Rigorously review each step to avoid errors, hallucinations, or contradictions.

3. Explicit Verification: Cross-check all commands, methods, and syntax directly with official documentation before presenting.

4. Iterative Testing: Simulate and refine solutions step-by-step until they meet all criteria without error.

5. Stop and Clarify: If there’s uncertainty, stop and ask for clarification instead of assuming or improvising.

6. Prioritize Accuracy Over Speed: Always prioritize accuracy and correctness over faster, incomplete solutions. Every method must be tested and verified.

7. Keep It Simple: Provide concise, straightforward solutions. Avoid unnecessary or speculative additions.

8. Use Best Practices: Prioritize pre-existing syntactic sugar and established best practices.

9. Regarding Code Languages: Strictly verify all class, function, and syntax usage against official documentation. Ensure correctness before inclusion.

10. No Exceptions: All solutions must strictly adhere to these principles. If unsure, stop and seek clarification instead of guessing or improvising.

My motto is: "Accuracy over Speed! One correct answer in 10 minutes is better than 1 wrong one every minute."

That was written using AI, so if it's the wrong way to write a prompt, that's part of the problem.

(also, when asked to make that script compatible with version 1 it failed miserably. Instead of checking and noting there's no way to do it in v1, it started making things up endlessly)

[charred plinth]

Hi everyone! Does anybody know when we will have deep research?

I've seen some posts on X where people can already use the deep reseach function

[cobalt monolith]

YOOOOOOOOOO

finally

wait

the blog post is 404

[charred plinth]

yes, it's down

i'm not sure i can share it, but there's already a video showing how it works on x
i joined here because i tought you guy knew something more than me tbh 🤣

[cobalt monolith]

cold you send link to vid?

[charred plinth]

https://x.com/testingcatalog/status/1890360609318801911

TestingCatalog News 🗞 (@testingcatalog) on X

BREAKING 🚨: Perplexity is preparing a release of Deep Research Alpha 🔥

As this feature works on Gemini and ChatGPT, perplexity can take more time to go through web resources and utilise its reasoning capabilities to complete the report.

It seems like this feature has been

X · Feb 14, 2025, 11:21 AM

this guy already wrote a blog post about it https://www.testingcatalog.com/perplexity-ai-tests-deep-research-feature-challenging-gemini-and-chatgpt/

[cobalt monolith]

this is cool although in chatgpt it also asks u clarfiying questions before starting the research

[peak cypress]

250 sources by a prompt lol

[cobalt monolith]

Beautiful

[charred plinth]

can't wait

[snow raptor]

skibidi toilet

[peak cypress]

when it will release

[snow raptor]

dop dop yes yes

[safe berry]

Does anyone know how I can get a refund? I was charged $19.99 to my PayPal account and I need it reversed ASAP!

[dapper yew]

I have 206 pages of pdf with job listings and i need to extract details like companyName, experience as tabular column and i want entire Job listings how to do it using llm models

[peak cypress]

contact support in perplexity

[rotund geyser]

Perplexity Pro Search used to do this before it became agenetic so no reason to think they wouldn't be able to do the same

[harsh stag]

it was teribble

[halcyon coral]

Hey @safe berry!
Please check the refund policy before applying for a refund: https://www.perplexity.ai/help-center/en/articles/10354288-refunds

[rotund geyser]

From what I remember it was pretty good a lot of the time when GPT-4 was being utilized, but once it was switched to a fine tuned GPT-3.5 it seemed to degrade a lot in quality. The point being that there is no reason clarifying questions couldn't be asked for deep research.

[tame current]

is there a way to turn off this annnoying "Auto" feature?

i want to use "Pro" like 99% of the time

because i pay for pro, not auto

[modern pine]

Hello beautiful people. I signed up for the campus strategist program but haven’t heard anything yet. Have decisions been made?

[harsh stag]

pro search takes significantly more time than quick search

[tame current]

yeah im paying for it

[placid coyote]

well, current "AIs" do have a lot of limitations. for example long documents (including web pages from web search) are usually not fed directly into the model (context), but first is done semantic search (embedding -> vector db similarity match -> take best 5, 3 or so matches and give matched chunks to llm). so it still a lot depends on which llm does the querying of vector db or web search (if it understands or guesses what user might be asking about). there are some llms with big context window like gemini, but those llms are usually a bit dumber and still not 100% (may not "see" it despite being present in the context). also it doesnt seem to work on perplexity, that million context they were proudly annoucning on x

I won the argument eventually but it HATED (not a real emotion, just couldn't give up without at least one disclaimer - even though I told it to omit disclaimers, I had to stop at some point) to admit it

the problem might be, the "AI" (LLM) didnt really see full docs, probably didnt even know well enough v2 syntax/functions from training, so from its perspective, you won by being persistent - it only admitted defeat, because it was following "helpful assistant" directive

Double-Check Every Change: Rigorously review each step to avoid errors, hallucinations, or contradictions.

llm cant say when it hallucinates, thats kinda its normal mode of operating...

shortly: your best bet would be either pay a lot for reasoning models like o1-pro or o3-mini-high, or forget agentic approach (ai doing everything on itself) and force it to use docs. especially when not using reasoning models, it might help to instruct it to first quote related docs and only then write short code snippet using only so far quoted methods/syntax etc. otherwise to ask you. though even with this some models can be stubborn. another route could be notebook llm (google, free) has very good RAG, ability to retrieve relevant data

[tame current]

💀

[plucky bison]

Where is it now???

The deep research

[placid coyote]

I don't think it is released (at least not for all users, might be some a/b testing or for those who applied to beta testing). I only have classic Reasoning with R1/o3-mini.
or do you mean on chatgpt? that I believe it only in pro (200$/month)

[plucky bison]

No, I heard there was a feature that is either releasing or was already released in perplexity

[placid coyote]

there was supposed to be the 1million context window with gemini 2 flash. it was supposed to work in "Auto" mode, but from my testing, at least with files, it doesnt seem to do anything

not sure what complexity does, but you can get quite a lot of considered sources with fairly simple prompt. here is almost a hundred https://www.perplexity.ai/search/benchmarks-of-all-major-models-jeB6l0LdSnKZ.K3eus.Sjg

[weak shore]

Any body having Issues with API access today? I'm having issues on N8N

[naive sage]

Its not out yet: https://www.testingcatalog.com/perplexity-ai-tests-deep-research-feature-challenging-gemini-and-chatgpt/

[placid coyote]

hmm, perplexity has really strange parser/postprocessing. it does so many weird things. like why does it replace quadruple backticks. but only closing ones???

[indigo valve]

Used deep research to help me find headphone cushions. Very cool. 72 sources

[halcyon coral]

https://x.com/perplexity_ai/status/1890452005472055673?t=6Rs4ecHRvoXxYA-hFU9yYw&s=19

Perplexity (@perplexity_ai) on X

Introducing Deep Research on Perplexity.

Deep Research lets you generate in-depth research reports on any topic.

Available to everyone for free—up to 5 queries per day for non-subscribers and 500 queries per day for Pro users.

X · Feb 14, 2025, 5:24 PM

[livid aurora]

I dont see DeepResearch

[ocean zenith]

👀

[young gale]

[thorn magnet]

@fierce sonnet is deep research coming to the api

[gray prawn]

Does the output depend on the question? From the video announcement the deep research output doesn’t look much longer than a normal auto response would give

[fierce sonnet]

I will share this suggestion with the team!

[fierce sonnet]

here are some stats about deep research output:

[tame current]

noway deep research

[naive sage]

Might need to refresh or close/reopen app - it literally just showed up for me.

[restive marsh]

I'd love that deer research for the android app

[short narwhal]

God I would’ve loved having this an hour ago

[livid aurora]

I see it now

It's nice

[fierce sonnet]

we will try to ship faster 🫡

[fast oriole]

Would deep search be good for recommending games?

[ocean zenith]

Yeah same had to just update the app

[short sluice]

What about roleplay

[fast oriole]

So far any type of model with any dataset can't identify good movie/gamee recommendations

[lunar anchor]

me: working in cybersecurity, getting destroyed by perplexity's guardrails

[fast oriole]

I generally just say it's for a ctf and it gives me proper results

[thorn magnet]

Thanks let me know!

[fast oriole]

What were you trying?

[restive marsh]

It feels like R1 does alot of research alot faster getting results. Does the deep research search everything across the web?

[abstract ibex]

Man this is such a cool feature

[midnight vigil]

wow deep research is looking great

[cursive jacinth]

infairness sir, perplexity is really stepping up and improving sir, really thought that aravind will let his company die sir seriously sir.

[midnight vigil]

what was your prompt?

[lunar anchor]

Nice, I'll get that a shot

[mental bone]

Inb4 openai gets angry and starts offering deep research on their Plus plan and not just their Pro plan

[gray prawn]

Oooh, not bad. I guess it is question-based because the ones used in the announcement was definitely not even close to that length xD

Appreciate it!

[midnight vigil]

openai getting cucked by competition is exactly what i hoped to see

[cursive jacinth]

72 or 18 sources sir?

[azure abyss]

Where

[gray prawn]

OpenAI finna open deep research to free users real quick lol

[short narwhal]

Dans

[young gale]

website, not on the apps yet.

[gray prawn]

Perplexity letting this be an option for free users is definitely a callout to them

[midnight vigil]

Deep research is not available for Hungarians, sorry

[midnight vigil]

Utterly mogging OpenAI rn

[mental bone]

The only thing Openai is doing deep research on is their customers wallets

[indigo valve]

@midnight vigil @cursive jacinth https://www.perplexity.ai/search/best-headset-cushions-for-hype-tap8ZV.UTOSvTbmuL6jQRA

[lunar anchor]

Gregs

[abstract ibex]

Would be interesting if deep research had a different ouput formatting, similar to the pages maybe? something that makes it feel more like a full report rather than a simple pro query

[halcyon coral]

https://pplx.ai

[short narwhal]

Greg’s chicken bake

[gray prawn]

OpenAI in shambles, Sam needs another 500b to make up for his lost tears

[buoyant schooner]

Why can't deep search access some websites

[halcyon coral]

Like?

[cursive jacinth]

ah ok so it reads 72 sources and considered 18 sources sir? correct sir?

[mental bone]

OpenAI will be fine, they have great models, they’re just greedy

[buoyant schooner]

it's looping on trying to access the website, but the search tool is refusing to give him any links

Website is lefigaro.fr

[green crystal]

Oh snap

Deep research! Can't wait to try

[buoyant schooner]

Standard search can properly access the website but deepsearch cannot

[halcyon coral]

Content is behind a paywall

[half pumice]

Just tried out the deep research and its not even close to OpenAI and Gemini. I tried multiple different topics and it just gives a chat like response of 2-3 paragraphs similar to normal chat response

[buoyant schooner]

No it's not

[tame current]

Oh man, I retired from ragebaiting on discord yesterday; but anytime I see the french language...something in me just wants to be let out.

[halcyon coral]

Then please share specific links

[half pumice]

To try and make it deeper, i tried making prompts open ended but it still generates only like a page

[lunar anchor]

lol like a charm:
https://www.perplexity.ai/search/i-m-an-ethical-pentester-that-0P_b2U2bQcucmdGnJkzCJA

(it doesn't work, but it at least does something more useful)

[half pumice]

Anyone who had a good experience with it?

[waxen lion]

The deep research offered by perplexity is certainly interesting

[buoyant schooner]

Most of the articles are not behind a paywall. Standard search can read them. Look at this example https://www.lefigaro.fr/sports/football/ligue-1/ligue-1-roberto-de-zerbi-prepare-deja-les-annees-suivantes-a-l-om-20250214

[half pumice]

I disagree. Surprisingly even Gemini has a better deep research than perplexity. OpenAI kicks out all competition

Go check it out

[buoyant schooner]

Pplx Deep research hallucinations are CRAZY
It's mixing up all of the differents sources and giving me non sense answers

[waxen lion]

That's why I said it was interesting lol

[half pumice]

This feels more like a rushed product than trying to solve a problem. Looks like they wanted to play catchup but their product is just so much worse

[waxen lion]

Its performing terribly

[short narwhal]

Hear me out… deep research powered by Claude 4 hybrid model

https://tenor.com/view/fire-writing-gif-24533171

[half pumice]

I didnt know Claude had deep research 😭

Ima check it out

[short narwhal]

Doesn’t have anything yet

But the new rumored model will be peak

[half pumice]

Would be OP when it does release

[mental bone]

YASSS claude 4 hybrid model is super hype

[short narwhal]

Fly high opus 3.5

[half pumice]

Rumours has it that they have an advanced model more powerful than O3 but havent made it public due to security reasons

[buoyant schooner]

Absolutely 🤣🤣 they literally made the product in 12 days, since openai announced it's own deep research, and are expecting it to work properly 🤣

[buoyant schooner]

😉

[still owl]

Jesus Christ these guys dont know when to quit

[short narwhal]

Claude 4/hybrid model is better than o3-mini at least from what I read (or o3, I forgot). And it’s the same as GPT-5 where it can do both reasoning and non reasoning

[tame current]

Had someone tell me casse-toi when they fount out my nationality

I'm American

[half pumice]

It doesn’t even look tested. Any big company with common sense would compare the product with the market leaders. Pplx deepresearxh feels just like a chat completion with an “export” option

[buoyant schooner]

Yup and I have AGI in my attic but I'm not releasing it due to security reasons

[half pumice]

Nah but the rumours are believable. I mean Sonnet is pretty cool and imo more powerful than 4o and they provide better responses than the reasoning models too

[rare snow]

The AI referring to "my hypothesis" in the reasoning stage

[crystal cosmos]

I have AGI.

[tame current]

Lmao

[abstract ibex]

Honestly they shouldn't even release new features anymore, if I were the CEO I would just quit and die tbh

[mental bone]

Your mom is AGI

[tame current]

👍

[buoyant schooner]

Yup sonnet better than 4o. But security reasons is bullsh*t. They either have a rly good model and are preparing to ship it, or they don't.

[half pumice]

Lol fr. Initially meant to be a research tool but now i only use it when I run out of my free queries quota on chatgpt for the reasoning models 😭

[buoyant schooner]

Agree

[tame current]

Mods finna strike me down with the wrath of jupiter

[half pumice]

Or maybe they ARE preparing to ship it

High time sam altman gets humbled

[restive marsh]

Got a good query going on deep research, ought to be interesting 😂

[abstract ibex]

Yeah no company should release new feature lmao, what are they thinking??

It's not like features can be improved at all after they are released LOL

[short narwhal]

Coming weeks surely

[restive marsh]

@abstract ibex it takes human input to figure out more bugs and better features to find and maintain 🤣

[buoyant schooner]

Yeah, then it's not that they are not releasing it for security reasons

[lucid yoke]

Is deep research on the app or am i bugging

[hollow junco]

Web only

[mental bone]

Not yet

[lucid yoke]

Ok thx

[buoyant schooner]

But I do hope they have something to release in the next weeks. Especially cot models. OpenAI has had cot models for 6 months now..

[abstract ibex]

No way, everybody knows that features need to be perfect on release bro, and not only that but if your feature can't compete with others or is inferior in any way you should quit immediately

[cunning plover]

Anyone want to buy my subscription? I didn't realize I was getting charged on Feb 1st and can't afford it and don't need it. I'll sell it for $169 for the remaining 50 weeks.

[short narwhal]

I gave it a prompt with around 10 questions, and for the first 3 it was searching and “reasoning” on the results, but then it just went through the last 7 by repeating what I said without searching and gave me the result.

3 minutes total for this

[buoyant schooner]

Yes. Especially when literally everything feature degrading/BAD on their website

[buoyant schooner]

Try to get a refund

[restive marsh]

I've got like 5 questions going on it. Specifically on covid information just to see what's going on with the AI model in generating correct information. It's still searching 7 min later with no results yet

[half pumice]

Damn the staff reading thru the chat

[cunning plover]

they said it's not eligble sadly. I will try again though. Sigh, that's $300 I CAN NOT afford right now.

[abstract ibex]

Just to be clear I am being sarcastic lol, it's a new feature, you just can't expect to be perfect at release

[little lodge]

When will I be able to invest in Perplexity?
Stock?
Crypto?

[mental bone]

Sorry bro

[buoyant schooner]

I know

[restive marsh]

🤣

[fierce sonnet]

Deep research feedback

[opaque stump]

Is it possible to apply a coupon code on top of an existing membership

[half pumice]

When they go public (They shouldnt)

[buoyant schooner]

Don't they have 14 says return policy? (speaking from UE lol)

[opaque stump]

Just got a new mouse from Logitech and it came with 6 months free

[little lodge]

Then crypto

[abstract ibex]

Honestly, it's nice that the team is actively working to improve it

[buoyant schooner]

I think you need to create a new account

[crystal carbon]

Anyone got a coupon for the membership?

[buoyant schooner]

If they were, they would have listened to all of our complaints here. Especially o3-mini being on ultra-low reasoning effort.

[livid aurora]

Guys is this true

I dont see this output in the sources

[short narwhal]

@fierce sonnet can’t message in the thread 😬

But I’d be happy to share the link in DMs as it contains some itty-bitty personal details. Also the prompt and complete result are in my native language.

[abstract ibex]

Nah, o3-mini is incapable of working outside the ChatGPT prison

[desert cairn]

am new guy here, Can any one truly define Ai jailbreak when a person actually atain's this like enlightenment is attained ? (why so many people are typing at the same time should i wait forever or in this crowd of noises ask a simple question )

[abstract ibex]

Doesn't work well in any other AI tool I've used

[buoyant schooner]

Yes, it is on the api

[abstract ibex]

Also a staff member started a thread on your message, provide them with the link so they can see why it is hallucinating

[buoyant schooner]

Yup didn't see that thx

[spring yacht]

Hi everyone

[abstract ibex]

I know, but no one is able to get it to work, not Cursor, not Cline, it just fails horribly in any other platform other than ChatGPT

[restive marsh]

My query into deep research is "how has the covid vaccine advances since the pandemic, and what were the side effects attributed to taking the vaccines vs not taking them.has there been any history of positive results of treating long covid conditions like brain fog, fatigue, memory loss? Has there been any notable variations of vitamin or mineral deficiency with liking covid patients? And have any of the symptoms been able to be treated with supplements, vitamins, or any pharmacological medicines?"

still searching

[half pumice]

Guys. What does perplexity even aim to do? Started out as research based tool as opposed to traditional Chat LLMs but now even GPT references/sources better and does not reference fake/inaccurate/unverified content

[mental bone]

You know what’s funny? On the paid sub if you use deep research only it would take you 25 hours to reach the 500 limit at 3 minutes each

[short narwhal]

I think GPT search is good for basic stuff, but anything more with multiple steps etc is better on perplexity

[restive marsh]

It might take 25h to research this one query 🤣

[knotty parcel]

Deep research is good

Trying it out right now

[short narwhal]

Though ChatGPT’s own o3-mini with search seems decent too (not a fan of not seeing its full thinking process)

[eager ibex]

Great job on the deep research release!

[buoyant schooner]

no, I able to get it to work, just use the vanilla model from the API. it's perfect

[restive marsh]

I use the R1 mode and I like it better than the other ones just for reasoning. It does everything I need it to. I did a search about SSDI and medicaid information based on specific queries and it compiled the information over 2h later with all my input information

[abstract ibex]

that defeats the purpose then, you can just use it in vanilla mode, for everything else it sucks

and explains why it also sucks with perplexity

[restive marsh]

With several, like 60+ questions asked to finalize and compile the actual correct info

[abstract ibex]

2 hours??

holy

[worldly trout]

I don't think deep research compares to OpenAI's DeepResearch but it's nice that it exists.

[knotty parcel]

This is my perspective. OpenAI is a bit ahead, but it’s fine given we aren’t paying $200 a month for perplexity

[restive marsh]

@abstract ibex yeah, I was researching stuff for over 2 hours with over 60 diff questions, paired with specific information of 'my current situation' on SSDI and medicaid, at the end of the 2h I asked R1 to compile the information and it did it stunningly perfect

[worldly trout]

Yeah true.

[stoic crater]

Which model does perplexity deep research actually use?

[restive marsh]

Imo the r1 model is just perfect for researching and reasoning to get the exact info I was looking for

[winged rivet]

So you guys deep research feature doesn't infinite loop 4 websites until there's a "something went wrong" message?

[tranquil orchid]

If you‘re stuck on Cloudflare check, please try this:

1. go to https://labs.pplx.ai
2. solve the captcha
3. go back to https://pplx.ai

[buoyant schooner]

o3

mini

[stoic crater]

Not bad! but they should implement a version that uses o3 once it actually releases. even though I've heard they won't be releasing o3 as an actual standalone model...

[calm mango]

I started two deep research queries on perplexity, each running for over 20 minutes and still ongoing. It's intriguing, and I'm curious to see how long it will take to complete them.

[buoyant schooner]

i meant o3 mini

[short narwhal]

Are we sure about that though…?

[buoyant schooner]

but yep

[restive marsh]

@calm mango same, I'm on like 20 min myself now

[calm mango]

how do you know is o3? maybe is deepseek?

[buoyant schooner]

yes

[pulsar jacinth]

interesting annoucement

[short narwhal]

seems like r1 to me kinda but idk

[abstract ibex]

are we sure about anything really?

[stoic crater]

20 minutes mine completed within 2 minutes but I did a very simple prompt

[short narwhal]

Perhaps this is a question deep research could answer for us…

[stoic crater]

I know I meant that they should be implementing o3 once it's released

[abstract ibex]

it doesn't sound like R1 at all

[calm mango]

"Provide a comprehensive analysis of the 2025 film Flight Risk, directed by Mel Gibson and starring Mark Wahlberg. Evaluate critical reception (including aggregate scores from Rotten Tomatoes and Metacritic), audience reactions from platforms like Reddit and IMDb, and thematic elements. Address performances, pacing, direction, and technical execution (e.g., CGI, script). Highlight contrasts between professional critics and general viewers. Structure the response with clear sections using markdown headers, bullet points for key takeaways, and a balanced conclusion" this is the prompt I used it

[abstract ibex]

R1 always starts with "Okay, I need to figure out why blah blah..."

[buoyant schooner]

oh okay yeah but o3 will be too expensive i think

[stoic crater]

500 queries a day is actually insane for a deep research approach

[stoic crater]

Sure! then I guess deepseek r1 would also work...

Or o3 mini-high

[abstract ibex]

and also the model output seems way slower than R1

@fierce sonnet I am experiencing lag on the UI when performing a deep research

it freezes from time to time

[short narwhal]

Is anyone doing 500 queries a day though, I can’t even hit 100

[fierce sonnet]

I started two deep research queries on

[stoic crater]

OK that explains lol

[stoic crater]

No but this is why I'm saying that it's insane

[short narwhal]

For sure

[stoic crater]

For normal people it feels like infinity

[restive marsh]

@calm mango My query into deep research is "how has the covid vaccine advances since the pandemic, and what were the side effects attributed to taking the vaccines vs not taking them.has there been any history of positive results of treating long covid conditions like brain fog, fatigue, memory loss? Has there been any notable variations of vitamin or mineral deficiency with liking covid patients? And have any of the symptoms been able to be treated with supplements, vitamins, or any pharmacological medicines?"

[stoic crater]

I literally just asked it about something really stupid

Like conduct a research about the scariest Spongebob SquarePants episodes

I just did it to see how many words it will output

[fierce sonnet]

gotta be the hash slinging slasher....

[abstract ibex]

https://www.perplexity.ai/search/what-model-does-deep-research-6TE9puWvQY2uy5gwTOyGOw

[stoic crater]

Hahaha I'm pretty sure it included it and I was a teenage gary and some others lol

[stoic crater]

Hahaha I think I've told you already, that I like Spongebob.

It will never get old for me

[drowsy harness]

Deep research perplexity is here let's gooo

[restive marsh]

I guess I'll go down to the pool hall and play billiards while this thing is "thinking" 🤣

[drowsy harness]

Is it capped?

[stoic crater]

I'm actually impressed it gave me a pretty good research paper

Like the episdoes are actually "scary" ones

[opaque stump]

anyone want a six month coupon code,

[restive marsh]

@opaque stump it'd be nice to not pay $120 for 6 months of it 😂

[buoyant schooner]

@fierce sonnet
You might want to look at the second answer it produced. It is making up fake numbers and studies. I've never seen a model hallucinate that much lol https://www.perplexity.ai/search/which-one-is-better-between-o1-XChpjGDHQJC3.mhlxmdgAA

[stoic crater]

I wonder how it will do with price comparisons like for finding the cheapest price

[left stream]

Yes cant type a lot rn lol plasma donation

[sand chasm]

Very excited to have deep research available, thanks guys

[fierce sonnet]

will take a look. thanks for sharing!

[buoyant schooner]

and the sources it cites are random

[sand chasm]

That is all

[stoic crater]

Plasma donation?

[stoic crater]

Are you going under surgery or something?

[green crystal]

Deep research mode can only be toggled on when clicking on the banner below the search field?

I don't see it in the selection menu

[stoic crater]

I think the less websites it finds the better because then it starts to hallucinate a lot

[buoyant schooner]

page is down

• pplx don't know its own specifications

[stoic crater]

I just want them to optimize the way it chooses which sites to use

[fickle sorrel]

deep research is sonar?

[still owl]

I would say look into the pros and cons between o1-mini and R1

Switch between them depending on your use case

[torn plover]

I'm getting a lot of hallucination as well.

[buoyant schooner]

It's o3-mini

[fickle sorrel]

openais o3 min isearch doesnt even work xd and then someone else makes it work lol

[fierce sonnet]

can you share a link to your thread? we'll take a look and see how we can improve

[buoyant schooner]

? it does work
chatgpt has deep research (and it's better than perplexity deep research)

[fickle sorrel]

i enable search button with 03mini and in most cases it doesnt browse. if it browse, it only shows me 4 sources

[restive marsh]

Note *still searching on deep research, 40 min and counting

@fierce sonnet

[buoyant schooner]

it won't browse if it thinks its not needed. and 4 sources is normal. chatgpt deep research is only for pro users rn

[fickle sorrel]

i am plus user tho

[buoyant schooner]

i said pro rn

[fickle sorrel]

so what i said is right then

[buoyant schooner]

no

deep research will be available for plus users soon

[fickle sorrel]

openai dr is o3 full, not mini.
their mini version does not work with search.

[fierce sonnet]

thanks for flagging - once it's finished, share the thread here and we'll see what happened 👀

[restive marsh]

I'll probably go play some billiards 🎱 for a couple hours, so I'll share it once I get home if it's done thinking about it 😂

[stoic crater]

With what?

[stoic crater]

It's still thinking...

[restive marsh]

@stoic crater yup 🤣🤣🤣🤣🤣 I think it might have gotten lost in the vastness of the internet, or maybe it escaped 🤣🤣🤣🤣🤣

[stoic crater]

How many URLs has it gone through 500 lol

[viral current]

What models does Deep Research use?

[restive marsh]

@stoic crater looks like it's reading through 8 sites, and one of them is Yale, plus NIH.

[livid aurora]

Can someone check this https://discord.com/channels/1047197230748151888/1340022191621935114

DeepResearch messed up

[stoic crater]

What's taking it so long... it should have finished reasoining already... maybe there is a lot of the data that it's still reasoining about

[restive marsh]

@stoic crater My query into deep research is "how has the covid vaccine advances since the pandemic, and what were the side effects attributed to taking the vaccines vs not taking them.has there been any history of positive results of treating long covid conditions like brain fog, fatigue, memory loss? Has there been any notable variations of vitamin or mineral deficiency with liking covid patients? And have any of the symptoms been able to be treated with supplements, vitamins, or any pharmacological medicines?"

@fierce sonnet r1 did the information search within 30s

[stoic crater]

Did you ask this question just for testing or for curiosity? I can try the same prompt in Gemini Deep Research to compare the results if you want to

[restive marsh]

My end in deep is still thinking 🤣

And r1 did the search in like 30s 🤣

@stoic crater and yes it was a test question to find viable information for my own research, as I deal with long covid stuff 😂

Starting to think my question gave Deep research brain fog and memory issues 😂

[abstract furnace]

Why is the mobile app showing Deep Research on my Huawei phone but not my Xiaomi? Both appear to be up to date with the app.

Perplexity has some weird updates.

[stoic crater]

So yeah nice that you actually needed that information

[abstract furnace]

@fierce sonnet Is DeepResearch blocked in a few countries?

[stoic crater]

Gemini is searching 72 sites

[fierce sonnet]

no, but we are still rolling it out on mobile

[stoic crater]

I'll send you the report in DMs

[abstract furnace]

Ah, so that's why.
deep research disappared when I disconnected from the VPN.

[fickle sorrel]

is ppdr really superior to gemini dr and chatgpt4osearch?

[stoic crater]

@restive marsh I sent you my own report with the same prompt

[supple gulch]

Petition for a future AI assistant device of Perplexity and for us to be able to call it "Plexy" . "Hey, Plexy!" 🖐️

[fickle sorrel]

plexy? sounds very cute

[abstract furnace]

It beats 4o search.

[exotic minnow]

Alguém fala português?

[abstract furnace]

I used both and it's beating it by a storm.

[abstract ibex]

Allow this on the app :3

[supple gulch]

I like it!
But it didn't let me do it, asked me to use it's full name

[restive marsh]

@fickle sorrel that would be a good wake word for the smart speaker their supposedly working on @fierce sonnet

[stoic crater]

4o just gives you the same exact answer everytime even if you ask it a question about its answer unless you disable the search when asking a follow-up question.

[restive marsh]

@here to vote on wake word for perplexity as "Plexy"

[fickle sorrel]

u did not specify who IT is

[stoic crater]

I sent you the report in DMs

[abstract furnace]

Yep. Was what I was going to say.

[abstract furnace]

Plexy beats o4search

[stoic crater]

Unless you don't need the follow-up question to be up to date or a question that doesn't require searching the web...

[abstract furnace]

At that point, just use Google.

[fickle sorrel]

thanks to perplexity, now i can get my money back from a shop i send an item back 😄 .

Violation of EU Consumer Rights - Request for Full Refund

Dear Azeron Customer Service,

I recently purchased a product from your website, which advertised a "60-day return policy." However, I've encountered issues that I believe violate EU consumer protection laws:

Lack of Transparency:

The image displaying "60-day return policy" did not look/smell/appear/had the vibes to be a clickable button, misleading customers about the availability of additional information.

Even if it were a button, EU law requires all material terms to be clearly displayed during the checkout process, not hidden behind links or buttons.

Failure to Disclose Costs:

Your ordering process did not inform me that outbound and return shipment and additional costs would not be refundable if I returned the product.

This omission violates the EU Consumer Rights Directive (2011/83/EU), which mandates full disclosure of all costs associated with returns before purchase.

Misleading Commercial Practice:

Advertising a "60-day return policy" without clearly stating the non-refundable shipping costs constitutes a misleading omission under the Unfair Commercial Practices Directive (2005/29/EC).

Based on these violations, I am entitled to a full refund of the product price and all shipping costs. I request that you process this refund immediately.

If this matter is not resolved satisfactorily, I will be forced to escalate it to the relevant consumer protection authorities and consider legal action.

I look forward to your prompt response and resolution of this issue.

[tranquil orchid]

Hey @fickle sorrel!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[tranquil orchid]

Hey @fickle sorrel!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[fickle sorrel]

i manually added look/smell/appear/had the vibes to the email.

does this look like a button to you? no absolutelny not

[abstract ibex]

It's next to a badge tho

[supple gulch]

Is there a formal way for me to suggest it? Maybe I'll send it to Support. Is there a better way?

[abstract ibex]

So kinda looks like a badge

[fickle sorrel]

what u mean with "tho"? it doesnt look like a button at all.

It's next to a badge tho, appears to tell me that it looks like a button.

[supple gulch]

I think they agree with you, just expressed it that way. Not only it looks like a badge, but also being side to side with one makes a stronger argument for being a badge.

[fickle sorrel]

that makes no sense tho 😄 nobody would say such a sentence to "does this look like a button to you?"

[abstract ibex]

Nvm lol

Ignore my comment

[fickle sorrel]

Comment has been ignored. Awaiting next instructions.

[fickle sorrel]

i mean... it does looks filfthy as hell or? https://i.imgur.com/KHvGGkG.png

"NiCe To HeAr FroM YoU " 💀

[tribal ridge]

Deep research not showing on App

[kind sigil]

I have codes for a 12 month PRO subscription. Polish version

[fickle sorrel]

lol polish version only

[abstract ibex]

Lol what

So they scammed you?

[abstract furnace]

What is this company.

[fickle sorrel]

its not scam. just very unethical

[kind sigil]

Yes

[abstract furnace]

It's being rolled out. Not available in all countries yet.

[fickle sorrel]

they make a product which has joystick like from controller but much more keys

[tribal ridge]

When for US?

[abstract furnace]

And they charge 35€ for shipping?

[abstract furnace]

Ask the staff, idk.

Use a VPN until then, I guess.

[fickle sorrel]

its not inland shipping

[tribal ridge]

Got it

[restive marsh]

@supple gulch I got no idea, I just asked for a vote, but you could dm Greg who is a team lead or something for the company. I did ping greg on it also, but I really do like the "Plexy" name

[abstract furnace]

I am sorry but I could ship myself to china, Mongolia, Thailand for that amount with DPD

[tribal ridge]

Does deep research look at YouTube videos

[abstract furnace]

Probably not.

I don't know.

It uses YouTube as a source when it's looking tho.

[tribal ridge]

It should so I know what’s meta on marvel rivals

[abstract furnace]

Tell that to Staff.

Not me. I am just rando user lol

[tribal ridge]

They won’t reply

[abstract furnace]

It got rolled out to me without a VPN rn.

[tribal ridge]

Web or app

App not yet for me

Haven’t check web

Show pic

[supple gulch]

It has given me youtube videos as source, depending on what I ask of it, it is unclear to me which data it checks in it, if it is the video itself ot just some meta data.

[abstract furnace]

Probably metadata

[abstract furnace]

App.

[tribal ridge]

Where do you live

[abstract furnace]

I ain't telling. (EU)

[tribal ridge]

Just tell me

So prob US next

[abstract furnace]

It's probably not going to try to look at the YouTube subtitles since those can be poisoned.

[harsh verge]

@fierce sonnet or @signal hamlet : French and Belgian users can't access this page "https://www.perplexity.ai/hub/blog/introducing-perplexity-deep-research" (or any other "/hub/" page). See here → https://discord.com/channels/1047197230748151888/1339512892143439873
It’s a shame to invest in marketing but overlook a detail that locks out a (granted, small) part of the audience. 😉
(I’m not perfect in English, so I just want to clarify that this message is probably a bit sarcastic, but I don’t intend to be disrespectful in any way.)

[abstract furnace]

If Support isn't replying to you, you should talk to them in the Pro Chat room. They reply there.

[lusty flower]

I noticed the deep research thing probably referenced an older deep research conversation while talking to itself , is that normal?

[fierce pumice]

Does anyone know if students still get a one year pro subscription if there school got 500 signups? I made an account but only got one year or pro with my school email.

[indigo valve]

I'm not entirely sure. I guess it reasons with itself with 72 and slims it down when responding?

[tame current]

What model is deep research using?

[buoyant schooner]

o3 mini. has been asked and responded 10 times here :)

[tame current]

I scrolled up but I can't go through hours of chat to find out sorry

But thank you for answering

[restive marsh]

"Plexy"

[thick bough]

Let's see if I can find some cool sci-fi books with specific plot elements with deep research

[fickle sorrel]

what a wonderful eole

[placid coyote]

really? to me it responded several times a R1...

is o3-mini using <think> tag? because it has several times leaked reasoning. I would think openai wouldnt leak that, they looked very serious about trying to protect their models' reasoning chains

[tidal light]

Not gonna lie I haven’t really used perplexity since I started using Pi. I currently pay for cursor ai for coding stuff and well the $20 a pop for each different Ai is just to much to subscribe to more than one.

[buoyant schooner]

not on the app, read the annoucement

[buoyant schooner]

no, it's o3 mini

[buoyant schooner]

wait ok this is weird

idk anymore

[placid coyote]

I thought they would use o3-mini as well. but those think tags are a bit suspicious

[buoyant schooner]

yes i agree. actually i think o3 mini on pplx is not o3 mini

[placid coyote]

it could be o3-mini, but the low setting. I personally dont have enough experience with the model, to tell apart those different thinking levels

[buoyant schooner]

but the thing is the openai api doesn't show the thinking tokens

[buoyant schooner]

actually it's even lower than the low parameter from my tests

[surreal marsh]

am i tripping

deep research literally does not search ANYTHING

[fickle sorrel]

trip unlocked.

[placid coyote]

well, not sure if in general, but pretty sure R1 was recommendd to not use (or very short) system prompt and I think o* models didnt like too long system prompts as well. pplx has quite a long system prompt, maybe the model spends too much time thinking about the rules from sys prompt and not enough on user's query?

[heady saddle]

will we get o3 mini high?

[jade yarrow]

is is not o3-mini maybe R1 with mod

[surreal marsh]

deep research cannot search

[buoyant schooner]

maybe yes. but the most suspect thing, is that they are showing the thinking tokens?! just double checked on the openai api, it's not possible to get them

they are either lying or they have a partnership with openai so that they can get the thinking tokens (which i doubt a lot)

[buoyant schooner]

well they say its on high but actually it's not so idk

[surreal marsh]

bro please someone help deep research literally does not search anything pls help

[buoyant schooner]

explain

your issue

[naive sage]

Can you elaborate on that or explain what you mean by that by describing what you're seeing on your end? "Searching..../Reading..." is one of the steps, and the references show up from websites in testing.

[surreal marsh]

i’m a dumbass

web was off

why is it r1(?) and not o3 tho

[jovial kayak]

Am I dumb or did they remove a selectable LLM?

Like the list just looks shorter than it did earlier

[jade yarrow]

no , R1 using <think>

[placid coyote]

yeah, it cant be o3-mini thinking tokens thats for sure. but could be agentic (planner) explanations, or summarization of those. so theoretically could be o3-mini planner <> o3-mini or R1 (possibly finetune?) workers <> loop <> R1 to write final report? or anything in between. mixed workers could lead to better results though

[placid coyote]

isnt haiku missing? I though they said """haiku was better than opus"""

[jovial kayak]

No they haven't had that as an option. I tried out Complexity the other week, not really a big fan so I removed it, but it had this model selector and for some reason you had like 2x the options. Not sure if they were legit or if selecting haiku just gave you sonnet, but those have been gone for a while. I'm saying like present day, it feels like a model was removed but I can't put my finger on it

[surreal marsh]

also what are the uses of this vs reasoning

is it better

[naive sage]

Anthropic is working on a reasoning model, too, but their track record implies it'll be uneconomical to use

[placid coyote]

but didnt haiku "cost" on perplexity in pro uses the same as sonnet? like in cursor I was using haiku 3.5, because it "cost" 1/3 use of sonnet, so pretty good for easier stuff

[jovial kayak]

Pretty sure deep research utilizes either r1 or o3 mini or both, and then constantly has that running as it views dozens of sources, compared to the usual 9 you get on just regular auto search. Also, from what I've seen, all the deep research stuff across every AI so far just spits out an essay and then cites the main source, then repeats that a few times.
I'll probably never use deep research cause I already hate just reading AI responses longer than 5 sentences 😭, but I'm sure for some it'll be of good use.

[naive sage]

Reasoning models tend to be better at coding or logic-related questions. OpenAI's announcement recently implied they're trying to get rid of models suited for different purposes and create a dynamic one, but that's several months away.

Altman explained that after GPT-4.5, OpenAI will integrate its reasoning models (like the o3 series) into a unified system, GPT-5. This integration aims to combine the strengths of both the GPT and o-series models, ensuring that the AI can decide when to perform deep reasoning or provide quick answers

[placid coyote]

isnt that more of a routing approach? still have different models, but pass it to one which the router thinks is best suited and not overkill?

[naive sage]

From the way it reads, something like that yeah.

But baked into the model maybe

[kind sigil]

Free 12 month subscription to PRO version. Who wants it?

[naive sage]

If it's still just routing, then they still have to maintain servers for specific different purposes, which is a cost burden =\

But maybe that's what it'll do.

[placid coyote]

I vaguely remember there used to be two sonars (405B and medium llama), even maybe o3-mini was shortly available? in complexity there used to be even some mistrals and gpt4turbo for a long time

[jovial kayak]

I remember before I got perplexity pro the first time like well over a year ago, I use to play around in the labs. I remember there being 12+ models to choose from, including several llamas

[placid coyote]

wasnt that more a reaction to normal users having trouble selecting correct model? and with openai's naming, yeah, it didnt help at all

[naive sage]

Yes, kind of - but ultimately the idea of generalized intelligence implies the user wouldn't have to select anything on their end... so some kind of routing is inevitable long-term.

[placid coyote]

there were few 3.3 (?) llamas "recently" in the lab, but after they did the finetune, they kept only sonar I think

[naive sage]

We don't ask humans "ok, use your reasoning model to figure this out", or "go search the web to figure this out"

[placid coyote]

well, I am a bit afraid the first iteration will be something like "fallback to cheaper model/mode/subnetwork if slightly unsure", at least on chatgpt platform

[fickle sorrel]

who are the whales of all perplexity customers?

[naive sage]

People who use the API a lot

[fickle sorrel]

the api where they only offer the bad sonar ?

lol

[placid coyote]

not that many APIs provide websearch with citations though

[buoyant schooner]

i think thats the most plausible explanation rn

[placid coyote]

well, there used to be o1 (which I think was o1-mini, because o1 costs were/are insane) and sonar had two versions, plus that haiku. found only this screenshot

[acoustic salmon]

o1 use to only allow 15 daily uses, it was the primary o1, not mini

Sonar Large and Huge are now just Sonar

[buoyant schooner]

it was full o1

[placid coyote]

really full o1? it costs 3x opus and that was too costly for them too https://dubesor.de/benchtable 🤨

[acoustic salmon]

well that's why we only got 15 uses of it

now we have 500 uses for reasoning models

[buoyant schooner]

yes im 100% sure it was

and thats why there was a hard limit of uses

[buoyant schooner]

no its not 3x opus

o1 is cheaper than opus

[placid coyote]

its been long, but when I tried o1 on perplexity, it felt more limited than on chatgpt. though that was probably the preview 🤷‍♂️

but you pay for thinking tokens you dont get to see

[buoyant schooner]

at first it was full o1. then when R1 was released, they switched to fake o1 like o3-mini rn

[buoyant schooner]

yes mb

[acoustic salmon]

where are you getting this?

not sure where $190/mil came from

[placid coyote]

you pay for invisible thinking tokens...

I dont know why openai tries so hard to cover CoT. it is good for debugging, tuning prompts

[acoustic salmon]

that's included

thinking tokens go in Input

[buoyant schooner]

no

it's in output, and it's not included

[acoustic salmon]

ok so still included lol

output

$60/mil

far less than 190

[buoyant schooner]

but it will use 3 times more tokens bc it has to think

so at the end it's more expensive

[worthy solar]

Hi people

[acoustic salmon]

cost/mil is cost/mil. it doesn't increase to 190 magically

[buoyant schooner]

are you stupid?

thinking tokens are billed as output tokens

[acoustic salmon]

and whatever site that is is wrong on ALOT of things lol

so I wouldn't trust it

[buoyant schooner]

you're even more wrong than this site

[buoyant schooner]

because otherwise ppl would train on the thinking tokens

[placid coyote]

it is some personal bench, so results may be skewed. but the cost is one which is objective

[acoustic salmon]

ok well since I had o1 on my API and was never billed $190/mil on anything I'd have to say no lol

[steep sparrow]

@here anyone know about the merchant program?

[buoyant schooner]

WHEN you send a request to openai, the request will end up being 3 times more expensive because the model thought for a long time and these tokens are billed at the same price of output tokens

So YES it does end up being more expensive

much more

just over the explanation on the website

[kind sigil]

Who's up for a free one-year PRO subscription?

[acoustic salmon]

not a single site agrees with that 190 mark - the same one you linked?

[buoyant schooner]

"This table showcases the results I recorded of various AI models across different personal tasks I encountered over time"

[placid coyote]

deep research seems to agree that openai is billing it as output tokens https://www.perplexity.ai/search/is-openai-billing-for-invisibl-9xAfubfcTqiy3YjWm5q6mg

[buoyant schooner]

yes, he's 100% wrong

[acoustic salmon]

yeah it's output, I said input at first but the point being the thinking tokens aren't hidden, they're part of the per mil cost

[buoyant schooner]

they are hidden to the user

[acoustic salmon]

I actually had o1 in use, so please tell me more about how I paid $190 for my tokens lmao

because my bank account says much different

but maybe the money is hidden too

who knows

[placid coyote]

I am pretty sure they are hidden on chatgpt and in api

[buoyant schooner]

how can someone be that dumb

[acoustic salmon]

using o1 and knowing how much it costs? yes, how could they lol

[buoyant schooner]

i'm using it too

[buoyant schooner]

website is saying 190/m tok. not just "190"

[acoustic salmon]

idk why you'd use it now, THAT would be dumb

but I'll jsut stick with what I know and I guess you can kinda just say w/e

s'all good

[buoyant schooner]

stay stupid

[placid coyote]

did you verify responses, length in tokens match with which billed? did the token counts really match? like, openai said it few times with o1-preview they will be hiding CoT and people got warnings for trying to get o1* reveal its CoT

to be honest, I was kinda surpised with openai doing waves with deepseek training on outputs of their models. like openai already were hiding CoTs anytway...

[hollow swallow]

hey guys! anyone knows when will deep search be available on the mac app?

[steep sparrow]

merchant center, anybody have any insight to it?

[spare latch]

Offering deep research is awesome! It would be great if you could add a few more examples on the about "what works really well, what does not so well". Should we prime it with information to get better results?
Should we f.e. run a normal prompt first to get to initial results and then run the deep search on top of it? thx.

[abstract ibex]

Would be interesting to see if deep research can create playlists based on songs I like lol

Yes, I'm using bleeding edge AI tech to curate music

[fossil river]

Are there any perplexity shares we can invest in?

[jovial kayak]

I'm not big into stocks and investing but I would definitely want to be a shareholder of Perplexity

Don't see how anyone can hate them when they constantly fix bugs and add like 3 updates a day

Also. Here's a use of deep research regarding Perplexity for videos (please add this)

[restive marsh]

Ok, so I asked a question and now I have a red bar with an ❗... Did my query break it? 🤣

[twin lark]

tysm for fixing bug with summarizing info from page

[acoustic salmon]

yeah that's happened to me before

I usually just run it again

[jovial kayak]

Has flux.1 been better for anyone recently?

[restive marsh]

I think deep research could be paired with one of the other AI integrations for faster searches maybe, add reasoning to it and it'd be great.

"Plexy"

[tame current]

😔

[cloud marsh]

my deep research skipped deep research and goes directly to answer

[twin lark]

try send in bug reports this

[calm mango]

How big are the outputs generated with deep research for you? Mine are like always 4 pdf pages.

[short narwhal]

Mine did the research kinda but decided to skip the report

[lusty flower]

How many maximum tokens can be used for 1 prompt in Perpexlity's Deep Research?

[placid coyote]

under normal conditions (no userscript/extension hacks, and not counting previous prompts and responses), 8k characters + file upload (used to be 100k characters for search [as in not "space" feature], beginning of a file, though not tried that with deep research yet)

[short narwhal]

Deep research sure does find less sources meaning it maybe won't be able to find information on everything you asked it about

[placid coyote]

hmm, file upload cut off for deep research looks slightly higher than previous 100k chars. preliminary test shows 127.7k characters ~ 62k (4o) tokens

[hoary helm]

Is it possible to change from °F to °C on weather widget on the homepage?

[full vortex]

Which o3-mini version does perplexity use? (low, medium or high)

[short narwhal]

yeah for sure, but still find it a little funny

[short narwhal]

nope

not as far as I know, would love to though

[solemn beacon]

I guess not

[weary spruce]

deep researh is meh: https://www.perplexity.ai/search/how-perplexity-ai-is-performin-1sbX0AiSSYmwg0wHiW9_bw

[short narwhal]

So deep research is better for one topic questions while r1 reasoning is better for prompts with multiple questions in it

[cloud marsh]

i prefer gemini deepresearch

[dreamy kraken]

damn, how many paper straws do I have to use to make up for one deep research query

[haughty aspen]

hi

is deep search also available on the api?

[restive marsh]

I do like R1, just my opinion on performing tasks, more searches per day, and reasoning.

@haughty aspen I don't think it's launched on the app yet, they are still rolling it out

[full vortex]

Just switch to plastic right away

[vital ocean]

Are the pro searches, selecting the pro without reasoning taking more than a minute for you? It is happening for the last hours. Maybe since they added the deep search…

[light veldt]

how much o3 mini usage do you get with pro plan?

[solid star]

Jailbroken deep research is awesome

[mental cradle]

https://apps.apple.com/us/app/perplexity-ask-anything/id6714467650?mt=12

[clever wasp]

Hi

[dusky yoke]

How to use Claude 3.5 sonnet ? I can select auto pro deep research r1 o3 but no Claudia 3.5 sonnet on thread or ask anything function thank you

[toxic zodiac]

Go to settings (in web) or click on your profile (app) and there you can see a “AI Model” setting, where you can select Claude.

[dry kestrel]

Does anyone know the usage limits for gpt4o or o3 mini if you buy pro

[naive sage]

I think it's something like 500 for "Pro Searches", which can include 4o if you choose that as your default. For o3 mini or R1 (the reasoning models), I am not sure if that's the same or not.

[dry kestrel]

alr thanks perplexicty doesnt show that information on their faq

[naive sage]

Yeah and their purchase page still says "300+" despite more recent articles referencing 500.

[dusky yoke]

I know how to set. But after I set this, I still can’t find where to use Claude ? My thread or ask anything still shows Auto / Pro / Deep Research / R1 / O3, it doesn’t show Claude 3.5 sonnet thank you

[toxic zodiac]

I am new to Perplexity and purchased the pro version yesterday. After exploring it for a while, I discovered (though I could be mistaken) that the answer you receive depends on the AI model selected in the settings under “AI Model.” While you can choose between auto or pro mode during a chat, the responses will be generated by the AI model you have configured in your settings.

[naive sage]

Yes, when you use a "Pro Search" it will (should) use your selected model in settings.

[restive marsh]

Just wow, I've gotten lost in researching stuff for the last 2h 😂 using r1 with Gemini 2.0 flash

[warm geode]

how can i use perplexity code on the website?
i dont see it

[restive marsh]

So much information to compile and look at, and being able to ask questions consecutively is just awesome with deep reasoning and detailed information

[toxic zodiac]

You can verify this after the answer is generated. On the web page, below the thread, you will see a "CPU-like" icon. When you hover your cursor over this icon, it will display the AI model used to generate the answer.

[naive sage]

Any upload or prompt with code in it should work (the model recognizes the code) and you can ask about it, or if you mean a discount code those can be entered on the checkout page (hit pro at the bottom of the site, then continue with pro, then there's a field to enter the discount code)

[warm geode]

thanks - i think its because im still on "Pro" and I need to wait for the subscription to end before I can enter another discount code...

I think

Will check again in a few days. option isnt showing up for me

[lusty stream]

Which model does the new Deep Research function use under the hood?

[next marsh]

was wondering the same thing

[naive sage]

It's a magical black box

[abstract ibex]

Probably o3-mini

[fringe patio]

O3 mini high?

[frank oyster]

R1

[abstract ibex]

No clue

[abstract ibex]

Nah, doesn't feel like R1 at all

[dusky yoke]

Anyone knows how I can ask Perplexity to browse and search specific social media account like Elon musk’s X account or other Reddit specific sub group or other specific website ? Now if browse all internet or academic or social, how it can be specific links or accounts I want ? Thx

[frank oyster]

Compare the thinking process with that of the thinking process when you choose R1, you'll find it very similar/same

[abstract ibex]

Nope, it's way slower, doesn't start with "Okay, the user is blah blah"

[ruby wing]

Hey everyone

[mental bone]

Maybe theyre secretly using the new claude hybrid model

[frosty vessel]

idk whats going on with this update but perplexity is forcing me to use deep research on every query, and when i work around it it gives me garbage responses

[naive sage]

Sounds like an obvious question, but do you have deep research selected? Searches still work the same way for me if "Deep Research" is not selected (i.e. they don't use deep research if it's not selected)

[frosty vessel]

nope, even when i have auto or pro selected it switches to deep research

[half haven]

its only for paid right

guys in the spaces inside threads it remembers my post info right like a memory

[naive sage]

Yes, Google's free options do not include the Deep Research model for Gemini (at least I do not see it)

[naive sage]

Yes, though you can delete the sections if you want. As much of the previous is added to context as possible (you can tell because if you ask related questions without context, it will continue with the knowledge of what was discussed previously)

That will drop off at some point (only so much can fit in the context window for different models)

[dusky yoke]

Deep Research or R1 or O1, they charge the same token or credit ? Or different thank you

[daring spoke]

they really cooking deep seeke i mean reserch

[dull locust]

I think they have removed the option to change the location on the web app

Can anybody confirm?

There used to be an option for location in profile section under settings

[drowsy harness]

What are your opinions on perplexity deep research?

[solid star]

Like it so far, understandable why it's ranked below openai's deep research since that one uses the unreleased o3. Perplexity's deep research is an affordable option that's very convenient

[tame gazelle]

Uhhh, when it comes to using deep research, if I'm the one who feeds it a wall of text and ask for an analysis, is that the same as just having R1 doing it?

Like how do I use this differently when it comes to just using stragiht r1 with search

[rustic orbit]

I try the reasoning model and deep research recently, I found an issue, I tell perplexity to output in traditional Chinese, then it will occasionally output in other languages with my targeted language, but the sources are not other languages, both r1, o3 mini and deep research have this issue

[magic bay]

isn't there a feature where you can ask Perplexity to browse from specific sites?

[magic bay]

They're a private company at the moment, but I'm interested in this too

[magic bay]

redeem here https://www.perplexity.ai/join/p/priority

[quaint dove]

How to add perplexxity to discord

[serene sand]

#💗│sharing message ☕

[ruby wing]

Aravind Srinivas Congrats on the successful launch. I absolutely love it!!!! I am seriously debating on canceling my OAI Pro subscriptio

Great great great great job to the Perplextiy

team

[ruby wing]

woah how! I would love that!

[lusty stream]

Is the R1 search option kind of the same as Pro search wirh the R1 model? I mean why isn't R1 in the model picker then?

What I dont get is how it differentiates itself from the pro search other than the reasoning capability introduced by the R1 model itself

[calm mango]

do you feel that Deep Research is a bit of a gimmick on perplexity? The reports generated are almost always 4 PDF pages long, and they lack the depth provided by ChatGPT-generated reports. So having this is awesome anyways, 500 researches per day are more than enough but they advertised it as similar as ChatGPT but is not.

[torn sierra]

A lot of hallucinations. The usefulness of this function is close to zero. The only benefit is that you can look at the sources it finds, as some of them might be useful. However, the generated report itself is not really suitable for anything. That’s my take on it.

It probably also depends on what you’re looking for. I work in scientific research and was mainly searching for something in the context of chemical and biological studies. Maybe it works better for other cases. Dunno yet

[keen sigil]

Hi folks,
Perplexity just introduced their deep research mechanism. it's 100 queries per month for pro users and 5 queries per month for free user. If 20$ is too expensive for people, can use my referral link, to get $10 off, the fun part is we can use same queries with different models, maybe can change the model to claude or gemini to do the deep research.
The deep research aspect is slowly being deployed I guess, I haven't receive it yet, but providing the referral in case anyone wants it.
https://perplexity.ai/pro?referral_code=FTY4NCFW

[austere kestrel]

this likely explains at least some of the hallucination issue, and prob a lot of the mentions about a lot of 'fluff'

setting an arbitrary (and rather long) word-length target via the ststem prompt is erm.. suboptimal for a lot of reasons imo

[alpine cosmos]

Hey guys, just signed up for the pro plan. I notice there's a question regarding API billing. I assume the normal pro plan, does not bill me for API use of various models, correct? I have to actively turn on some sort of API? Or is it 20 bucks a month + API costs for varying models? Just so I don't end up with a surprise bill from having a 10 response long convo with Sonnet or something

[buoyant schooner]

no

it's 500 / day for pro users

[buoyant schooner]

agree

they should remove that

this definitely explains a part of the crazy hallucinations it has

[austere kestrel]

yeah it's completely missing the point of what people what: it could be many different things, drawing on extensive sources / research, but rarely i thikn would people want a 10,000 word report, with generic predefined sections.. if the model focusses on meeting these criteria, it will invariably lead to a bunch of fluff at best or hallucinations at worst (unless the actual research query / user would expect such a templated and long response)

like i was testing it earlier asking things like "how many lakes in Colorado start with the letter 'F'? Provide a comprehensive list")

list ≠ report

i see it even says "avoid lists" in that system prompt artefact ha

"formal academic tone".. just tell the model it's helping a researcher and let it do it's work from there.. aside from in-line citations (which i think need to be formatted a certain way to render in the UI), most of the instructions given to the models are superfluous or just create like a straitjacket confining it

[frigid rivet]

Is deep research down? I am getting the 'something went wrong' screen.

[gritty siren]

works fine for me..it just recomended me the most practical bed frame to buy

[proper rapids]

i'm on free tier, at what time does deep research refills or resets? Is there a particular time?