💬│general

[burnt sparrow]

I'm working on a python app anyway and dont reallly need a ui

It did it again to me in the web interface evne with the extension uninstalled

Just blank

[white lake]

how do i get the pro role

[indigo valve]

That's so weird

@burnt sparrow You can try resetting site cookies

[burnt sparrow]

It was my ad blocker

I turned it off for the domain and seeing whats going on. It explains why the extension was freaking out though.

But for right now until I get the API script running without issues I'm just using the macos client and labs for testing. I did one API call already and the token return was way too high.

I have to find the right amount of data vs cost

[wary yoke]

does perplexity pro allow you to generate images? if so how

[cobalt monolith]

Yes although I think it's not like others ai
You have to write something for example some prompt send it and then click generate image on the top right

[thick smelt]

Have u resolved your charge issue?

[cloud sigil]

anyone know any ai that can summarize many pdfs/audio stuff/downloads and can answer questions like chatgpt?

[cobalt monolith]

Well you can either use notebooklm or upload files to perplexity

[wary yoke]

could you elaborate. I don't see any option to do som

When I chat and say generate this image, then it lists some online ai tools

[wary yoke]

There are 4 tabs library, home, search , where is this option?

[cobalt monolith]

in chat
for example you say

"hello perplexity"
he responds with some nonsense
and then at the top right you will see this option

[wary yoke]

thank you so much

does it generate the image that we searched in prompmt?

[cobalt monolith]

I guess so

let me try

yes it seem like it somewhat did

although i wish the facilitation of image generation was better here

[serene sand]

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

[cobalt monolith]

oh cool didnt know about that

[serene sand]

https://discord.com/channels/1047197230748151888/1111786888626438245 hop in

[wary yoke]

thank you , Now I'm able to generate images

[cobalt monolith]

@serene sandare there any new cool features the perplexity team is cooking?

[serene sand]

https://www.testingcatalog.com/tag/perplexity/

[cobalt monolith]

interesting

[night lintel]

Hey guys do you know why perplexity doesnt accept my microphone?
I run grapheneos on a Pixel 8 pro.
I gave all permissions installed Google speech recognition and tried many different settings. I want to use the assistant and voice bot.
Microphone via tts works.
Also tried installing Google assistant works without any problems.

[night lintel]

I think I found the fix.
I had Microsoft swift key installed and it crashed everytime I used the voice Feature.
I now deactivated it and now it works like a charm.
Strange but it fixed it for me!

[left jasper]

Is it just for me the ctrl/cmd+. shortcut is broken since the new reasoning change on the web? Quite annoying that it stopped working once the o3 option was added.

[sharp turtle]

There are some bugs in the Android client, such as every time you have to click Pro Search to switch to o3mini in the input box, it will not remember the previous selection which still use Pro Search not o3mini, and the Rewrite function, I clearly use o3mini, but it still shows that it is based on Claude-3.5 in the settings

[fresh pivot]

Excuse me, why can't I see the reasoning process today even though I chose the R1 model?

[serene sand]

works fine on my side

[gray arch]

perplexity so laggy

[stable radish]

Hi, anyone having PRO search completely messed up today? As in, activating module and not having actual search results?

[lusty cave]

Yes it sometimes switches between models

[lusty cave]

Yes. I think they are in a wave of adding features. I think after all these ai companies take rest from releasing new models every week they will start to optimise the workflow.

When i ask perplexity about what models are offered in the pro plan it shows an old list of ai models. 😂 Even if i use R1 or o3 mini they themselves missed their names in the list 😂

[spare kernel]

Hello

[cinder comet]

while this is all good but how about we focus on the search service itself

adding models wont improve the search accuracy, it looks to me that the main objective from pplx has diverged to this new mentality to add whatever model is released

[wheat geyser]

hey peeps… started using perplexity a few weeks ago and it has quickly become my favourite ai app for news and research! i love the innovation and i’m here for the ride 🙂

[stable radish]

Make sense, but I do enjoy using a stable service... One day there's no PRO search, the other day, 2 model are disappearing...

[short gale]

the excessive hype and media coverage surrounding R1 further exacerbated things

[cinder comet]

i mean r1 reaction was deserved tbh giving how it was free and near o1-pro performance... but the thing is that even deepseek search service is more accurate than pplx

pplx had this deep search feature but they removed it

[serene sand]

#💗│sharing message ☕

[green crystal]

And add that r1 caused other ai companies to work on affordable/free reasoning models lol

Gemini flash thinking improved capabilities quite a bit
Now it just needs to start properly working with it's plugins

[cinder comet]

yea It changed the whole rules of the game

openai probably had a 400% margin profit from o1 pro plan ( $200/month )

turns out it was cheap to run such model

or they've just reached what deepseek already achieved with r1

[blazing owl]

I want to know how many billions of parameters are deployed in Perplexity's R1 model.

[cinder comet]

or sama is blatantly lying

[cinder comet]

they said its the full version

the one with approx 600b

dunno if its true

[blazing owl]

671 billion parameters？

[cinder comet]

yea

[blazing owl]

好的，谢谢

ok thanks

[cinder comet]

but they may be using some 3rd party provider

like together.ai or hyperbolic

openrouter...

[cinder comet]

🫡 🇨🇳

[blazing owl]

haha

[blazing owl]

🫡

[outer arch]

Hi! I have the pro subscription but how do i get my pro discord role here?

[blazing owl]

I encountered another issue today. I selected the R1 model, but why didn't it interact with the RI model? Also, I can never use the o3-mini model. Every time I select o3-mini, it automatically switches to the R1 model.

Has anyone else encountered the same issue as me?

[tame current]

altman said they were losing money from the 200 $ / month plan

[buoyant raft]

i have doubts that its actually using r1. the output quality is extremely bad, compared from the outputs on the deepseek app

[tame current]

they prob using distilled version which is worse than the full

thats why i left pplx cuz they're not open about wtf they using

what o3 mini version am i getting ?

[buoyant raft]

seems i could get at least some beliavable info out of it, lol

Based on the extensive evidence, there are several concerning practices by Perplexity:

Content Theft and Misrepresentation

• Perplexity has been caught scraping content without permission and bypassing website restrictions using hidden IP addresses
• The company plagiarizes content from major publications like Forbes, WIRED, and Bloomberg
• Their chatbot generates false information and attributes fake quotes to legitimate sources

Deceptive Practices

• Perplexity claims to directly access sources but often reconstructs summaries based only on URLs and metadata
• The platform's interface shows a "reading" animation that misrepresents how it actually processes information
• The company falsely claimed to be Forbes' second-biggest referral source

Legal Issues

• Multiple lawsuits have been filed against Perplexity:
  • News Corp sued for copyright infringement and misuse of trademarked works
  • Dow Jones and New York Post sued over alleged misuse of copyrighted content
  • The company faces potential legal claims ranging from infringement to defamation

Recent DeepSeek Integration

• While Perplexity claims to run DeepSeek R1 on US servers, investigations show concerning privacy practices
• The implementation appears rushed and raises security concerns about data handling

The evidence suggests a pattern of misleading practices, unauthorized content use, and misrepresentation of capabilities to users. The company's rapid valuation appears to be built on questionable practices rather than genuine technological innovation.

[tame current]

wait where do you get this from?

it's ai generated

[buoyant raft]

from perplexity lol

[tame current]

did you use perplexity to tell you about its concerning practices ? 😄

i wonder how much is true from this

[buoyant raft]

it has sources but not sure if allowed to post links, so i removed them

[cloud marsh]

"Content Theft" 😂 scraping content without permission? isnt that how internet works?

[buoyant raft]

it doest even understad the question and repeats itself in the second reply. imo a patetic atept of a failing business to copy chinese innovation.

[tame current]

they dont copy anyhing its open source

its not perfect and i'm guessing its like i said distilled version (so a bit weaker) but idk for sure

[opal laurel]

I can not use my selected model in Perplexity right now

[harsh hazel]

How to get pro role? And should r1 be available in android app? Mine doesn't have

[buoyant raft]

not the issue, unless ur a IP lawyer. i I always considered the content stealing one of the best qualities of AIs,... but it doesn't change that perplexity just sucks and this is scam.

[buoyant raft]

Bad wording on my part. If they would copy it, that would be awesome. But it's a failed attempt of imitation, while misleading customers

[indigo valve]

Make sure your app is updated

[clear stump]

Wait this is the wrong server

[modest pulsar]

With what name do you address perplexity? I call it "plexy"

[magic talon]

the drink

[cursive jacinth]

is the 1 million dollar really just for US sir? this is bias sir. I also want to be a millionar sir.

[harsh verge]

No million for all the countries that produce fentanyl, only one possible winner: a resident of the country that consumes it. 😇 (And hello, sir)

[halcyon coral]

https://x.com/perplexity_ai/status/1887897060902318263?t=nY0WGz964KT17ahSeN03pA&s=19

Perplexity (@perplexity_ai) on X

Perplexity now offers file and image uploads with an expanded context window of 1 million tokens.

Free for all signed in users in “Auto” mode.

X · Feb 7, 2025, 4:11 PM

[cursive jacinth]

Gemini or all models sir? its pretty vague sir.

ok so i tested it sir, i used claude model and uploaded a file using auto sir. so my assumption is that probably they are using gemini to read the content first then the one will answer is the chosen model sir.

[tepid sequoia]

Wow, a million-token context window, how amazing 😄 but I want to write something, for example, about Harry Potter characters, and I want it to be at least 2000 words long 😄 and suddenly we always stop at 1000 words 😄 what does it mean that to achieve a 1 million context window you have to send a lot of messages that are already summarized, just in such a text your context is shortened, so great as a marketing move but again, in reality, we haven't improved anything 😄 basically, we added Gemini 2.0 which is available everywhere, we added the R1 model which is open source and the only argument is that it's hosted in the USA, like yes, it's no longer censored because we deal with Tiananmen Square somewhere in China in 1989 every day, instead, we got an effective message shortener, I don't know how, but I would exchange censored Tiananmen Square in China in 1989 for the possibility to write long texts and detailed answers.

[short gale]

Still confused why updates like this can’t also be included in #📢│announcements

Not everyone uses that god awful website

[tacit igloo]

Yeah. A proper changelog would be nice

[grand silo]

The fact it requires Auto suggests some kind of model mixing or auto selection of model rather than sticking with your chosen option.

[cursive jacinth]

Yes sir but atleast they are trying to improve now sir i can see that sir

[grand silo]

That or are desperately trying to find something ‘special’ to justify the VC funding

[mystic nymph]

Hallo

Im fron poland

[muted galleon]

Can someone point me to where I can get some help with the API? It is constantly 504’ing. I can’t get anything done.

[cursive jacinth]

what about it sir? VC funding sir?

[abstract ibex]

I wonder why Gooddawg uses the word sir like you would a comma

Should I ask perplexity?

[abstract ibex]

https://www.perplexity.ai/search/give-me-a-list-of-20-healthy-s-6xweIcoqSVC21OTZXxhmfw

What

Oh the links

It says "I cannot include direct links"

why can't you include direct links?

I cannot include direct links to specific products on Walmart's website for several reasons:

1. Policy Restrictions: Many platforms, including this one, have policies against providing direct links to external websites for various reasons, including security, privacy, and content control.

2. Dynamic Content: Product links can change over time due to updates in product listings, inventory changes, or website redesigns. This means that even if a link is provided, it might not remain active or accurate.

3. User Experience: Providing direct links might not always offer the best user experience, as users may prefer to navigate through a website themselves to explore related products or promotions.

4. SEO and Affiliate Policies: Some platforms may have restrictions related to search engine optimization (SEO) or affiliate marketing policies that discourage direct linking to external sites.

Instead, I can guide you on how to find these products by suggesting keywords to search for on Walmart's website. This approach allows you to explore related products and promotions while ensuring a more personalized shopping experience.

Huh

My theory is that it just doesn't have access to the links due to how perplexity searches

[short gale]

https://discord.com/channels/1047197230748151888/1301576498200711239 I wouldn’t hold your breath

[abstract ibex]

o3-mini sucks in every other AI tool I've seen

This model only seems to work on chatGPTs website

I mean maybe the model can detect benchmarks and be smarter do it can score higher

Maybe we need to prompt it like it's a benchmark lol

Maybe it will actually do something

[cinder comet]

hes lying

[sick smelt]

Does anyone know if the focus feature is gonna come back?

[tepid sequoia]

But he's not lying, he really is losing money, it's just a bit different. When you piss away billions, and your investors are knocking on your door, you have to say something. But then again, let's consider this, he said this before the release of R1. Now he's happy to get at least 200 dollars. Well, it's one thing to make ChatGPT a profitable service to cover costs, and then another to start making money for your investors. OpenAI is currently only valued on paper, by the stock price, that's all. But if something were to happen to the market, if there was a crash, you'd see some crying. But then again, I see it this way, at some point, OpenAI will definitely become an important part of the world, so even if it's not making money, someone will have to subsidize AI so it doesn't collapse.

[tame current]

hey guys, i need help with perplexity wake up word on android

ping me if anyone has suggestions or ideas

[tacit python]

gm, anyway to change my finance watchlist?

want to takeaway some i had as only two get shown

[willow kettle]

Flash 2.0 thinking is mad about Perplexity + DeepSeek advice/response ```This AI response is intentionally or unintentionally deceptive. It uses selective citation, misinterpretation, and outright fabrication (re-misattributing the 77.7% figure) to create a false impression of Drotaverine's efficacy for stool frequency and incomplete evacuation. It is clinically unsound, irresponsible, and potentially harmful.

Crucial Disclaimer (STRONGLY EMPHASIZED AGAIN - THIS IS CRITICAL):

This AI response is DANGEROUSLY MISLEADING and SHOULD BE COMPLETELY DISREGARDED for medical decision-making. DO NOT use Drotaverine or any medication based on this AI's recommendations. IMMEDIATELY CONSULT A HUMAN DOCTOR (GASTROENTEROLOGIST) for proper diagnosis and a SAFE, EVIDENCE-BASED, and INDIVIDUALIZED treatment plan for your symptoms. The AI's repeated and escalating misrepresentation of medical evidence demonstrates that it is NOT a reliable source of medical advice and could lead to ineffective or harmful self-treatment. Your health is at risk if you rely on this AI for medical guidance.```

[spring shuttle]

Made a community purely for developement with AI. If this sounds like your cup of tea, shoot me a dm

[shut dagger]

so we now have 1 million for all models?

no more 32K?

[jaunty dove]

@fierce sonnet umm, all files or just pdf? It doesn’t let upload epub.

[crystal carbon]

I love this app

[hollow junco]

Only when ‘auto’ so not for all models, only their default model

[shut dagger]

auto always gives me output from model I selected in settings

[gusty wasp]

Deepseek v3 when

[hollow junco]

Auto uses llama 3 70b/their sonar model

lol what

well it used to use llama 3

[shut dagger]

don't ask it, just check the model name on output, there's a small badge on the right corner of output

[outer glen]

is deepseek model free?

[shut dagger]

Auto used my default model selection settings

[zealous kayak]

new announcement is the vaguest thing ever
so what, is gemini acting as like a file summarizer, basically?

it's definitely gemini flash only as no other model has 1mil

[hollow junco]

claude has a max content of 200k and when you use the 200k context it costs like $2 per query so idk about that lol

[shut dagger]

I presume other models would max out their context?

[hollow junco]

probably but i know you can get llama 3 to 1m context with llama 3 gradient

[shut dagger]

they probably have special deals, after all, Amazon guy has invested into perplexity

same guy who funds anthropic

[hollow junco]

idk abt that tbh

but tbh it doesnt matter we'll find out more soon

gemini flash is cheap as dirt tho

[shut dagger]

yeah, most probably we will find disappointment

😂

[hollow junco]

💯

[outer eagle]

I’m new to perplexity and loving it. Is there a way it can also be used as a chatbot so I don’t also need a ChatGPT subscription?

[zealous kayak]

yeah no

even anthropic themselves only offer 500k on their enterprise sub

let alone the costs

[fierce sonnet]

hey, currently we do not support epub. We support simple text, pdfs, markdown, and code files, plus a few common proprietary formats such as .docx and .pptx

I'll pass along the request for epub support, thanks for the feedback

[pliant sand]

I would like to get in touch with Sales team at Perplexity to explore Enterprise plan and features. What's the best way?? I can't seem to find a way to reach out from the website.

[shut dagger]

I thought max context of claude was only 200K

[zealous kayak]

on API and claude pro, yes - but they offer 500k on the enterprise sub (60/70$ a month iirc)

it's on the announcement for the enterprise plan

[shut dagger]

I meant cost not context

[cursive jacinth]

Hi sir, context limit 1m applies to all model sir? can you elaborate sir?

[red forge]

Only on Auto

[cursive jacinth]

so it doesnt use any model sir? I dont understand sir, who reads the 1m text sir? there is a model called auto now sir?

[viscid bloom]

Is the deepseek on the website the 671b model ?

[clever chasm]

I missed the Xfinity rewards as I just found out about this rewards program!

[rustic kestrel]

what perplexity been freezing for anyone on browser?

[indigo valve]

Yes

[viscid bloom]

Oh nice

[strange cypress]

https://youtu.be/XiXLti_Y_is

[rustic kestrel]

🙏

[signal yoke]

@signal yoke

[short gale]

#📢│announcements message thank you

[burnt sparrow]

This was exactly the same bug I had yesterday when Installed that 3rd party browser plugin. Then the UI just went blank, I turned off my content blocker so the page would reaload but ended up uninstallling the 3rd party chrome extension and that fixed it.

[burnt sparrow]

kinda want to use this in the API but alerady burned 2 cents with sonar pro reasoning with one qurery. I'm back to using labs for now since the API for some reason doesn't have weird token bugs like the mac client does. I think each clients values are different between web, labs, local MacOS, and API. I ended up installing the local MacOS client it just kept running like trash and giving me terrible answers when I was doing my research.

Really wish we had acccess to the other models in labs and the API.

[stone kettle]

It was a big day for the family 💪

[unique umbra]

Who the hell still uses 3.5 sonnet

[feral trail]

After getting LinkedIn Premium how many days will it take to get Perplexity pro code (premium perk)?

[unique umbra]

Isn't linked in premium like 100 bucks a month..

[torn sparrow]

Hi, my perplexity android app does not have the option to choose o3-mini, anyone else has the same problem?

[blazing owl]

Excuse me, why can't I still use the RI model on the web version? Why can I only use professional search?

[feral trail]

depends on the plan and region

[rough minnow]

https://tenor.com/view/6m-rain-wallstreet-make-it-rain-gif-1159512466922260118

[spring bridge]

Any promo codes?

[rough minnow]

La expresión "con cabeza" puede tener varios significados dependiendo del contexto en el que se use. Aquí te presento algunas posibles interpretaciones:
Con la cabeza en su lugar: Esta expresión se refiere a alguien que tiene una actitud sensata y juiciosa, manteniendo la calma y el control en situaciones difíciles.
Tener cabeza: Significa ser inteligente o tener capacidad para pensar de manera lógica y racional. Por ejemplo, "Ella tiene cabeza para los negocios."
Ir con la cabeza alta: Indica orgullo o dignidad. Por ejemplo, "Salió del lugar con la cabeza alta después de defender su posición."
Con la cabeza fría: Se refiere a mantener la calma y no dejarse llevar por las emociones en momentos de estrés o crisis. Por ejemplo, "Debemos mantener la cabeza fría para tomar decisiones acertadas."
Con la cabeza en las nubes: Similar a "tener la cabeza llena de pájaros," indica que alguien está distraído o tiene la mente en otro lugar, no enfocado en la realidad inmediata13.

nop

one year promo

pls dont hack the code

jep

dime la verdad que piensas realmente de e l o n puro marketing

[torn ravine]

Is reasoning down?

[rough minnow]

https://tenor.com/view/dashiexp-dashie-dashiexp2-pause-pause-on-that-gif-25929270

[wise folio]

I was thinking about buying a pro sub just to get access to deepseek r1 without being rated limited on the native deepseek webchat, but I am curious if I would be able to turn off the web search function if I use deepseek r1 in perplexity

Does anyone know

[torn ravine]

Yes u can

[wise folio]

It would basically be the same as deepseek r1 in the native deepseek chat, where it would just give you the result (for example, the answer to a coding problem) right? I just don't want it to be formatted like a search result that it would typically would even though there is a toggle to turn off search

[torn ravine]

I'm not sure but I guess Pplx still passes some system prompt onto DeepSeek

[wise folio]

I see

[rain lodge]

Its not the same. First they have removed the censorship pieces. Second its hosted in the US by perplexity and lastly they did fine tuning on the model so it aligns with the rest of the platform. You will get similar responses but they are tailored to fit the perplexity platform.

[warm cave]

Sonnet 3.5 is the goat

[unique umbra]

Literally how

It's 2 years old

Or a year but it performs worse than others

And its expensive

[warm cave]

It’s currently the best “non-thinking model” if you work with code you know thinking models can be less see less to work with and are slower

[unique umbra]

Gemini flash moment

[warm cave]

R1 + sonnet is really good

[unique umbra]

Which is better

And faster

[warm cave]

R1 is the planning model in cline and sonnet is the exicution model

[cerulean glacier]

When do you think we'll see the 'Writing' focus added back in?

[warm cave]

O1 is stupid expensive cost $186 for this bench

Sonnet cost $15

R1 + sonnet was $13

[unique umbra]

Again like

There's gemini now

And o3

I think even deepseek v3 is better

[warm cave]

Gemini is pretty new, can’t say if it’s good or not yet, o3 is still pricey, and thinking models are annoying for multi-turn long context with code, I would say o3+ sonnet would be good

But I compare sonnet to GPT4o because they are both non-thinking models, so both are easier to work with multi-turn long context

Hopefully Gemini flash lives up to the hype, I tried it a little and was not impressed, but hey that’s just a few tests, I could be way wrong, will test it more

[unique umbra]

Newest Gemini model gets like 60 tokens per second while performing ab 10% below r1

[warm cave]

What bench mark? Just wondering because that’s a big jump, Gemini 2 flash experimental was really bad at code

I will definitely give gem flash some more time tmr

[unique umbra]

https://artificialanalysis.ai/models/gemini-2-0-flash-experimental

They only tested experimental so far

I'm guessing the released model is a bit better

[warm cave]

Nice, those look promising, I hope to see it tested on aider benchmark soon

[frosty vessel]

171 sources :o

[naive sage]

So it can be similar to Google and OpenAI's Deep Research. I wonder if anyone's done a detailed comparison between the three.

[frosty vessel]

kind of

i used a tuned prompt

[tame current]

whats the prompt ?

[acoustic lion]

Is there any usage limitation with Deep Seek R1 in Pro account? I can't use Deep Seek R1 since yesterday, I choose Reasoning with R1 model, but it only comes out result by using "PRO" model

[frosty vessel]

600 per day i believe

[frosty vessel]

"perform at least a dozen queries that are related"

[naive sage]

The pro search has a limit of 'at least' 300 per day... I'm assuming some cases let you go above, but you should expect at least 300. To compare, o3-mini has a limit of 150 per day on OpenAI (on the $20 plus plan), so it's a good value... though most people don't use that much. (Edited: Correction - the limit has been upped from 50/day to 150/day for o3-mini on OpenAI, Perplexity is still double that)

[acoustic lion]

I didn't use that much, around 50 per day in the past few days since subscription

[tame current]

Ohh it worked 296 sources 💀💀

[frosty vessel]

yep

[tame current]

@frosty vessel How did you find that tyep of secret prompts bruhh or you just experiment it

[frosty vessel]

just testing stuff out, i started with more lengthy prompts but the simpler it was the more results it gave

[naive sage]

(I'd imagine that with too many sources, the context length becomes a potential concern though)

[acoustic lion]

Thank you all for the reply. I definitely didn't pass the limitation. Do you guys know where can I find the offcial support?

[naive sage]

I'd report it at #1140622008086970420 ... someone else mentioned a similar issue here so it seems to effect multiple people: #⚡│ask-community message. Someone also opened a thread in bug reports already for it, it seems.

[acoustic lion]

Okay, I'll have a look now. thanks a lot

[naive sage]

Hope they get everything in order in time for the Super Bowl event

[frosty vessel]

@raw seal $20

[raw seal]

ye thz

i got it for 10€/month for a year (flex)

i have chatgpt plus too

idk slightly unnecessary ik

[frosty vessel]

i pay $35 a year

[naive sage]

In the next month or two deep research will be coming to chatgpt plus, so you won't need the pricey pro plan to do it... that'll be interesting to compare.

I don't know what kind of limits they'll have in place yet, though.

[frosty vessel]

oh we're talking about chatgpt?

[naive sage]

Since you mentioned you had it. 🙂

[frosty vessel]

i was referring to perplexity

[naive sage]

Yeah, I think it'd be neat to compare web searches and results between them.

[frosty vessel]

im a big fan of perplexity

[shy cloak]

but currently they disabled R1

[naive sage]

I don't have pro so I can't test, but does it still work with R1 on the Pro searches?

[frosty vessel]

works for me

this is what i used

[shy cloak]

It has problem now

[frosty vessel]

yeah, you can do pro search without reasoning but cant do reasoning without pro search

[frosty vessel]

i just used it rn

[shy cloak]

Come to hackerrank, pick a hard problem

and you will see

https://www.hackerrank.com/challenges/build-a-string/problem?isFullScreen=true

[frosty vessel]

send prompt pls

[shy cloak]

Find the optimal solution

[shy cloak]

include these 2 images into the prompt

and send "Find the optimal solution". Choose R1. You will see. Many people are facing this problem

Few days ago. R1 provided a chain of thoughts in around 240 seconds. Then when it was wrapping up, Perplexity ended the response. No answer provided but only the thinking

[frosty vessel]

it gave me an answer instantly so i assume no reasoning happened? but same thing when i rewrote with o3-mini

[shy cloak]

I have no idea why today it's automatically switched to Pro model instead of alerting "R1 is unavaiable for this kind of prompt"

[shy cloak]

exactly

the reason I know there is no reasoning because of the answer quality

Pro model is currently so bad. It can't answer these kinds of questions

[frosty vessel]

when i ask it a simple question it reasons and answers. but if i ask the same question as an image it doesnt reason and uses gpt4o. interesting

[shy cloak]

But, if I prompt to know "what Apple plans to do in 2026", R1 is enabled.

[shy cloak]

this is because

for some reasons, the server knows this is an easy response

that R1 can provide a short chain of thoughts

thus R1 is enabled

[frosty vessel]

i wonder what it being an image has to do with it though

bc its the same question

[shy cloak]

I have tried easier questions but it still happened. No R1

[shy cloak]

2-3 days ago

it was working

R1 was enabled for all prompts

[naive sage]

If people have been reporting it for a couple of days, has anyone from Perplexity acknowledged it as an issue or discussed it on here?

[frosty vessel]

thats funny

[shy cloak]

I believe it's intentional

[frosty vessel]

perplexity acknowledging issues 😆

[shy cloak]

drop some bs OpenAI's reponses

OpenAI is always a bs

[naive sage]

4o's truthiness seems to be not great lately, but o3-mini seems better so far.

Claude's due another Sonnet refresh soon, I'm sure they're working on reasoning capabilities too - but Anthropic has been struggling with capacity issues b/c their models aren't as efficient.

[shy cloak]

Claude is not good for logic and coding.

anyway, I don't think auto switching to Pro model is a good idea. Just pop up an error message to say that R1 is unavailable and stop

I don't know, but it seems like there is no other reason than mine. R1 or I leave. Auto switching to other models is just a bs idea. not worth it

[raw seal]

for what?

perplexity?

[frosty vessel]

yes

[raw seal]

wha?

oh

reading chat history

a bit late

[raw seal]

this is the pro discord so write pro support

[shy cloak]

I just found that if I attach an image, R1 is not enabled

[raw seal]

works for me

wiat R1 gets disabled if u attach an image? @shy cloak

duck.ai has GPT 4o and Llama 3.3 now

idk about the llama model tho

is it any good?

what is it comparable to, 4o?

[shy cloak]

when I attach an image and click submit, R1 is not activated although I chose R1 for my prompt

[raw seal]

duck.ai aka DuckDuckGo AI chat

incase someone doesn't know

i think it is maybe it just doesnt show

is it the same as me?

[shy cloak]

as I said, stop playing with OpenAI because all of their products are bs

[raw seal]

i attached a recording

[raw seal]

tf

its ai

who cares if its openai if it works for u and u dont care about them having ur data

[shy cloak]

why you don't care if it can't give you the answer?

[shy cloak]

seems like an intern's project. you need to focus on Perplexity only

[raw seal]

welll.. yeah i do care

[raw seal]

you don't know Duckduckgo ai chat?

you do know what duckduckgo is right?

[raw seal]

why?

seems reasonable but i want a reason

or a short explaination if ur willing to give me

[slow linden]

We need mistral back in the mix

[hard veldt]

Grateful if Perplexity could clarify whether the 1 million context window applied to Gemini flash only, and whether there’s any change to the 32k cap for other models👍

[frosty vessel]

use le chat

[slow linden]

But then I’d be using a different app

[green crystal]

R1 cant see images

though a little disclaimer wouldnt hurt tbh

[placid coyote]

hmm, does the gemini 1M ctx work for anyone? I tried with 3 different files in Auto mode and am always getting Sonnet, not Gemini, and sonnet only sees first 100k characters... so like 27k (4o) tokens which is pretty far away from 1000k tokens they mention in announcements 😑

[glad otter]

Can you fix whatever these numbers are for please.

[past sluice]

I really enjoy using Perplexity and appreciate all its features. I have a couple of questions:

1. When will the deep research mode be available on Perplexity?
2. Are there plans to enhance the free plan, for example, by offering at least 10 pro searches per day?

Thank you!

[serene sand]

https://x.com/aravsrinivas/status/1884801300027589007?s=61

Aravind Srinivas (@AravSrinivas) on X

All Perplexity Pro users now get 500 daily DeepSeek R1 queries (without censorship and prompts not going to China).

Free users get 5 daily queries.

X · Jan 30, 2025, 3:10 AM

[mystic rune]

1 million tokens is only in auto mode?

[cursive jacinth]

thats my question yesterday sir but someone just reacted to my question like a 🤦🏻 sir, clearly he doesnt understand my question sir.

People these days should stop smoking w**ds sir, their comprehension gets worse sir.

[cursive jacinth]

here sir, i was being sarcastic sir but the guy got hooked on my sarcasm sir. 😂😂😂

[pulsar crater]

R1 missing from android version again?

[forest star]

I want to work with a team of AI engineers one day

[serene sand]

#💗│sharing message ☕

[past sluice]

deal 😄

[austere kestrel]

pretty syre file uploads still get rag'd / become a 'source', rather than the raw contents of the uploaded file or pasted text being injected into the prompt (e.g. as happens using Anthropic's chat UI)

and whatever handles that process, has a cap at ~32k tokens

This is ~45k tokens but no file upload (input across muliple messages.. obviously tedious / impractical beyond testing ha). Snnet3.5 shows it has visibility of the beginning, middle and end of text
(whether it goes to 200k - or 1m for gemini - i dunno.. but yeah it's the file upload/rag process that remains the limitation, rather than model context windows, so far as i can tell / fwiw) https://www.perplexity.ai/search/gday-mate-dSq4sInqREWgs6Qx0va1cQ

[tame current]

how perplexity r1 giving faster response than original website ( there ar ethinking process is also little differnet ) ?

Are they different ?

[rustic kestrel]

does the perplexity website freeze randomly

for anyone else

[stone kettle]

Good morning everyone today's focus is on crypto 💪

[abstract ibex]

I also love scamming people

[stone kettle]

Commenting without doing research is ignorance at its core 🤡

[abstract ibex]

Yeah I also love that

[acoustic salmon]

Research done, scam detected lol. Leave the 🤡 emojis to the professionals.

[abstract ibex]

Lmaoo

[acoustic salmon]

my emoji, you no take emoji

[sweet oxide]

realistically how much does [1 million tokens](#📢│announcements message) equates to in a file content?

[rustic kestrel]

1 million words i believe

tokens r kinda jus character sets

[acoustic salmon]

1 token ~ 4 characters in English

according to OpenAI docs

Wayne Gretzky’s quote "You miss 100% of the shots you don't take" contains 11 tokens.

it's about 41 characters, so pretty close

I guess it rounds up for that 1 character

[naive sage]

Depends on the tokenizer tbh, some work differently than others but the general rule of thumb is treat each word as a token: https://medium.com/the-research-nest/explained-tokens-and-embeddings-in-llms-69a16ba5db33

[cobalt monolith]

i am a bit confused by the announcements do all users get 1M tokens per file?

[sharp turtle]

O3mini what happened?

[cobalt monolith]

also a bit confused by "Free for all signed in users in “Auto” mode."

@fierce sonnet could you please elaborate

[urban crest]

I am a pro subscriber & am trying to load a 10MB markdown file to a perplexity space. I get the error ‘unable to process file’

Any thoughts on how to proceed ?

[acoustic salmon]

do you think it might be busting the token limit?

[tame gazelle]

Wait so 1 million tokens. Is that only auto mode or all modes? Is it just for words in an uploaded file (but the chat itself is still 32k?)

[acoustic salmon]

believe it's auto-only

[limber saddle]

This is the worst support I've ever had. Perplexity is again telling me I dont have pro and not letting me use the pro models when my user says I have pro.
And the support just forgot about my ticket

[echo wind]

[cobalt monolith]

i guess so its choosing gemini?

[stone kettle]

We post nothing important here 🥱 and yes you are still a 🤡

[acoustic salmon]

even if you set it to another model?

[cobalt monolith]

i have no idea king, i am speculating

[acoustic salmon]

I don't have a good test document that gets close to the 1 mil token limit unfortunately

[naive sage]

Yes, gemini is the only one on Perplexity that has a 1M context

Unless they have some amazing thing in-house they developed

So I'd imagine they use it whenever there's uploads in order to ensure they can handle it... though it might make sense to check the text size, it's easier to just default to it.

[cobalt monolith]

doesnt i have 2M?

it*

[naive sage]

The Pro has 2M, I think flash only has 1M

Yeah, Gemini 1.5 flash and the newly released (as of a few days ago) Gemini 2 both have 1 M. Gemini 2 Pro is in experimental status (with hard usage limits, so can't be used by Perplexity yet) and has 2M

[cobalt monolith]

oh i see

i guess it makes sense why its for auto then

i hope one day we will have in built solution for all models

of 2M tokens lol

[winged axle]

yall what do I do if I have a reeeeeeeeeeally long chat history so much so that it's become laggy to scroll down. How do I transfer the context to a cleaner window?

[naive sage]

One trick people used to do is ask it to generate a summary that preserves the important details and context of what has been discussed so far, then copy that old chat to a markdown file and start a fresh one.

Then you can feed that summary into the context to start the new one... but obviously that's not a perfect solution

[winged axle]

I'll give this a try thank you. I wish I could just minimize the old chat so that it doesn't render everything all the time

[naive sage]

That might be a good suggestion to post in #🌐│feedback-web , I think someone else said it can become slow when there's a lot of stuff too

(Though it is important to keep in mind that in general asking an LLM questions with a large context history is naturally slower than one with a smaller context, so it's hard to tell if the issue is 'rendering/browser performance/problematic js', or just model slowness in general)

If the problem is just when scrolling and it's not generating, then that sounds more like a web/js optimization problem though, which could benefit from minimizing/hiding content if it had that capability

[pine wadi]

Why perplexity removed O1 reasoning model and lama?

[naive sage]

If it's true they're using o3-mini-high, it outperforms o1 on GPQA and Coding, but doesn't do quite as well on Math and is nearly the same on Reasoning. So while o3-mini-high is also cheaper, it could be that they considered o1 replaced by o3, or it could be a cost-saving measure. I'm not sure what variant of o3-mini they're using though, the default is medium on ChatGPT. Removing LLAMA is less clear, as 3.3 at least is still considered capable and relatively inexpensive to run.

[oak sentinel]

P

[random ocean]

I have observed that pplx hallucinates a lot more with same query with R1 than the R1 hosted at deepseek chat.

That’s specially true for recent events.

[placid coyote]

if the RAG is still utilized for file uploads and anything larger than some fixed size of characters (used to be 8k) turns into a "file", then it is pretty deceptive advertising 1M context window which a user because of platform limits can't really use... 😕

[random ocean]

243 sources for a query, that’s kinda a record for me.

[leaden lagoon]

What?

I am not a mod

There are probably people in here that can mod

[proud plaza]

s

[gentle bison]

Since being on grapheneOS for the last week, the voice chat feature has gone. Is it only on my end and if so, anyone have any insight into why?

[twin lark]

[tame current]

MiniMax

[azure abyss]

How much does it cost to run Perplexity

[jagged sphinx]

I ask what todays date is (which I could obviously use my phone for I was just checking if it provided up to date information for everything) and it answered wrong, please just add a function that tells that date when you ask what is todays date or something like that and is correct based on the location you placed in your profile settings. This isnt really important but I’m hyper-fixating on it.

[sleek vortex]

millions lol

[azure abyss]

the owner is rich

[left stream]

Hard to decide which one was better, decided to have both

[frosty vessel]

How much does it cost to run Uncovr

[sleek vortex]

Not much yet

Like, under $100/mo

Opening subscriptions and reforming a bit soon though

[azure abyss]

What kind of api do you use?

[sleek vortex]

many many apis

basically all the model providers (openai gemini aws/bedrock openai->/azure, anthropic), mix of web search apis, youtube,image,x,file storage on s3, embeddings, other apis

[azure abyss]

Send the web src

😂

[sleek vortex]

?

[frosty vessel]

how much is a search on average @sleek vortex

[sleek vortex]

maybe $0.003 in ai costs + index costs ??

Can go much higher though, this is using optimisations and basically the cheapest models

maybe as i just averaged my ai costs from the last like month

i actually don't have good data points on that honestly

-# ...and i really should

[jagged sphinx]

-# no

[glad otter]

But they don't link to sources. They just show numbers. Think the linking system is broken.

[sleek vortex]

lol

[sleek vortex]

It’s a glitch from the model

aka it hallucinated a different source format

Hence perplexity didn’t catch those sources

You can probably just go and manually count to source 2 and 9 for now but

[mint oak]

What is the model for image generation in perplexity?

[naive sage]

When you're speaking of operating costs it's very difficult to get a number since it depends on several factors, such as the model used and even what hardware you're using to serve the query. But if you're talking about how much it costs the consumer those prices are here: https://docs.perplexity.ai/guides/pricing

[frosty vessel]

on his site i mean

[lusty flower]

Hey guys!
I’m working on a project and was wondering if posting links and files in the space section would increase the chances of the model hallucinating and if it is advisable to add stuff to those sections?

[naive sage]

Perplexity lets the user choose between three models for AI image generation: https://www.perplexity.ai/help-center/en/articles/10354781-generating-images-with-perplexity

1. Playground - for quick and efficient image generation
2. DALLE 3 - OpenAI's image generation model
3. Flux - A more recent model from Black Forest Labs

[mint oak]

Thanks, I think Flux is the best one, right?

[naive sage]

It's subjective - it's easier to get more vivid and creative things done in DALLE most of the time, but Flux is designed to be hyper-realistic I believe

See: https://www.goenhance.ai/blog/flux-1-vs-dalle-3

[mint oak]

Gonna take a look, ty

[naive sage]

If you have 16GB of VRAM, you can try flux yourself (I already tried it, and even the low end 10GB flux doesn't run well on my system, since I have 8 GB VRAM, takes forever)

Generations are also limited to 50 per day. See: #👉│feedback-general message

[jade breach]

Why does the Ai lie? Like it says the word count and its wrong

[naive sage]

Actual logic and reasoning is a more recent undertaking which is improving, but traditionally what LLMs are built for is analyzing content, building relationships between words that it reads in, and outputting the most likely 'next' token... so it's not really 'lying' or 'hallucinating' (the term most commonly used when LLMS seem to answer arbitrarily incorrect things), its just that the model sees the words as the 'best/most likely' response based on what it was trained on. Things like generating documentation, summarizations, explaining things in other ways - LLMs excel at that - so Perplexity's primary use case, summarizing and presenting news that interests people, and answering questions about the news is a good fit. Actually answering programmatic or logic questions that can give a wide variety of answers from one to another is hard if it hasn't been trained on those questions (programming is similar, it only appears better b/c there's tons of code samples online, so if there's documentation out there for how to do something, it's likely been swallowed up in training, but more obscure tasks or where documentation is just wrong or outdated it's still bad). Reasoning models work a bit different, generally some up-front processes to try to actually have more understanding about what to do, so things like counting letters/words generally work better in things like o3 or R1, but it's not perfect. Some reasoning benchmarks show only 25-40%ish success on the newest reasoning models... they'll continue to improve over time, though.. but I think tasks should/will ultimately be offloaded to specific compute units, like regions of a brain, that are more purpose built to do task X or Y efficiently.

i.e. function calling or tool calling as it's referred to

Then, when you describe what your tool does, it can see a question and decide 'Oh, I know how to do that thing'... and call your tool by inferring whatever parameters it needs (for example, maybe your user context knows you're in Shanghai, so when you ask for the weather it'll call the weather tool and pass in Shanghai as the city), instead of relying on its pre-learned data or always searching and taking in dozens of articles to try to figure it out

[frosty vessel]

why does search keep defaulting to r1 reasoning no matter how many times i change it back

[sacred zinc]

Hello!

[random helm]

This is also a best ai app i used so far

https://apps.apple.com/us/app/tryfit-ai-outfit-changer/id6740171699

[naive sage]

:sus:

[random helm]

??

[naive sage]

Seems like an ad

[random helm]

Bruh i use it

[short gale]

well with an endorsement like that sign me up!

[austere kestrel]

yeah I don't get this announcement - seems shamelessly misleading (or maybe there's just a massive disconnect between their marketing and dev teams.. dunno)

admittedly only a single test - so hardly rigorous / conclusive. but fwiw. using Auto (web off) and uploading a file with ~100k tokens of text, it seems it cuts off after 24k tokens – or, more conspicuously, at nearly exactly 100,000 characters

this is the file used, and here is the thread from the test https://www.perplexity.ai/search/see-attached-material-and-inst-E4GIqfr5RQ.gfIM17QzRrA
would be interesting to see if it's reproducible (or an anomaly / user error)

for comparison, same file and prompt given to gem-flash2 on google aistudio

[short gale]

The 1M context claim seems pretty dubious at best

[autumn meteor]

Been ripped off. Paid for Perplexity for a Pro subscription because it gave me access to ChatGPT and DALLE-E. So I could pay for one product instead of three. Now is says:

I currently cannot generate or provide images directly, even with a Pro subscription. However, you can use AI-powered image generation tools like DALL·E, MidJourney, or Stable Diffusion to create the image you described. These platforms allow you to craft highly specific prompts to generate custom visuals.

[naive sage]

After making a search, images are generated in a separate special prompt (in the web I think it's on the right side).

It doesn't support multimodal image generation (within the LLM prompt itself) - so those models have their own prompt

[autumn meteor]

no. they took it away. bait and switch. should refund those of us who paid for this. but the option is still there.

[naive sage]

Is this there after you do a search: #💬│general message

[autumn meteor]

I used it two weeks ago. You do it right inside the prompt. The dialogue box you talk about was thre before. But I did not need it before. It worked right in the prompt. That dialogue box does not appear to have the same capability. So I will have to go back and pay Open ART

[naive sage]

Their site is confusing - how many images do they let you generate a month?

[autumn meteor]

one. maybe. It actually fixed a problem for me that https://openart.ai/home (Dalle-E) has. One problem with Dalle-E is, and this is very odd, when it creates images the text in the image is 90% of the time spelled wrong. (Perplexity lets you use those 3 image generators I show in the screen print above.). I was having to edit a manual image editor to fix spelling errors in images it create. Dalle-E does a beautiful job of creating images but for some reason it cannot spell.

Well maybe they just moved it to the box you mention. I just used it to create this image with the Flux image generator

[naive sage]

Flux is supposed to be a bit better with text from what I've read. But yeah, lots of them have that issue - getting better slowly though.

[autumn meteor]

Look. Proof. I just did this with Dalle-E. I asked it to write "God Bless Ukraine". As you can see the text is garbeled.

[placid coyote]

dalle3 is not perfect with text, though usually better than flux1 dev (at least for me). use the wrench to input prompt for image gen ai directly. tested it now and all looks okay

[harsh verge]

Here is an example with text in the image.
The prompt is:

"A neon sign with the word 'ULTRA' glowing brightly in vibrant red neon lights, outlined with a glowing green border. The sign is horizontal and prominently displayed against a dark night time background, mounted on the side of a building. The neon tubing has a slightly vintage look, with realistic reflections on the wall and subtle details like wiring and supports. The overall vibe is classic and eye-catching, evoking the atmosphere of a retro diner or late-night establishment."

[autumn meteor]

I put the same thing in Dalle-E. I made this.

https://flux-ai.io/flux-ai-image-generator/ seems to work better

[harsh verge]

Flux.1 dev demonstrates superior performance over DALL·E in text generation, realism, and detail, as well as technical performance.
Superior prompt adherence, exceptional visual quality, complex scene handling, dynamic lighting and atmosphere, improved human anatomy, etc.

[bright parrot]

1. which model is recommended for academic article writing?
2. once i start a chat with a specific model can i change the model in between of the chat? if yes, will it have a negative impact over my results?

please someone if answers these two questions. do mention me

[mental plover]

1. i'd say Claude 3.5 Sonnet or 03 Mini
2. test it 🙂

[frozen roost]

gm everyone

[lusty stream]

I dont find the image generation ai model selection option anymore. Did it get removed?

[harsh verge]

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

[digital dust]

I would say: gpt-40

while I did not have a chance of use the o3 model, maybe it will work better

claude is better at coding

[severe crystal]

Interesting comparison between Perplexity and Google Deep Research. PErplexity comes out a little better. https://www.youtube.com/watch?v=te6GqOWkHxY

[cobalt monolith]

https://www.youtube.com/watch?v=27AxmEh3qEA

https://github.com/dzhng/deep-research

[versed plaza]

My million dollar question "how is babby formed"

[hasty basin]

If only I was in the US to take part in the sweepstake

$1M is crazy though, I didn't realise perplexity was that successful

[keen prairie]

Slide me 3K atleast 🫡🫡🫡🫡🫡

[high pulsar]

Jeff Bezos and Nvidia invested in this lol

[keen prairie]

lol

[muted jetty]

#📢│announcements message why US only? 🤨

[high pulsar]

Also Perplexity is valued at 9 billion Dollar, funded 500 mio. Dollar last funding round

[abstract ibex]

My million dollars question will be "Why only for the US"

[sweet ember]

haha

[cobalt monolith]

its for the US only?

damn

I thought I had a shot at being a millionaire

[sleek vortex]

💀 literally giving away money

what a joke

[placid coyote]

okay, I officially don't know what the x post meant...

Perplexity now offers file and image uploads with an expanded context window of 1 million tokens. Free for all signed in users in "Auto" mode.

I don't see any difference in poor performance compared to before the "update" 🙄
https://monnef.gitlab.io/by-ai/2025/pplx_M_ctx

if it doesn't work for uploaded files, why formulate the text this way? it looks like it should work with file uploads
and even if it isn't supposed to work with file uploads, what is the use case then, when any text longer than 8k characters (~2k tokens) turns into an "uploaded file"?

[oblique parcel]

How is the team going to avoid people outside the US in the giveaway using VPNs? 🤔

[fast meteor]

some form of verification/kyc before redeeming 1mil?

[sleek vortex]

yeah probably

they wont give the money to a non-real or non-usa person

[short gale]

on top of the ~$6-8M they’re paying just for the commercial

[fast meteor]

5. Winner Verification: Each potential Winner may be required to prove eligibility, including, without limitation, proof of age, residence and identity, which may include submitting a copy of a driver’s license, passport or similar government issued identification within three (3) days of being notified of being a potential Winner.

A potential Winner (or any alternate potential Winner if one must be selected) is not official and shall not be confirmed as the Winner until he/she has completed, signed and returned documentation required by Perplexity including, but not limited to, a notarized Affidavit of Eligibility, release of liability, publicity release form, and a completed IRS W-9 form, and the eligibility of the potential Winner has been verified.

[silent patrol]

How can I use the real time camera on Perplexity just like in ChatGPT?

[fast meteor]

i think this marketing is okay, as long as the churn rate is not so terrible

You can have 26 total entries. 1 for entering the sweepstakes, 5 for asking questions, and 20 for **referring **friends.

[cosmic forum]

Lmao that sounds very unsuspicious, why does even Preplexity team wuants to offer 1 million 😂

And why is it only in the US 😭

[cursive jacinth]

Sir its for fentanyl sir. We have to support them sir.

[abstract ibex]

to get more people to sign up obviously

[heady wing]

can we use incognito mode for the sweepstakes?

[cloud marsh]

illegally downloaded from libgen...? 😂

[frosty vessel]

lol

[cobalt monolith]

Lmaoooo

Also rip Aaron

[fast meteor]

https://www.perplexity.ai/search/ocr-ublU9zl.RxmARRDQIz8r2A
just found out omni is crazy good at ocr handwritten notes

[cobalt monolith]

What's OCR? Don't mind my ignorance

[fast meteor]

Optical character recognition

[dense heart]

And we are working on

[spice helm]

Does anyone know which version of o3 mini Perplexity is using? Also the number of params on the R1 model?

[obsidian pike]

I don't see the banner

[shy cloak]

why Perplexity can't show flowchart?

[plain schooner]

@fierce sonnet

can i dm

[cobalt monolith]

Greek?
And I don't think o3 mini is disclosed but afaik I saw the CEO saying r1 is full

[earnest latch]

Today marks the end of my free Perplexity subscription. It’s been an amazing tool for research, but as a student, I can’t afford the cost. Still, it’s a fantastic search alternative! Huge thanks to the CEO @AravSrinivas and team @perplexity_ai for this great experience! 🙏

[harsh verge]

There is an excellent extension called "Complexity" that allows you to do this in PPLX.
More info: https://www.cplx.app/

[shy cloak]

How can you have the Pro - Sonnet? on my side, I can't choose it manually in the bar except in the setting. However, I got many gpt-4o responses.

[shy cloak]

Not working. I have installed the extension already.

[hollow junco]

i think it's just the o3 mini from the api (not the dated version). Low effort.
For r1 they don't say. But judging by the speed (esp compared to their sonar huge which is just llama 405b), i'd say it's a quant at least and maybe distilled? Unless they have confirm it isn't? @signal hamlet

[harsh verge]

Did you configure the pre-prompt?

[harsh verge]

CPLX plugin called "Language Model Selector."

[shy cloak]

Got it. It's working. Very helpful. Thank you

[gentle bison]

WOW, thanks for that!

[cloud marsh]

just use claude

[spice helm]

Cypriot 🙂 wbu?
Okay thanks!

[spice helm]

Oh okay I see, thank you!

[primal ravine]

Here's a link for today's $1M sweepstakes!

1. Download the Perplexity mobile app
2. Enter for a chance to win
3. Ask a question

https://perplexity.sng.link/A6awk/cx24/r_a5e44950f5

[ancient sphinx]

They could add LLM - Le Chat by Mistral AI among the models in Perplexity in the future

• 13x faster than ChatGPT
• 100% open-source
• Completely free (vs $20/month)
• Privacy-first architecture
• Capital efficiency over hype
• Strategic government backing

[shy cloak]

I'm a Claude user for very long time. I came to Perplexity for a reason

[acoustic salmon]

Stuff like this should really be headlined more
https://www.perplexity.ai/sports/nfl

I would have used that all season lol

[fluid spoke]

Perplexity would have been amazing in the Olympics

[neon wyvern]

Download or update the Perplexity app, create an account, click on this referral link, and ask questions for a chance to win $1M! https://perplexity.sng.link/A6awk/cx24/r_64dff2fd67

[thick bough]

I'm on the app, I'm logged in, but see no link to enter

[abstract topaz]

Do i have to subscribe to pro or can i enter the challenge as a free user? I cant see the entry button on the app?

[naive sage]

I believe it's open to everyone

[unique umbra]

Is it random?

[naive sage]

Not sure why some people don't see the button...

Recommended someone try reinstalling the app in #⚡│ask-community

Reinstalling the app seems to have worked for that user

[rotund geyser]

I had to reinstall

[dry fog]

Roblox scripts never works:/

Not even when it comes to basic

https://tenor.com/view/discord-loading-buffering-waiting-to-load-gif-16464415

[pulsar gull]

add the url for perplexity to number 1

Test Your Knowledge in the Perplexity Million Dollar Question Competition

Think you have what it takes to craft the ultimate question? Join the Perplexity Million Dollar Question Competition and showcase your skills.

How to Participate:

1. Download the Perplexity app here: www.perplexity.ai to stay ahead in the competition.
2. Use this link to create your own challenging questions: Perplexity Million Dollar Question Creator.
3. Compete with others and prove your expertise.

Don’t miss your chance to stand out in this exciting challenge. Start now!

[frosty vessel]

its bc nobody cares about the nfl, they gotta add sports like mma with loyal fanbases

[azure creek]

Hello

I joined because I can use Chatgpt 4o within 1 subscription alongside with other models

[burnt sparrow]

Is it working properly now becaue it was giving me a headache a few days ago.

[azure creek]

When I have perplexity pro sub It is chat gpt Plus right? Or is it the free version with limits of chat gpt-4o and does it actually use gpt-4o mini?

[small tree]

when you accidentally waste tokens

[burnt sparrow]

Running API?

[small tree]

No I no longer use perplexity for coding tasks. It's useless

https://sourcegraph.com/cody/chat

I use cody from sourcegraph

it's amazing

[burnt sparrow]

How does this compare to co-pilot?

[small tree]

is co-pilot the microsoft ai assistant?

[burnt sparrow]

I'm also not using perplexity for coding either

The one built into VS Code.

I have a pro account for it.

[small tree]

Oh, github co-pilot

It's no longer worth it

paying $10 a month for an autocomplete

Pay Cody instead, it has built in chat + access to o3-mini

and o1

$9 a month

and it has autocomplete which is faster than co-pilot

they have a free plan you can use to test it out

200 free o3-mini chats a month and unlimited auto complete (On free tier)

[burnt sparrow]

I don't pay for it, its part of a dev package I have.

[azure creek]

Bro what in the scam is this app

[unique herald]

Download or update the Perplexity app, create an account, click on this referral link, and ask questions for a chance to win $1M! https://perplexity.sng.link/A6awk/cx24/r_8f9c505a82

[small tree]

Oh, if it's free, then github copilot is fine, but all copilot is is chatgpt 4o turbo isn't it?

[azure creek]

I bought pro specially to get gpt Plus included

But it's not gpt at all???

[burnt sparrow]

o3 is a newer model for reasoning only.

Yes its o3

[small tree]

it's a neutered version of GPT

it's inferior

[azure creek]

But it said gpt-4o when I paid subscription

[small tree]

a handicapped disabled version of r1, gpt

[burnt sparrow]

Anyway, I'll look into what you said Gordon, but I have a OpenAI account along with this. So as of right now I'm not really paying for much.

[azure creek]

What a scam. Only good for coders and not for ordinary people

[burnt sparrow]

I need things for research, not just coding.

Rebel you have other models you can use its not just o3

[small tree]

Yea, perplexity is great for searching

that's it

[burnt sparrow]

yep

[small tree]

it used to be good for coding too

not anymore

[burnt sparrow]

Thats all I need is to find abstracts for models.

[azure creek]

Chat gpt is the most versatile out of all AI bots I tested

[burnt sparrow]

I kind of have a pipeline going, but I'm going to go over to Azure soon since I have credit and just my own models.

[azure creek]

Copilot has gone down , while it used to be my favorite

[naive sage]

o3 mini is better than 4o

[azure creek]

But o3 mini it says it's for mathematics and coding

[burnt sparrow]

Rebel when you say "tested" are you saying from a consumer side of things because if you go on Ollama and Hugging Face there are far superior models to what OpenAI makes. If you know how to run models.

[small tree]

yea o3 excels in stem tasks

[azure creek]

I mean as a random human customer with above average IQ

[small tree]

superior models in what sense, fine tuned models for specific tasks? Or there are generally superior models?

Only thing that comes close and even surpasses openai in a lot of cases is DSR1 671B

[azure creek]

All I wanted was Chatgpt plus with extra options/chat bots to try and I didn't get that... I got a standalone chat bot using AI bots itself and then transcribe it and respond that to my inquirys

Well I got it for 12 months so I will see what it could do for me

:/

[small tree]

@azure creek Yea, perplexity was disappointing for my use case, I wanted 1 single subscription, now I need multiple lol

They used to let you select models and it'd produce direct outputs without tampering and modification by Perplexity

[azure creek]

Same I just want Chatgpt plus and I can't afford it now lol

[small tree]

Now perplexity limits its max output tokens. Limits its input tokens

[azure creek]

Even when being pro subscription

?

[small tree]

Perplexity gave away its pro subscriptions and then they had to rush to save money because they are burning through investor money

R1 kind of saved their asses

if R1 wasn't released, perplexity would be 6 feet under

they removed o1 because they couldnt afford it

did you see how fast they took the open source model, rebranded it and released it as "Sonar Reasoning"

[azure creek]

What model should I use that comes closest to Chatgpt 4o with R1 and web search , without limitation

[small tree]

yea, pro is actually free tier, their "quota" means nothing

[azure creek]

How they have 4.8/5 stars on Play store.... Probably fake reviews

[small tree]

It's not that, it's the users don't know better so they dont demand better

We're the "Edge case" users

[azure creek]

That's just a scam at the end of the road

[small tree]

they dont care about what we want

[azure creek]

Damn

I got refunded by Google play

F yeah

Free subscription lol

[small tree]

Most perplexity users were given a free pro subscription, users who have never used AI models, so they interact with this Level 1 AI chatbot thinking it's impressive

[small tree]

well done

[azure creek]

Ohh I get it , just like all the morons using fake chat gpt apps like chat smith and stuff. But actually for me it's good. The less people use AI the smarter and better I become lol. Chatgpt is my best friend and changed my life forever. I mean it when I say this.

It's like crypto I feel. When I told everyone to buy in 2017 nobody cared. Now they tell me to buy lmao

[small tree]

Yea, you're ahead of the curve, use the tool to your advantage

[azure creek]

I literally won several cases in court without lawyer thanks to Chatgpt. Of course you have to input the good prompts and triple check. And double check with copilot for example. They can correct each other lol

[naive sage]

o3's MMLU benchmark (92.3) also outperforms 4o (I've seen it recorded from 82-90, based on improvements over time perhaps).. what tasks do you think 4o is better at? As far as I've seen, o3 consistently beats 4o, or can you find a source where that's not the case.

[azure creek]

No it's just said that in perplexity app. It says that o3 is meant for other sruff

o3-mini (the version in perplexity) Limitations:
Lacks multimodal capabilities (e.g., image or audio processing) and creativity in language generation compared to GPT-4o

Primarily designed for reasoning tasks rather than general knowledge or creative writing.

So I can't upload 40 pages long pdf file and make a 40 pages long pdf as a reply using all the context from the upload and my prompts

[naive sage]

The only thing I've ever seen do 40 page replies is deep research stuff (from watching reviews of it)

But right now you need to pay $200/month with a limit of 100 of those queries per month in ChatGPT for that, quite steep... though costs will come down over time.

[azure creek]

I made whole ass court documents without lawyer

But I'm from Europe and I think it's different than in USA

[naive sage]

Was that with a custom GPT?

[azure creek]

No just the official app

[naive sage]

(Some lawyers have gotten into trouble doing that, because it hallucinates references and stuff)

[azure creek]

But I had to wait every day for like 12 hours to continue

[naive sage]

(There are some law-related custom GPTs though, which improve it a bit)

[azure creek]

Those are specialized for USA, I found one early on the Ai bot craze. In my country the whole law is online so makes it easy for Chatgpt 4o to do it's thing

[naive sage]

I don't trust it for that purpose, unfortunately. Would have to do a ton of proof-reading and research still... the truthfulness of any model isn't there, but some things have fact checking features now (I think Google tries to do fact checking by validating against web results, and highlights which page had a match - but they still warn you to go to the reference to validate manually)

Here's one example of it, there are several: https://www.legalcheek.com/2025/02/another-lawyer-faces-chatgpt-trouble/

[azure creek]

Yeah I Always let several AI , mostly copilot Chatgpt and deepseek check for errors back and forth lol . I sent it to my lawyer and it was 100% accurate and impressed her.

[azure creek]

Law in usa is very very different. Here in Belgium it's not the same at all. Such mistakes are impossible for instance.

But I understand your point

[naive sage]

That was in Australia

[azure creek]

Oh ok well that's also a whole ass different system

In my country there is no such thing as refer to other cases

[naive sage]

Eek, mistakes are impossible? Hmm... I see errors sometimes when asking queries still.

Would be nice if fact checking features become more ubiquitous, that'd help some.

[azure creek]

Well I didn't have any mistake in the 5 lawsuits I won so...

But like I said every single law and royal arrest is public and easy to check. So ai can't make much of a mistake

It's official source website of my government

[azure creek]

True. Lots of people gonna lose their jobs haha

It was a nice little chat, 2 am right now for me. Good night 🙂

[naive sage]

One of my coworkers has a habit of just replying with ChatGPT responses to people at work, it's super obvious lol

[azure creek]

Yes that's the wrong way

[naive sage]

Oh wow, it's late there lol - good night 🙂

[azure creek]

You should correct it and perfection it several times

[azure creek]

Haha yeah

One of my favorite prompt when writing a reply email is : write is as if I have below average IQ and used X kind of dialect (becomes very realistic hahaha) just wanted to add that . Byee 😌👋

[somber bobcat]

Hi I have referred a ton of people tonight for the sweepstakes but I see 0 referrals in my account on my laptop = help! Thank you!!!

[naive sage]

Might need to make sure they are installing the app, logged in, and performing at least one search as well.

But if that's all done then I'm not sure, have not tested that.

[naive sage]

Ask what year is most likely for it to start

Most people say 2027

[acoustic salmon]

https://tenor.com/view/ryceballz-oof-oof-size-gif-21066840

[hazy lantern]

@naive sage https://www.perplexity.ai/search/run-a-simulation-on-the-best-a-pgNSx0m0S7GWqLz.vz_AxQ

[frosty vessel]

Perplexity has massive potential to completely take over Deep Research. I'm not sure why they're so insistant on solely becoming the next Google

[small tree]

google can destroy them with the snap of their fingers

look at gemeni 2.0

[minor shard]

Agreed. There really isn't anything Perplexity is doing that another company can't replicate.

Always feels like they are in catchup mode

[azure mauve]

I don't see the Focus option anymore, why?

[cobalt monolith]

It was changed to togglables modes

[thin summit]

Perplexity was a one-stop shop, but does feel more generic now. Beating Google would just be the start. We'll see if Big G can get out of their own way, fix their search, and then make their AI models more comfortably useful.

• I mostly get value out of Gemini through AI Studio, not through the main app; the only exception is with transcription e.g. "take a note"

Somewhat related, though, you can select the models in your Spaces.

[ripe warren]

Does perplexity partner with all the top AIs to formulate an answer or does it just decide which AI model would be the best fit for your inquiry

[thin summit]

That second thing. Light sentiment analysis of the query, and send it to the one best suited for it based on that heuristic

[limber saddle]

Do you work for Perplexity? I've had this sneaking suspicion that Auto heavily favors whatever option is cheapest for Perplexity.

[ripe warren]

Cheapest? They pay them every for every time they're used? Its not like a yearly license thing?

[limber saddle]

Different models cost different amounts per input and output

[ripe warren]

Ah cool, didnt know that

Did the winner get announced yet?

For the $1 million sweepstake

[abstract ibex]

Just discovered that you can create pages with perplexity, that's a very cool feature. I can send personalized articles about any topic to my family!

[tame current]

[harsh verge]

I think you got your answer. 😉 (Don't forget to update → v1.6.1.0)

[tropic beacon]

omg! im glad that ppx brought grok into their selection of models because other models are so sensitive that they dont want to help me assess my writing

[green crystal]

not with their search being so bad

[limber quest]

Yo Buddy

[cobalt monolith]

damn what did u write

[visual latch]

Hi, I have activated perplexity ai pro but it is not yet activated. i hve waited for 2 busoiness days, sam from perplexity is not helping me yet. i dont want to lose money. if it happens i wil lose trust

[burnt sparrow]

For doc writing and research so far this has been decent. As for models, if you know python technically anyone can up their own models if they have enough compute power. But then you also have to maintain all of it yourself which is a ton of work.

What I would like to know is which model is best for writing manuscripts in latex.

[thin summit]

They said that's what they did a while back, but I don't have a reference. When I had an AI job, it's what I did, so it just made sense

[burnt sparrow]

Cost vs performance trade off is how you make money.

[wind bramble]

youtube video by @karpathy “Deep Dive into LLMs like ChatGPT” turned into text & interactive Q&A with answers grounded in the video with http://cofyt.app

[hallow flume]

where i can post polls?

[green helm]

Where can i see the limits again?

[fast meteor]

https://www.perplexity.ai/p/api/v1/user/settings

[silent patrol]

Hello, guys. How can I use the real time camera on Perplexity just like in ChatGPT?

[visual latch]

Hi friends i am happy now. I am able to use perplexity pro now. The issue is solved

[zealous blade]

hey guys how can i start using perplexity

[halcyon coral]

Hey @silent patrol!
Please check this announcement: #📢│announcements message

[lusty egret]

How can I make Perplexity replies sound more human?

For example, I want to create a feasibility study. When I ask Perplexity "Can you help me create a feasibility study?", I get "Creating a feasibility study involves systematically evaluating a project's..."

Whereas on ChatGPT, I get "Of course! What is the feasibility study for? Are you looking to analyze a business idea, a project, or something else? Let me know the details, and I’ll structure it accordingly."

[radiant quarry]

that's a great question actually

Can we perhaps prompt it to sound more human?

[lusty egret]

Yes

Am I better off with Pro and R1 off completely? I tried using Sonnet but same

[radiant quarry]

Also is it normal for it to use sources with every question? Is it what chatgpt does but behind the scenes

[naive sage]

If you turn off the Web, Academic, and Social options then I don't believe it uses sources.

It'll be more like just talking to the model directly at that point, similar to the other platforms

[lusty egret]

Thank you

[zealous blade]

#💬│general how can i start making use of perplexity am new

[wanton drum]

Hello guys good night

[split night]

the fileupload is very frustrating

i have a code in which i want it to make some changes

but as the code is 400 lines

i cant paste it and it goes as file

now it wont read the file properly. and rewrites the codes in its own way

which is frustrating

[gaunt beacon]

Well thats the thing with ai, i suggest sending only the snippet you want edited

Or if you have some money spend in api keys and then use a tool like aider to code

[split night]

they should let it send a text

rather than file

i mean 400 is not big of lines

[gaunt beacon]

Well yes and no, in 400 lines you can have 40.000 characters of code and 400 characters of code

[gaunt beacon]

Also what mosel are you using in my experiences even when the file uploads it works okay?

[split night]

i am using deepseek r1

it wasnt working with anything

so i had to copy paste it to the deepseek website

and it worked

so problem isn not with deepseek

[gaunt beacon]

Well as far as i know, perplexity does different algorithms with the llm

[split night]

yeah

i can see it

pretty sure the file is just read brifley beucase of which the code they give out next is all over the place when the instruction is clear

cant really send the snippet and changing one thing requires more than one part in the source to be adjusted

and thats also the reason we are using ai

[forest mortar]

Anyone else got the issue of copying since the new perplexity update? i usually copy bits and pieces of text for a quick ctrlc ctrlv, but since the new feature of looking things up directly from a response, its become really buggy and hard to distinguish

[thick lion]

When Perplexity is gonna have something like deep research like OpenAI?

[naive sage]

They're probably working on it. You can kind of do an emulated deep research "lite" with some manual work by having an LLM break the question into several and then asking each one separately.

[placid coyote]

it is a known limitation of perplexity, their RAG (file handling) is one of the weakest points. for more details see: https://monnef.gitlab.io/by-ai/2025/pplx-tech-props

[serene sand]

welcome on board!

deep dive ➤ #💬│general message

[lament epoch]

I am totally not able to setup a new thread on perplexity

it picks up content form previous searches

how to fix this

[split night]

not good

[serene sand]

https://discord.com/channels/1047197230748151888/1140622008086970420

please share your examples, threads in this chanel, thank you

[green crystal]

Anyone seeing perplexity automatically switch from deepseek to standard pro search after a query?

[weary crescent]

@vapid onyx not sure what perms you have if any but I think this guy is trying to scam

[buoyant parrot]

Perplexity pro worth it?

[tame current]

👍🏻

[buoyant parrot]

Better then the payed version of gpt?

[acoustic salmon]

depends on what you use it for

[gilded finch]

Is 01 better than 03

[abstract ibex]

Omg dude please teach me all your secrets I want to be rich like you without doing anything and evading taxes.

[naive sage]

Do not envy the wicked in life

[abstract ibex]

Let me annoy the scammers

[umbral coral]

Malicious

sound:0:3

[vapid onyx]

Handled, thank you!

[solid star]

In certain stuff yes, but depends. Both have strength and weaknesses. O3 mini high beats o1 in coding while o1 beats o3 mini high in reasoning.

[naive sage]

(it's very close in reasoning though!)

[shrewd pollen]

why is my perplexity in chrome lagging like hell

its getting stuck also

[feral gulch]

Who won the $1M ?

[stiff bone]

1M?

[azure mauve]

So there is no writing mode now? Or search reddit mode?

[unborn flicker]

are we sure the data is safe and not going to the ccp? 😂 😂

[cobalt monolith]

For now it seems writing mode was deleted from the web UI but exists in the app

We will have to wait and see

[azure mauve]

Very upsetting, as I used it more often for my production. Now it seems it just becoming a search engine. Need to rethink my subscription. Used perplexity for a while now. It's been a great tool for my research and writing purposes.

[cobalt monolith]

Best to ask the CEO on Twitter

[azure mauve]

Hi, @signal hamlet any words?

[harsh stag]

@cobalt monolith @azure mauve toggle off the "web" option, it's equivalent to the old writing focus

[rare widget]

Hello

[chilly pawn]

is there any way to make AI login using my university login details so it has access to all paywalled journals?

[pulsar drift]

Why can't i use o3 mini with pictures

[wraith crow]

o3 mini does not support vision based tasks

[pulsar drift]

didn't it use to? Does any reasoning model support pictures?

[drowsy harness]

O1 support images

[wind sentinel]

yeah, who did?

[hallow flume]

Is there any hint, that Perplexity get its own App for Windows in Future?

[kind rampart]

The web interface of Perplexity has been unable to use R1 these past few days. Although the results indicate they're generated by R1, there's actually no thinking process involved.

[indigo peak]

Hi, new pro subscriber here. Liking everything so far except the three UI elements beneath the ask anything bar. How can I remove them?

[harsh stag]

There's no way to hide them unless you force it through browser extension

[indigo peak]

Ty! I've blocked the elements. They still work if i click, but I don't do that and they're not throwing ad garbage in my face on launch which I prize most of all. Stupid that Pro tier does not have a toggle for this.

[green crystal]

Is that Jian yiang?

Ah it's him

It'll work out just fine

[dark wren]

in my experience r1 is better in coding than o1 or o3 mini

[green crystal]

After all his corrupt uncle in Beijing will help us out

[dark wren]

it better

[dark wren]

🫠

[green crystal]

[blazing ingot]

O

[stone furnace]

oi Perplexity family! Just joined the discord and have signed up for Pro. Really enjoying the reasoning models with Perplexity search and looking forward to contributing as I find my way around with it. Look forward to meeting some of y'all, cheers!

and a big thank you to the Perplexity team for creating such a nice UI and set of tools for everyone.

[twilit dragon]

pls play its diffrent Join at www.kahoot.it Game PIN: 6849640

[proper mural]

Complexity addon on ur browser

[bright jungle]

@cobalt monolith Anime Wolfboy, black and gray toned, with victorian england era style clothes, with golden eyes, surrounded by a dark ominous aura. Wolfboy has a cut on his face, , wolfboy has a dagger in each hand , humanoid looking

[bright jungle]

Hmm 🤔

Try one more but add 4k graphics,

[cobalt monolith]

give me the whole prompt

[bright jungle]

Draw Anime Wolfboy, black and gray toned, with victorian england era style clothes, with golden eyes, surrounded by a dark ominous aura. Wolfboy has a cut on his face, , wolfboy has a dagger in each hand , humanoid looking, 4k graphics and lighting.

[bright jungle]

Ah I see was hoping it would be on par with this one , ai I’ve been using for art

[cobalt monolith]

i guess its not really worth it honestly

[bright jungle]

Probably not for art but probably worth it for other things

Thank you for helping me test it🙂

[cobalt monolith]

np

[solid star]

Where did you test them? I used o3 mini high and o1 on chatgpt and both beat r1 in coding for me

[ocean owl]

Salut

[vale crest]

bro Sonar is insanely fast

[gray veldt]

who is sonar ?

[vale crest]

perplexity's model

here see

[gray veldt]

does really 100k+ peoples talking about AI here ?

[vale crest]

of course

[tame current]

right ? imagine shoving politics down your throat even when you're paying

[gray veldt]

that's uncomprehensible

i wasn't knowing anythink about perplexity not far about 1 week ago

[fickle sorrel]

does perpletixy use the full r1 without destilation?

[cobalt monolith]

Idk about destilation but it does use the full one but a modified one that isn't uncensored

[fickle sorrel]

is it possible to ude perplexities model just for chatting, like in openrouter, without the research/google aspect?

[cobalt monolith]

Yes just turn off web

[vale crest]

bro i literally came to ask the same question

[cobalt monolith]

It was recently made into a toggle

[fickle sorrel]

i can turn this off?

[cobalt monolith]

Yes

[vale crest]

i guess it is distilled maybe

or they are hosting it on US servers

prob

i hope it is full R1

[cobalt monolith]

They do host on us servers

It is the full r1

[vale crest]

any source

?

proving that

[fickle sorrel]

but u said it could full distilled.
iirc full means non distilled

[cobalt monolith]

I asked

Let me find it rq

[vale crest]

k

[vale crest]

yup

[cobalt monolith]

Ah

Then not distilled

Maybe it wasn't me who asked but someone let me find it on Twitter

https://twitter.com/AravSrinivas/status/1884509590684934211

Aravind Srinivas (@AravSrinivas) on X

@julheetel8890 Full.

X · Jan 29, 2025, 7:51 AM

[vale crest]

thanks for this

[cobalt monolith]

No problem

[fickle sorrel]

once u made one prompt in a new chat where u disabled browsing, u just need to ask stuff which requires browsing and it will browse tho

[cobalt monolith]

We all have the same questions

[cobalt monolith]

Are u using the web version of perplexity?

[fickle sorrel]

perplexity.ai

[cobalt monolith]

So the web version

Weird that it searches

[fickle sorrel]

u cant even disable browsing <D

[cobalt monolith]

You can

This is why they made it a toggle now

[fickle sorrel]

for a reason probably like not using it for big chats like in chatgpt

[fickle sorrel]

yeah just when u click on new thread.

[cobalt monolith]

Oh you want to change mid thread?

[fickle sorrel]

i dont want it to browse at all

i wanna use R1 😄 thats all

[cobalt monolith]

Should still work without web hmm
I guess report in bug reports or something

[jade yarrow]

is New Sonar model in Perplexity pro released yet?

[vale crest]

yes

[vale crest]

it is

[jade yarrow]

XD Seriously what is difference? I asked support but support in email said Llama 3.1 70B

[vale crest]

i am using it rn

Sonar is out and it is delivery 1200 t/s

fastest LLM

[cobalt monolith]

Is good?

[vale crest]

these should answer your question

[jade yarrow]

Support stop using AI in your support that is XD answer what you gived me

[sly pivot]

Hi, My perplexity have all permission needed to make a call to any of my contacts but it it says that he cant make a call as hedoesnthave a permission. 😏

[naive sage]

Sonar with LLAMA 3.3 just released today: https://www.perplexity.ai/hub/blog/meet-new-sonar

[sly pivot]

Aking me for a number..

[cobalt monolith]

Damn ALHAMDULILA
this sexy

[sly pivot]

i dont even know my own

[jade yarrow]

I don't what happened with Perplexity why they didn't say it in settings that is new version , only Sonar and enough

[sly pivot]

Is sonar work with aichat from sigoden?

[jade yarrow]

👏 👏 I understood XD

[naive sage]

Sounds like not everyone has access to the new Sonar yet, so maybe that's what he meant... might link him the announcement post from Perplexity and see what they say

[jade yarrow]

But I didn't know how new version looks like

[mild swift]

Can anyone explain why perplexity on desktop won't answer a basic prompt which is Hello while on mobile it does. All i get is a reminder that the app is an ai search engine

[jade yarrow]

No comments , i am waiting for new models from Perplexity

[empty kestrel]

Hey guys

is there a page on rate limits?

[abstract ibex]

Perplexity should launch a search engine based on AI to make this easier for us to find it

[empty kestrel]

lol I found it

[frozen spindle]

Anyone planning on purchasing Nvidia Digits?

[serene sand]

https://x.com/perplexity_ai/status/1889392617479082323?s=61

Perplexity (@perplexity_ai) on X

Perplexity's Sonar—built on Llama 3.3 70b—outperforms GPT-4o-mini and Claude 3.5 Haiku while matching or surpassing top models like GPT-4o and Claude 3.5 Sonnet in user satisfaction.

At 1200 tokens/second, Sonar is optimized for answer quality and speed.

X · Feb 11, 2025, 7:14 PM

https://tenor.com/view/rápido-fast-snail-robot-gif-15498737

[junior adder]

web version is extremely slow (nothing new here) and android version keeps popping "something went wrong" not letting to type in anything

[naive sage]

I wonder if it's related to ChatGPT having problems... the ChatGPT site wasn't loading the side panel or user options and was giving errors, even though I was logged in, until a couple minutes ago. Not sure how long exactly it was going on, but I noticed it about 10 minutes ago.

It is fixed now, so might try refreshing and seeing if it is still happening

[tight prism]

https://x.com/aravsrinivas/status/1889398205076451697?s=46&t=Un1yLqIRg3sDiqpmnWHBfg

Aravind Srinivas (@AravSrinivas) on X

We’ve post-trained some really good models on top of Llama 3.3 that far surpass 4o-mini and 3.5-Haiku and match 4o and 3.5-Sonnet for answer quality, despite being way cheaper and blazing fast! Users are loving it already based on retention metrics. This is thanks to our work on

X · Feb 11, 2025, 7:36 PM

[abstract ibex]

Me 🫡

(guess the type of scam in the comments chat)

My guess is: "I just need money to send it"

Bingo

Whoever guessed the same: gg

[naive sage]

Quincey "fisher"... 🤔

[abstract ibex]

Poetic

[naive sage]

Also a new account... 🤔

[abstract ibex]

It's obviously a scam, you give them money to pay for shipping and that's it

Oldest trick in the book