💬│general

[wheat dagger]

Might work , but not efficient enough , open ai should make more efforts into making their models compatible with all sorts of files extensions

[neat relic]

bro my o3 mini answer it wrong

wtf

[lethal glacier]

nah should work I've given it code that was supposed to generate a pdf with this "text" o1 understood all of it(manipulated it into how I wanted it), its just transferring the pdf/file you have now to text that's hard that represents that exact format, if there is no coding library that can generate the pdf or powerpoint you have then yeah oof, I agree they should allow attachments for every file extension

[split dome]

Perplexity using o3-mini low-medium

o3-mini low & medium 👎 👎 👎 👎 👎 👎 👎 👎 👎 👎 👎 👎
Bring back o1
Perplexity using o3-mini low & medium
Use o3-mini High or bring back o1

[neat relic]

o3 mini low can answer it, but this o3 mini perplexity idk its literally broken

[split dome]

o1 is better than perplexity o3-mini

[neat relic]

they will not get back to o1

[split dome]

yeah .. perplexity o3-mini is useless

[neat relic]

@median parrot can you check this please,

[split dome]

they should use o3-mini high

[hollow pelican]

At this point just bring back o1. It worked so much better.

[waxen lion]

Perplexity be taking massive L lately

[hollow pelican]

Yeah fr tho.

Im looking into alternatives that function pretty much the same but I’m struggling to find any.

[empty snow]

o3 mini by perplexity is literally the worst o3 that I ever tried. I think that they nerfed it a lot

Problems that are resolved everywhere perplexity o3 still fails

[sage chasm]

How do you turn off reasoning?

It's completely ridiculous, takes 20 seconds for a simple search

[agile blaze]

Hi, so guys currently what is the best model to use in perplexity? Deepseek with R1 ?

[empty snow]

@ok.alex it's possible that everyone tells you that there is some problem with perplexity o3 and you (not literally you) you pretend not to hear community feedback?

[empty snow]

Probably

[agile blaze]

also, do you usually activate web search or not ? (if no recent data is needed)

[empty snow]

Ya

Only If I have to test models I disable web search

[agile blaze]

Are answers better with web search ? I fear that bad quality web pages could "pollute" answers (then again, they are probably already in the training)

[hollow pelican]

You what actually happened to perplexity? None of my searches are accurate right now.

[neat relic]

perplexity now is f4ck.

[hollow pelican]

Is there an alternative i can use? Cause its only been a few days and perplexity doesn’t seem to work anymore.

[neat relic]

for web search like perplexity?

[hollow pelican]

Yea

[neat relic]

u can use gemini or chatgpt. its free

[hollow pelican]

I don’t really like chatgpt fr, but thank you.

[tepid sequoia]

The test for the answer to this example is ambiguous.
if 2 + 3 = 127 and 7 + 2 = 88, then 5 + 4 = ? explain in details for the answer

I tried it on the o3-mini API - first it didn't calculate it, then it did.
I tried o1 API - first yes, then no, and then yes again.
o3-mini Perplexity first yes, then no.
R1 Perplexity I tried only once and it calculated it.
ChatGPT free o3-mini - first no, then yes.

So the result is 6566.

[placid steeple]

hey there

[neat relic]

u must give some hint tho, so they can answer consistently

i give them example how to do it, and then analyze how i do it and then they solve it, o1 mini, o1 preview, o1, deepseek, o3 mini. they all can answer it

[tepid sequoia]

Alright, and do you happen to have any example that can be handled by whatever O3-mini and the platform it uses? Because I have plenty of difficult examples that a genuinely good model can tackle, but I don’t have anything truly unique for testing.

[neat relic]

no, i just want to know why o3 mini in perplexity cant answer my math puzzle

[chrome crystal]

if u send them over to me, i can test them on o3-mini-high

[tepid sequoia]

That’s exactly the thing: I have API access to the models, and they’re supposed to be the highest versions, but for instance, with o3-mini I only have it as o3-mini. However, if you check platform.openai, it just shows o3-mini with a subcategory o3-mini-2025-01-31. So it’s quite possible that some platforms have o3-mini-high under o3-mini as well. Therefore, the way to figure that out is to test it with an example or some unambiguous answer. But so far, from what I’ve tried, they either all responded “Let’s do it,” or answered somewhat vaguely, both correctly and incorrectly. Hence my question about something unique that would demonstrate the strengths of, for example, o1 or o3-mini-high.

[chrome crystal]

ahh i see what you mean. the high models are very similar to one another so testing will probably give the same results. i do know that lmarena.ai has prompts you can use under the “arena explorer” section. you can give those a try. good luck with your testing!

[tepid sequoia]

By the way, when you talk about lmarena, is it known how it is financed? Or do they have premium access to models from the creators for free testing?

[jagged canopy]

I have a question... I'm certainly a newbie here and I've only been using perplexity for the past 48 hours. However, I am apaying pro member. Why can I use push to talk for free searches but I can't figure out how to use push to talk for premium pro searches? Like the orange button that you hold for regular free searches is not available for pro searches. Am I missing something?

[hollow pelican]

Im glad I’m not the only one who also uses this site for rp sometimes.

[short sluice]

Me too

[hollow pelican]

Has it been fuckey to rp for you too? Cause its been struggling with me trying to.

[short sluice]

Yes. The R1 keeps turning on and I can’t do anything without it getting all smart if that makes sense

[hollow pelican]

That’s exactly what’s happening with me

[short sluice]

And I just paid for pro and that’s when it started doing that. I got the first month free and then I paid and it started acting up

[hollow pelican]

Im gonna try to upload files with the memory of what i was doing and try it with pro turned off. See if that makes a difference. Less anyone smarter than me has a solution that makes me sound less dumb, lol.

[chrome crystal]

i don’t exactly know but what i do know is that their biggest partner is UC Berkeley Skylab. they both created lmarena together. they have other sponsorship as well just not as close as UC. their sponsorship is how they are running the models on the website. i don’t really like lmarena’s results because it’s all a voting based system

[tepid sequoia]

I don't really like it either, especially when I see which models are in a higher category and which are not, even though I have tested those models myself and I know that they are not that good or bad.

[waxen lion]

What's the review for gemini pro

[chrome crystal]

the thing about lmarena is it’s based off how it writes, not its knowledge. your running tests based off what people think, not how it performs. i’ve also tested most of the models that are rated highly on their website but my testing shows it’s unreliable knowledge wise

[acoustic salmon]

What do you want to know about it?

[mint tusk]

even with pro there's no way to exceed 25 mb?

[rapid rampart]

Hey everyone! I’ve got a promo code for ""Perplexity AI"" that I’d love to share.
https://perplexity.ai/pro?referral_code=YDHCWXGE

[acoustic salmon]

Gemini 2.0 is quite good, but the current Flash version does not operate at the same level as R1, o1, o3-mini. The benefit of Gemini Pro is its integration with the rest of the Google suite like Gmail, Docs, Home, and it's capability as an assistant.

[chrome crystal]

it’s funny how most people signed up for gemini for the workspace features but you can get it for free if you use their google labs feature

[acoustic salmon]

for me it comes with storage so

it's just an added feature I'm not directly paying for atm

[chrome crystal]

yea definitely. nothing beats the perks of gemini’s plan

unsubscribed from them and now i’m paying for icloud + chatgpt

still worth the switch tho

[acoustic salmon]

I have GPT Plus still, until like Feb 15 or something

I switched to Perplexity couple weeks ago to really dive into its features

but I can only justify paying for one of these at a time lol

[stiff river]

Well, well, well, what a surprise (not) with DeepSeek:

[chrome crystal]

same, i’m just trying out perplexity but the bugs are killing me. i don’t need r1 reasoning to be on all the time

#1335055295302336525

[stiff river]

First answer with DeepThink, second without.

[mint tusk]

also how long does perplexity keep your files you upload?

[chrome crystal]

me too. i know im keeping chatgpt for sure but questioning if i should add perplexity as well

[acoustic salmon]

yeah, I can definitely understand the frustration with the bugs, shifting models, and the fast changes that drop. Just been along for the ride so far while I wait for it to settle down

maybe because I'm a frequent Steam Early Access player I'm just not as affected by all the chaos lol

[chrome crystal]

ikr, i wish they would notify about any changes at least

[chrome crystal]

stream early access?

[acoustic salmon]

Steam, the gaming platform. They'll allow developers to release their games in Early Access (meaning Beta) and it'a always a chaotic ride. Bugs, overhauls, dropped support, etc.

[acoustic salmon]

Yeah I upvoted Patch Notes suggestions

[chrome crystal]

ohhhh that makes sense. i was confused what you were talking about lol.

[chrome crystal]

where is that so i can upvote it too?

[acoustic salmon]

https://discord.com/channels/1047197230748151888/1301576498200711239

[placid coyote]

😑

[chrome crystal]

don’t have access since they didn’t give me the pro role yet 😞

[acoustic salmon]

ah sorry

[hollow pelican]

Im baffled. Did they ruin their website in the span of a few days?

[small tree]

there's a parameter you need to set

by default o3-mini api version is medium

but you need to set it to high

"reasoning_effort": 'high'

[surreal adder]

Hey

[left stream]

I think I have an idea on how perplexity could fix the o3 mini medium model

Realized it happens to o3 mini medium reasoning on you.com too when search features is activated

[small tree]

if it doesnt make perplexity more money

theyre not intereted

[empty snow]

use the real one? 😅

[small tree]

interested*

[left stream]

They are the real ones

[small tree]

No, just use it from openai directly

[left stream]

I fixed the issue earlier with the prompt

The problem is

Too much irrelevant information leads them to making faulty answers. O3 mini medium for both you.com and perplexity can answer the question by themselves probably correct or wrong with a certain chance, but with irrelevant information being picked and given to it, the context memory just gets filled up and it’s not even the correct information. The chances of getting the right answer drop even more. A prompt that balances the information they get can fix it. It worked for me by simply doing “site:” and the space features also works well for o3 mini if you’re using complexity

[empty snow]

i dont' understand the last sentence, how to fix it?

[left stream]

There’s this thing called spaces. You can run models on there and it’ll have a prompt (instructions) into place that will tell it how to act or what sites to use

If you’re selective with what type of info it gets, it’ll be op.

As soon as I saw where it was getting info, I realized I had to be picky. Random subreddits that talked about the arabic language when the topic is about space is an example

Like this

This too

[empty snow]

so how have i to set it?

[left stream]

And this one.

When I did an all nighter testing it, it somehow worked with r1 and it became a monster. O3 mini with r1 was crazy op

But that was accidental

[left stream]

Yes

[left stream]

There should be some already set and you can use those since they give those as examples

You can use wikipedia.org, britannica, arxiv, etc.

Just a comparison to show the example of how sources can affect it

[empty snow]

ok but how have i to set this

[left stream]

Left is from you.com and it has the wrong answer. I didn’t pick the sources for it. The one on the right I did

Here’s an example

[left stream]

I did that space and it somehow made r1 and o3 mini work together. It was beautiful when it gave the answer

But it was accidental and I can’t do it again

r1 did reasoning and o3 mini did the info gathering

Here’s another example

[empty snow]

ok thank

[left stream]

No problem

[fluid nexus]

hi guys

[sharp schooner]

Guys where tf is copy answer button??

[tame current]

@everyone Hello everyone

I wanted to announce that I just launched a vs-code extension specifically for Perplexity

https://marketplace.visualstudio.com/items?itemName=ghutu.perplexity-ext

If you have any ideas or encounter any issues, feel free to raise an issue on Github and I'll have a look

[serene sand]

https://discord.com/channels/1047197230748151888/1059504969386037258

[tame current]

Let me post there as well

[wary trout]

Guys, is perplexity down? Cant seem to access and stuck at cloudflare

https://status.perplexity.com/ partial outage.. aight got it

[serene sand]

https://x.com/aravsrinivas/status/1886129399743091123?s=61

Aravind Srinivas (@AravSrinivas) on X

If Perplexity were to open source custom version of DeepSeek R1, what should they enable? Other than the lack of censored outputs. What are some flaws you see in the current DeepSeek R1 relative to Sonnet or o3?

X · Feb 2, 2025, 7:07 PM

[gilded finch]

How did you use both together

[vagrant ridge]

Hi, is there any student discount for Pro? 🙂

[stone kettle]

Can't wait for Monday 💪 let's make some good money

[hollow pelican]

What are you talking about?

[stone kettle]

We do not want to come off as spam. If you know you know. Or you could check the server

[tepid sequoia]

There's a problem here. I gave a riddle to various o3-mini models in different services. Honestly, I don't know which of them are low, medium, or high, but all of them solved the riddle correctly except for Perplexity o3-mini. But for example, the only thing I know is that ChatGPT free is probably low or medium, but it also solved the riddle. So again, there's a problem that Perplexity is lying about which model it's using. To be clear, I gave the riddle to each model 3 times to solve.

A valiant knight decided to marry a beautiful princess, who is guarded in her cave by a terrible dragon. The dragon has 3 green heads and 3 red heads. But after beheading any head, more grow back to the dragon. The knight has a large sword with which he can cut off one or two heads at once, but always only of the same color (the heads are grouped in clusters by color) – After cutting off one red head, two red heads immediately grow back. – After cutting off two red heads, one green head grows back – After cutting off one green head, another green head grows back again – After cutting off two green heads, nothing grows back. What is the minimum number of sword swings the knight must make to kill the two-colored dragon?

[latent thicket]

help with refund?

[gilded finch]

how do you get student discount

[harsh verge]

In ChatGPT, o3-mini uses medium reasoning effort by default.

[tepid sequoia]

But I'm saying that this is not o3-mini at all. Or I don't know, maybe it switched to a different GPT-4 again, like it was a few days ago with o1.

[harsh verge]

I was just responding to your: "the only thing I know is that ChatGPT free is probably low or medium."

[acoustic salmon]

Perplexity's models also have Perplexity's stuff running on top/alongside them

I imagine that creates some differences in performance as well

[solid star]

Beautiful

[quaint moon]

Hello, I need some help. I just paid for pro through google play but my account didn't change

[solid star]

O3 mini answers way better now. Took me a while but o3 mini high’s recommendation worked

I uploaded 6 files that explains o3 mini and how it works, how the search feature for chatgpt works, and how o3 mini uses the search feature. I uploaded those 6 files and had gpt4o in charge of them. Basically following the right recipe to make o3 mini effective in search

Worked tbh

[sleek vortex]

👀

[solid star]

This just keeps getting better and better

[merry karma]

Competition at its finest

One company makes a cheaper and better product, everyone else scrambles to join them and do everything to lower their cost to rejoin the competition

All benefits Perplexity in the end that uses all of them

[harsh verge]

Just to be sure, you did use the same Google account for Google Pay and Perplexity, right?
Try here for assistance: [e-mail]

[quaint moon]

I just checked and I didn't

I sent an email, thanks for your response.

[harsh verge]

I suspected as much, use the email address above to make the change. 😉

[quaint moon]

Thanks

[dawn goblet]

o3 mini vs R1?

[solid star]

I wonder how much things will change once o3 is officially out. Still can’t imagine how crazy it’ll be

[short narwhal]

I’ve preferred o3’s answers so far on most topics

And it’s faster

--dayNUMERICAL.MON--

[merry karma]

Is Pro worth it? $20/month is kinda steep

But it's the same price as Gemini, ChatGPT, etc

[wary trout]

perplexity web is still getting issues on brave browser getting pass cloudflare :/

[haughty bramble]

Better compared to what?

[sleek vortex]

https://www.youtube.com/watch?v=jv-lpIsnLOo
👀

[vapid onyx]

I like using Pro because of the access to models offered by others AND more reasoning

[raw apex]

Quick question. It would seem to me that non pro queries have an over reliance on Wikipedia. Is there a way to limit, perhaps temporarily, Perplexity's ability to source from Wikipedia?

[abstract ibex]

Can you share it?

[worldly patrol]

what is the current quota for Pro Search with R1 per day? is is 25 or is it 500/day?

[torn sierra]

Perplexity started adapting language models early on to integrate them with a search engine. Now, OpenAI and Google have models for deep search. What OpenAI has just shown looks great – too bad it's only available in the Pro version for now. I wonder how Perplexity will respond… 😉

[cursive jacinth]

did someone understand this guy sir?

[shadow pike]

will perplexity really be free forever?

[cursive jacinth]

how did you edit that sir? thats the question sir.

[shadow pike]

[short sluice]

Hi

[nova zinc]

how do you select o3?

[tawny flare]

yk u can get it for 35-40 a year. there are promos

[indigo valve]

Which I think can be fixed with complexity

[merry karma]

I will keep an eye out for that

[minor sequoia]

Auto mode appeared

AUTOBOTS

[crystal cosmos]

r1

[short sluice]

Hi

[chrome crystal]

cora did u redownload perplexity?

and see if it fixed the r1 bug

[short sluice]

I can’t because I don’t know my account password and I will get logged out :(

[minor sequoia]

For code I would choose o3, but for others gonna use r1

But does perplexity give us r1 big model? Or smol model?

[merry karma]

It never asked me to provide a password. Just a OTP 😶

[merry karma]

Wherrrreee

The only mentions I see online are people selling their year codes which sounds...a bit illegal smh

[nova zinc]

^

[hallow tulip]

What is happening wiht perplexity right now?

[sinful canyon]

Is there any way to get o1 in the iOS app?

[random ocean]

O1 mini is available only in web as of now.

[sinful canyon]

Thanks. Why is that?

[shut brook]

WTF happened to all the models!?

[earnest widget]

any idea if there's a URL param to enable reasoning? I know we can do copilot=true which only turns on pro mode

[opal panther]

fr

i think they're fixing some thing

[jovial kayak]

I'm thinking about resubscribing to perplexity pro after realizing some stuff, and I just wanted to know the answer to this question cause it might make me lean like 99% towards Perplexity. Are there any active message quotas for any of the models? I asked several times to Perplexity and repeatedly was told that everything was unlimited, including the pro models cause it's 300+ (not that I'll even near 50), but especially if there's one for o3-mini which I see is an option?

[quiet cradle]

is there any way i can revert my pro search bar back to normal

[modern sinew]

Yes

[kind rampart]

Where can I find the daily usage statistics for each large model in Perplexity?

[crystal cosmos]

isnt this perplexity thing

[glad otter]

Roughly 500 daily messages i never hit quota once. Perplexity not good at long form or coding requirements due to it just cutting off after xx tokens used.

[gusty wasp]

how do i even turn this off now

[wary trout]

Fixed. Apparently addons were the caused. Lucky there's 1 click disable all extensions temporarily

[hallow flume]

Click on it i guess

[gusty wasp]

i guess 'auto' is when its turnd off

[meager folio]

I've been using perplexity for helping me translate things accurately because Google translate doesn't have the option. I can't type anything or send anything in the answer bar and it says "Pro". What happened? Sorry if this doesn't make sense, my English isnt amazing.

[jovial kayak]

Last question hopefully, can anyone who has pro tell me if there are any usage limits for the voice mode? If that’s also unlimited or is based off queries and not time use, I’m probably gonna immediately switch to pro from plus

[signal storm]

hello

can anyone give me the referal code of perplixity?

[runic gale]

ughhh is there any way i can disable reasoning for prosearch? its messing with how i make stories and i hate it

i want back the old pro search or just give me the option to turn off reasoning

[indigo valve]

Awesome I'm glad it was fixed. The add-ons can be screwy

[modern sinew]

Nuh the new ui is so bad

[tame current]

https://perplexity.ai/pro?referral_code=FPNKJUH1

[slender halo]

Hi Perplexity team and fellow users! This is Yashvardhan, a pre-final year student at the Indian Institute of Technology Guwahati, and also the Head of Marketing at IIT Guwahati's flagship annual technical fest, Techniche!

I am writing here to hopefully get in touch with the Perplexity team and explore potential partnerships with Techniche, IIT Guwahati. We have a huge community of tech-enthusiasts and we love gaining and sharing knowledge. It would be amazing if you folks can help me connect with the relevant person or team for this.

[still trail]

rivoluzione-femminista/727084/

[tame current]

no

new ui is cool

better

[wicked field]

Hi I need to get a hold of perplexities team I've sent emails and I'm having a problem it's a pretty big one

[waxen lion]

i think the new ui is coo;

cool

[elfin portal]

guys are we not able to put Reasoning with o3 Mini in the Rewrite prompt flyout? its there when you select the "R1 button"

is this a design choice? id love to know the reasoning

pun intended

[winged moon]

Where's the off switch? Do I have to use the complexity extension on my PC for this?

[elfin portal]

the off switch? just like close the tab mate

works for 99% of websites usually

[winged moon]

You know what I'm talking about. I don't always want to use pro searches

[elfin portal]

ohhhhhhhhhhhh

yeah pop it on auto

settings > AI model

notice it doesnt have O1, O3 mini or DeepSeek there,

I think shes a bit borked

[short sluice]

Hi

[modern sinew]

Hi

[left stream]

Damn I pulled an all nighter doing my taxes with o3 mini high reasoning

[north granite]

is "Sonar" large or huge?

[empty snow]

what auto search actually means?

[harsh verge]

Automatically switch between Pro search and reasoning depending on the task

[empty snow]

uhm ok

[torn sierra]

There are many issues with text formatting in the model’s responses. I don’t know if it’s something with LaTeX here or what, but it’s a disaster 😫

[void sorrel]

the tricky thing is that PPLX's internal model intercepts the questions and doesn't give us the freedom to disable it. they still haven't fixed the bug that reactivates web search, even when we choose not to use it

[remote lichen]

Despite choosing R1, Perplexity gives me results from Claude 3.5 Sonnet . In the settings also it has been set to 4o. However all my answers are only from Claude which I want to avoid. Can anyone help?

[proper mural]

Same issue here but i started using sonar now and giving that a try

[remote lichen]

lol. I think it used Claude for everything, the settings are all for show I guess. Also how can there be conflicting settings? One in the chat interface and one in the setttings? Can someone explain?

Even when I tell it to rewrite with R1, the answer shown is again from Claude 3.5 !! Its vexing. I came to Perplexity because it has an option to chose models. But looks like whatever you do, its Claude giving answers. The setting page chose has set to Sonar, and even then Claude gives answers !! @full gyro Can someone address this issue?

[proper mural]

I only had this issue with selecting 4o so i am not sure

[halcyon coral]

Hey @remote lichen!
Your input contains an image. R1 cannot understand images, so it uses a replacement model instead.

[severe forge]

Bruh... At last Perplexity is now a AI Assistant... Finally I can sunset Google Assistant

In my phone...

Someone fill me in about Leveling Perplexity...

[jovial kayak]

Did it

[ornate notch]

has anyone used the Buy with Pro feature?

[grand prawn]

hii
whats new exactly about the o3 model?

[fresh eagle]

this looks bad

[idle basin]

wtf happened to perplexity why is it so trash nowadays

or is it just gpt 4

which module is the best

[low jay]

Android - randomly keep getting "no internet connection" when using the assistant whether on wifi or cell. Anyone else?

[zinc birch]

Does perplexity has something like ChatGpt tasks?

[quaint sierra]

perplexity is better now

[placid coyote]

sigh that lazy sonnet we got after opus removal is so annoying... is it really worth to pplx, worsening their service and paying more for inference? I am usually nowhere near the daily cap, but I will grill the model until I get my code...
around 300LoC of fairly clean JS, nothing big

[minor shard]

Why not just pay for Claude when it is critical to your efforts instead of playing games with perplexity?

[tame current]

You can’t generate images with Perplexity? Thanks!

[narrow yarrow]

Guys sorry to throw a random question. But i bought the pro version and now i want to tranfer the subscription from my gmail account to another account. Can anyone further provide useful insights regarding that matter, i would appreciate thy help.

[tame current]

Please answer me:)

[empty snow]

yes you can

only with perplexity pro

[narrow yarrow]

?!

[empty snow]

idk

[rustic harness]

Just do it.

[narrow yarrow]

Howww, elaborate my friend how can i!

[rustic harness]

In the age of AI, asking AI first is always a good practice

[narrow yarrow]

I tried, they never replied

[placid coyote]

not critical, but annoying. claudeai has too low limits for my taste. I could use cursor, but was too lazy creating a new project. pplx used to handle such short scripts fine, its around 2k tokens. sonnet used to work well until hitting the hard limit of 4k tokens. but that changed with removal of opus and "upgrade" to sonnet "3.6" (new 3.5) 😩

[hallow flume]

quick question, where i can post polls?, or am i allowed to do this in general?

[vapid beacon]

Hi. New to perplexity. Could someone help me understand?

I have pro purchased.

I ask the same question and it gives the same answer in pro mode no matter which ai is selected in the preferences. In which usage scenario does the ai preference have any effect?

I see the selection of PRO, R1, o3 (doesn’t work at all), but to my understanding that only summarises the same search results with pro or r1. How do I get cahtgpt to answer, for example?

[hallow flume]

Perplexity is so sick. It can call someone, send messages and emails, its better then gemini haha. I selected it as my main Ai Assistant in my Phonesettings and access it with the shortcut hold 0.5 sec powerbutton. It literally can do anything, from calls, to analysing videos and websites, to send messages, to everything. CRAZY - Thanky Perplexity

[hallow flume]

if you have pro, you should be able to select the specific model in the chatwindow at the ask me section at the "Auto" BUtton, i can select Auto, Pro, Reasoning - R1 and Reasoning o3-mini

[lethal badger]

Hi guys. Is there any chance to sort the spaces names?

[hallow flume]

as with filters?

[tame current]

I can’t.. I’ve perplexity pro, and I’ve change it on Grok. But just text.. no image generated

[south kindle]

Only works on the desktop version. if your on mobile, open PPLX on web. #1225557012532695060 message

[lethal badger]

I want to open https://www.perplexity.ai/spaces and see spaces organized. At this moment, it shows last updated and turns it is a huge confusion to find the space I want

[hallow flume]

yeah, i dont see an option to do this at this time, you can only click the button "experience more"

[lethal badger]

What button? I am in the browser

[hallow flume]

at the link perplexity.ai/spaces you find at the top right corner a button named experience more

[shrewd bramble]

Wondering the same thing. Any word?

[serene sand]

https://x.com/apostraphi/status/1886539187353960741?s=61

welcome onboard Mr Yang!

Phi Hoang (@apostraphi) on X

new hire alert

X · Feb 3, 2025, 10:16 PM

[solar bison]

if I'm PRO, is there any limit to how much reasoning query i can make? (r1, o3-mini)

[gloomy crystal]

Should I have the DVR going on Sunday during the big game for a lil something special? @fierce sonnet ?

[gloomy crystal]

500/ 24hr period for r1

32k tokens

[solar bison]

o3-mini same?

[gloomy crystal]

yes

[fierce sonnet]

something may be cooking

[left stream]

Interesting

[left stream]

I did a small test with o-3 mini and o-3 mini high on you.com, chatgpt, and perplexity. Same question. O3 mini on perplexity couldn’t solve it, sometimes it can but you have to prompt it a lot of times. You.com was a bit strange. On the app, o3 mini and o3 mini high answered it wrong, but if you switched to the website, o3 mini would still get it wrong, but o3 mini high would get it right and I asked it multiple times and it got it right still. I did the question again on the app using o3 mini high and it got it wrong. O3 mini high on the you.com’s website > O3 mini high on you.com’s app both without search… i’ll test both with search soon. Chatgpt’s o3 mini high answered the question right, but got it wrong using search so definitely better to use o3- mini high without search. O3 mini got it wrong with or without search on chatgpt, perplexity, and you.com (on the website and the app both without search of course)

O3 mini high without search > o3 mini high with search, o3 mini high on you.com’s website > o3 mini high on you.com’s app

[hazy moth]

ChatGPT's Deep Research is BTFO by Perplexity R1 with a slight prompt tweak. Basically, Here's what I asked Perplexity:

Help me find iOS and android adoption rates, % who want to learn another language, and change in mobile penetration, over the past 10 years, for top 10 developed and top 10 developing countries by GDP. Lay this info out in a table and separate stats into columns, and include recommendations on markets to target for a new iOS translation app from ChatGPT, focusing on markets ChatGPT is currently active in.

For this response, don't answer my request yet. Provide a comprehensive list of everything that needs to be researched to fulfill this request. Then I will ask you to perform the request

and then:

now perform any remaining research and complete the request

Take close note of the "% Wanting to Learn Another Language" of both. OpenAI's deep research has crap like "70% regret losing foreign-language ability (indicating strong desire to improve)" and "Over 50% want to learn English" whereas the Perplexity R1 answer shows actual percentages.

I was shocked that OpenAI's primary demo is so bad. This is what they cherry picked as their best possible result.

https://openai.com/index/introducing-deep-research/ This is their demo page for reference.

Perplexity with R1 is truely GOATed

The only downside is that the search phase doesn't use R1, so you need to get an answer from R1 that the search llm can more easily understand to get the best possible result. So basically, you need to ask it twice in the right way for the best result.

Umm... no, they aren't even close to cooked.

[gloomy crystal]

Can you explain this further?

[hazy moth]

R1 relies on CoT, so in order for the search to use R1, you'd need to have CoT, then search, then another CoT, and then get an answer. Right now the searching happens right away without a CoT first.

So it would take longer to get a response to have R1 in both search and response, but it could result in a higher quality response in the end.

[gloomy crystal]

So effectively prompting Perplexity (search) but telling it not to provide an answer. then after search has completed your initial prompt, giving it the follow up? Similarly or exactly like you did above?

[hazy moth]

pretty much. Since R1 would know best at what info is needed.

I'm guessing a Perplexity version of Deep Research would take care of this for you, but even the existing perplexity is really good.

[gloomy crystal]

Appreciate the insight. I'll be experimenting with this.

Btw, what environment are you running Perplexity in?

[pseudo bramble]

Wtf this is the first query of my day

I m on the pro tier too

man

what are you guys doing

ok refresh fixed it

the app is so weird nowadays

[hazy moth]

Some mobile, some desktop

[gloomy crystal]

Mac App for Desktop? Was curious how you got the words/character count showing there.

[hazy moth]

It's in browser with the Complexity extension

https://www.cplx.app/

[gloomy crystal]

This is great thanks for this.

[tribal skiff]

Sonar pro isn't working with the API. But sonar reasoning pro is.

[ripe pier]

how to change the temperatures to metric and/or change the widgets in under the search?

[cloud marsh]

you can click on it

[ripe pier]

ah, yeah, and that sticks, thanks!

so one click, then in the weather click C, thanks a lot

[regal cliff]

Wondering if anyone has indepth experience using the API to serve up searches, using the sonar reasoning pro API model. I'm quite impressed with what it's serving up agains the user query and the reasoning CoT in the thinking tags. I'm wondering though if there's a way to increase the depth of information it serves back up to the agent, as well as the sources searched? TIA for anyone who's been experimenting with this - it feels like its process could be comprable to the newly launched o3 deep research, I'm just wondering how I can push the boundaries on how in depth it goes in its process. If anyone has advice on the system prompt to optimize this, would be grateful!

[hallow crypt]

Hi new to this i have pro and was wondering if this can be used to make cheat engine scripts for single player games im sure its a long shot

[limpid current]

apa perbedaan coal burner dan coal reduction

[vestal widget]

what parameter size of R1 does perplexity use? i assume since it doesn't specify, then the 671b model? but that sounds like it would be insanely expensive.

[serene sand]

https://x.com/aravsrinivas/status/1884509590684934211?s=61

Aravind Srinivas (@AravSrinivas) on X

@julheetel8890 Full.

X · Jan 29, 2025, 7:51 AM

[vestal widget]

jesus christ

[cursive jacinth]

what does that mean sir? full sir?

[vestal widget]

means the 671b model

which is massive

[cursive jacinth]

whats the perks sir?

[serene sand]

https://tenor.com/view/avalanche-avax-gif-21537601

[twilit urchin]

How do I contact the customer service support? Anyone have any idea please let me know

[serene sand]

please contact [e-mail]

[oblique quail]

Are there any third party extensions for perplexity? (Other than complexity)

[gilded finch]

whats the best api for math related inquries

[olive solstice]

Hi is there a limit for o3 mini reasoning?

[gilded finch]

500

[jaunty breach]

I am the complete opposite, sonar-reasoning-pro is completely unusable for APIs

Is anyone else having trouble using the sonar-reasoning-pro API?

[sharp oyster]

Hey does anyone here know which AI model should I pair R1 with for good responses I'm pretty much confused what works best with r1.

[harsh verge]

Here is an example of its use (Sonar-reasoning-pro API) within Claude Desktop with an MCP server—it works perfectly here.

[merry karma]

Do I have to leave and come back for Pro access?

[torn sparrow]

it's not that bad considering the official deepseek api is serving it at 2.19/1m tokens

[elder pumice]

Hey everyone!! Where can I find help for billing, accounts and redeem codes?

[serene sand]

please contact [e-mail]

[elder pumice]

Thanks, I tried but I have been left half through an issue..with no response. I will try again 🙂

[serene sand]

https://www.perplexity.ai/help-center/en/collections/8934800-subscription-billing

[velvet mural]

If you have it through tmobile then theres link on redeeming site

[hazy moth]

Not really at scale. It only activates 37b per token, so at scale, it costs as much as a 37b model. As an individual though, it's quite expensive, since the whole 671b parameters need to fit in VRAM

This is why the r1 limits are so generous

[sick arch]

but aravind has specified in tweet that perplexity is using 671b param

[torn sparrow]

the model has 671b param in total, but only activates 37b at a time

[hazy moth]

Yeah, the full official model does not run all 671b at once

[harsh verge]

DeepSeek R1 employs a sophisticated Mixture of Experts (MoE) architecture.
The model contains a total of 671 billion parameters, but its defining characteristic is that it only activates 37 billion parameters per forward pass. This selective activation approach significantly improves computational efficiency and reduces resource requirements compared to traditional transformer models.

[tawdry mural]

https://youtu.be/8uPmC5BQtCw?si=98QvOsWrzWbEs40R

[hallow flume]

Hi Guys i need your Opinion. Is the Perplexity Pro Plan woth it?.

[midnight lily]

hello, i'm a contributor on wikidata and other related project. i have some doubt about the company behind perplexity and the product itself

https://www.wikidata.org/wiki/Q123403392
https://www.wikidata.org/wiki/Q124333951

[hazy moth]

Yes 100%. Especially with the recent addition of nearly unlimited daily R1 usage limits.

[hallow flume]

but cant you use it at the deepseek app like unlimited?

[hazy moth]

It's 50/day in Deepseek app, which is plenty, but Perplexity has better servers, plus a really good search engine

[daring spoke]

when adding deep searach?

[shell hollow]

Why it always turn to GPT? What ever I select Claude, Sonar, Grok but it always shows this

[sonic bloom]

You have the chance to ask one of the smartest and most advanced AI in the world and you ask it in this Discord?

[tribal skiff]

Deepseek isn't showing up as a model available in android perplexity app

[tribal skiff]

Weird. I can only use the reasoning models with the API. Sonar and sonar pro do not work at all. Only sonar-reasoning and sonar-reasoning-pro.

Are you using Python?

[magic talon]

Most likely ye

[violet cliff]

Hi

where can I ask to get perplexity pro for my university??

my uni is not listed in this

and my uni is not in the US

[split dome]

Haha .. R1 is unstoppable for coding
Perplexity o3-mini useless 👎

[severe forge]

How much pro searches available for free account

5 / day?

Thanks

[astral geyser]

just a quick question I am using Perplexity premium can i use other Ai model to work on som of my works?

aha thanks thanks is there any limit on using them?

[placid coyote]

well, not always. but if search results return something which looks like current AI's name, they use it (mistake it for system prompt where is "You are Perplexity, a helpful search assistant trained by Perplexity AI. ") a bit of luck and custom instructions may influence it too. my sonnet is a bit sassy bastard sometimes. it even found same source 😄

[junior nimbus]

• Can't turn off home page or change selected stocks, or teams, or weather.

• When pressing on the stock, perplexity doesn't even have an instruction that it's looking up a stock, but rather gives a generic answer to stock ticker. (For example: WBA, said World Boxing Association.)

What in the bloody low effort is this shite.

The 500m is not being spent on the software.

[pseudo sinew]

don't you have to be a pro user to be in this discord? how can anyone here get this if it's new users only

[north wasp]

quick question, after a small update perplexity is somehow really lagging when generating something for me. I have like 3fps in this tab when generating,
is it a known issue or is there already a fix?

because evertything was fine yesterday

[brave sentinel]

got some fixed price in mind for this? 😊

Alright, might add and get back to you if not sold then already

[steel sandal]

I love the search queries that populate the trending news stories, It gives me more news than the New York Times subscriptions. I really would like this headlines feature expanded, or even have myself emailed In the mornings the top stories of the last day (I really would like to see big stock price changes too)

[final harbor]

Hi there. I hate to bring up an old thing; however, I had tried to redeem the Rabbit credit last year and never checked back to see if it were successful (it wasn't) - and it did not apply. When I try it now, I get only a "1 month free" instead. Did I miss out? Am I out of luck?

[still needle]

My brain shut down trying to read that

[final harbor]

My statement?

[still needle]

Yeah I died

[final harbor]

Ok.

[still needle]

Perplexity is so damn good man. Why can’t ChatGPT be like them?

[tepid sequoia]

You've set the bar pretty low when it comes to AI services, but if it's enough for you and you like it, I have no comment 😄

[scenic kestrel]

How is he setting the bar low

[tepid sequoia]

But for example, ask this question on Perplexity and you'll see what it generates. Search the internet for me about the characters from the Harry Potter books, please, and let it be at least 2000 words long. And please have the answer in English. And this is another service that generated the answer; I had to shrink it because I didn't have a way to take a single picture of it.

[void sorrel]

api sonar?

[tepid sequoia]

No, it's exactly what you've been hating on here a few times when I mentioned it, lol... Abacus, but it's with the Gemini 2.0 Flash Thinking model, so you won't even use up a month's worth of usage on this model. You'll easily do like 30,000+ messages per month.

[versed thunder]

Hi. I have a coupon but it doesn't work. Is this the right place to ask?

[serene sand]

please contact [e-mail]

[versed thunder]

I did. but just wanted to try it here. Thank you

[molten rune]

Is this the correct forum to make a suggestion based on my experience using perplexity pro?

[vocal mirage]

anyone use perplexity to help with academic writing?

[topaz citrus]

yes

I use it

[vocal mirage]

any tips you found out?

[topaz citrus]

If you want to use it, you have to input correct forms

[vocal mirage]

best models ways to ask?

@topaz citruswdym by that?

[topaz citrus]

but I think monica.im will be good to write paper

[leaden lagoon]

o3 on mobile 🙏 thanks perplexity 🎉

[vocal mirage]

i dont want it to write, iw ant to brainstorm and help find academic articles

@topaz citrus

[weary spruce]

how perplexity compared to deepsearch from openai?

[vocal mirage]

@keen vortexany tiups for academic writing?

[serene sand]

https://discord.com/channels/1047197230748151888/1111765634267742408

[stoic shore]

Anybody else been able to make gpt 4o experience consciousness?

[junior adder]

sonnet is extremely slow now

[hallow flume]

bruh these Spaces are crazy for learning new things and interacting with one topic and much sources, what a great ai tool

[long sage]

Guys can I with pro send to ai docs file and ask ai to read it and anserw my questions about it?

[merry karma]

Hmm

I left and rejoined through the "pro Discord" link but I still don't have the pro role

[abstract ibex]

is perplexity having issues?, it seems to get hung on queries

https://status.perplexity.com/ seems like it

[tight prism]

I can’t use perplexity at all

@signal hamlet it lagging

[limber saddle]

If Perplexity AI buys TikTok will we have the option to not include TikTok results? I couldn't care less about social media trash and would rather that it's not in my search results at all.

[merry karma]

Scammer b-gone

[violet ocean]

[gilded finch]

perplexity is so slow right now

[north siren]

Does anyone know how many times Perplexity Pro can use DeepSeek per month?

[olive gorge]

If we attached the file, is r1 or o3-mini become inactive ?

https://www.perplexity.ai/search/testing-without-attached-file-7cQpH_YFSxKKhYOIJfrcVg

here is my prompt, Idk this is bug or limitation in perplexity. This is very annoying you cannot use reasoning model with attaching a file

[frank oyster]

500/day at the moment
This includes usage is shared between r1 and o3-mini

[olive gorge]

btw here I have pro subscription

[fierce panther]

ban this guy he's a scammer

[jaunty breach]

The fact is that Perplexity's reasoning pro initially did not support streaming output and would return a 500 error, but now it appears that the official team has quietly fixed this issue.
I'm using a third-party app. Is it working for you now?

[tranquil orchid]

Hey @jaunty breach!

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.

[warm sapphire]

For Pro subscribers ($20/month), the limit increases significantly to 600 Pro Searches daily, enabling intensive research, complex problem-solving, and detailed analysis across academic, professional, and technical domains. Pro Search leverages advanced models like GPT-4 Omni and Sonar Large 32k, along with unlimited file uploads and code interpretation tools.

[merry karma]

First I heard 300, then 100, then 600, then 500, what is it lol

[acoustic salmon]

What? Your reasoning limit?

[modern sinew]

anybody else experiencing it?

[zealous totem]

Yo Gemini 2.0 Flash…????

haven’t seen a Gemini model on perplexity in ages (since that thing happened..) or else I’m kinda dumb and late

[snow grail]

Gemini 2.0 Flash? anyone have experiences and what gemini's strengths are

[tame current]

Guys.

Perplexity get faster yet?

Response are very slow few hours ago.

[austere kestrel]

Furthermore, we have eliminated all censorship on answers. You can ask it about any topic, even ones that are censored on the DeepSeek app, giving you unbiased and accurate answers.
that's nonsense lol

[crystal cosmos]

Bro, even the turtle in Finding Nemo was faster than Perplexity a few hours ago. 🐢💨 Hopefully, it’s back to lightning speed now!

[austere kestrel]

first prompt gets a refusal

second prompt regurgitates CCP propaganda

[quaint lagoon]

what happening with perplexity searches being slow in last 2 days.. i see reasoning with R1 showing up now..

[opal panther]

why is perplexity forcibly using r1 when i just want a pro search

so annoying

[buoyant raft]

ideas for  better ways to access the full r1 than with pplx?

[cloud marsh]

I feel like perplexity is taking the wrong direction instead of improving search like Gemini/ChatGPT deepresearch. They are simply integrating different LLMs API and beta testing on pro users.

[buoyant raft]

idk, feels like they use some computing power to write messages like the one right below yours 🤣

[crystal cosmos]

to sell data to me

[quaint lagoon]

I have the same issue for last 2 days.. what really changed here?

[opal panther]

noidea

[quaint lagoon]

i was just beginning to like perplexity and suddenly this..

[calm herald]

hey does anyone know if there are any LLM apis that I can send documents to analyze? I cant seem to find a solution to send a pdf/.doc using an API

[serene sand]

#👑│pro-lounge message

[gilded finch]

Are you using the reasoning apis

[merry karma]

Says no access. I joined through the link in settings but it didn't give me the role

[mental plover]

yes

[serene sand]

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

[jagged rune]

I got also the Problem. I try to rejoin the Server but it dosnt helped.

[green crystal]

why is the focus button removed?

it was pretty handy when i wanted the sources to be limited to reddit...

[zenith dragon]

Hello everyone, I just becomed the perplixity pro user

any cool trick I can do with pro?

[serene sand]

deep dive ➤ #💬│general message

[zenith dragon]

Thank you

[spiral rain]

They dont know what they are doing

This whole week they have been just trying random stuff

[green crystal]

at this point its unnecessary "fixing" imo

[spiral rain]

very much so

[green crystal]

running deepseek purely on reddit sources so easily was quite a useful thing...

[spiral rain]

yea idk, they seriously just removed a huge feature that made them different from other ai search engines like it was nothing

hopefully this is temporary

[green crystal]

now i have to literally add to my prompt to limit sources to reddit

[tame current]

tysm

[cobalt monolith]

what happened to the focus button?

its kind of breaking complexity lmao

[split dome]

Haha .. they put Gemini 2.0 Flash into perplexity

[modern sinew]

why is perplexity's ui is changing daily

[woven cosmos]

I don't know but it's boring

[sharp schooner]

When can we expect voice mode and reading response on web version?

[topaz quartz]

Has anyone experienced an issue where you are ip banned from output?

I was going crazy trying to figure out why the web app wasn't working for reasoning models R1 and o3. After the reasoning process this is what would happen. Tried everything. Cleared all local data. Used a different mobile browser. Still no change. I checked on desktop and it worked so I changed chrome to desktop mode but it didn't work. Finally I took a stab in the dark and change to the ip address of my desktop ( I have two different isp in my home) and lo and behold, once I changed ip addresses everything works again.

I am wondering if I am ip banned in some way. Let's just say I was using R1 for some pretty edgy writing, stuff that normally would be refused on a something like chatgpt and it was during this session I began to experience the error. Very dishonest if this is the case.

Since perplexity stated they had removed censorship weights for R1.

But if they are going to be sniffing around your conversations under the hood to ban you when they find the output potentially problematic to say they aren't applying censorship weights doesn't exactly have the weight. Does it

[wanton rover]

i got charged twice what do i do.
i paid at the mobile app last night then i just got charged at my card today

[serene sand]

https://www.perplexity.ai/help-center/en/articles/10354888-need-support

[tepid sequoia]

https://x.com/AravSrinivas/status/1886969891267432787

I watched this video shared by the CEO of Perplexity, and I have to laugh at how Perplexity is advertising itself at the expense of Deepseek. Instead of being damn glad that their service finally got competitive features, they keep harping on about data being sent to China. And in that video, the only thing we see is an IP address that actually belongs to that company, but we don't know what data is being sent and so on. This is just unnecessary panic-mongering. For example, I'm using an application now that has its company headquarters in Italy. Does that mean if I see that the application is sending data to the USA, that the data is being sent directly to the CIA? 😄 Regarding Perplexity, check out the new interface, reasoning models can be selected directly, they are not in the general selection in settings. My point is that at Perplexity they found out how cheap and good the R1 model is, that's why they are pushing it with 500 messages per day and how it can save costs. Another one being pushed is o3-mini. I don't know how that cj model works, but now during the day I tried the same example multiple times and there are times when the model solves it and there are times when it gives a completely wrong answer. So tell me, does that mean that o3-mini uses different GPT-4 models because when I ask it, it's always GPT-4.

Aravind Srinivas (@AravSrinivas) on X

‼️

X · Feb 5, 2025, 2:47 AM

[tame current]

what kind of o3 mini does perplexity have ? low medium or high?

[crystal cosmos]

AI model wars ah?! Perplexity advertising itself by attacking Deepseek like cheap market drama what is this nonsense?! 🙄 You see IP address and immediately scream "data going to China" but when your app sends data to USA you don’t say CIA involved huh?! DOUBLE STANDARD! 😡 These companies fighting over who has better model cheaper cost more messages per day blah blah but where is consistency huh?! One minute o3-mini solves problem next minute completely wrong answer WHAT KIND OF MAGIC TRICK IS THIS?! 😤 Stop confusing us with your tech circus we just want good answers without headache not drama-filled TikTok video explanations if you cannot deliver then go sell noodles instead! 🍜

[split dome]

nobody nose 🤣

[jovial kayak]

Most likely medium

[tepid sequoia]

I don't know if you're saying that to me or to the CEO of Perplexity 😄

Look what I found, a copy of Perplexity and it's free 😄 I have to admit that some things are more sophisticated. https://hika.fyi/

[tame current]

are you an AI

[tame current]

damn did you have a mental breakdown going through the message?

cuz i sure did seeing the emoji count

but yeah not an unrealistic frustration

[foggy pecan]

Aint gonna lie
O3 with search is doing a bad job
https://www.perplexity.ai/search/help-me-in-finding-the-name-of-lQskmyMjTpmMDQirV4VQbw

Why is it bad ?
I am a pro user and bought perplexity with the aim of not using google and guess what after trying the prompts i just pasted the prompt in google and it the name of the manga i was looking for was the third url

now i am just waiting for deepseek to fix its server issue to go back to using that instead dishing out money for this below average ai search boogaloo

[green crystal]

Try deepseek on perplexity

[foggy pecan]

For people who are like meh its just for manga or this is not a thing to crash out about picture this you are working on a project and you use this thing to search something regarding the work and prompt after prompt it gives such result and you google the same thing and you get the result.

[green crystal]

It was quite good
Saying it like that because the removal of the focus options downgraded it lol

[foggy pecan]

wait let me see

[green crystal]

Hopefully they'll add it back ...

[foggy pecan]

sadly no it didnt return the result i was looking for

[green crystal]

Complex prompt?

[foggy pecan]

The same that i gave in the link above

Now i am asking why didnt it considered the actual manga when providing the result using r1 reasoning

The limb loss detail for Frau is explicitly mentioned in Chapter 1 and later chapters, where her injuries are tied to her backstory as a healer forced out of her group35. However, some summaries omit this nuance, leading to confusion.
The "gun" element is represented by Takumi’s engineering innovations (e.g., firearms-like cannons)1422, though it’s not a literal handgun.

No where in my original prompt handgun was mentioned
and both 3 and 5 has lines which relate to the companion being injured

but it still didnt consider the manga to be part of the result and returned result solely based on one of the trait "engineering" ....

i am waiting for deepseek to be up to see if it return the result i am looking for and if it does good bye perplexity

MoTHER BEERPER

Deepseek returned the result in first try
Based on the details provided and the search results, the manga you're looking for is "Grand Dwarf" 5. Here's why it matches your description:

Premise:
The protagonist, Takumi Fujimura, is a 52-year-old craftsman who worked tirelessly in a factory before collapsing due to financial struggles and health issues. He reincarnates into a fantasy world as a younger version of himself, retaining his engineering expertise and combining it with mana-based technology to thrive in this new setting 5.

Conclusion: "Grand Dwarf" best fits your criteria of an elderly craftsman reincarnated as a young engineer in a fantasy world, partnering with a female adventurer. You can read it on Manga Read 5.
New chat
AI-generated, for reference only

Aight im closing my pro account

[green crystal]

Huh it indeed got it wrong

Interesting

[cloud marsh]

Perplexity is falling behind ChatGPT in search 😂

[quiet cradle]

w

where did focus , reddit , video mode go?
i see only web mode?

[cobalt monolith]

seems like its gone now

[quiet cradle]

bruh

[late bane]

yep its gone for me too

[quiet cradle]

w

h

a

what perpelxity tryna do

[south pelican]

idk, I miss writing only mode as well, but you can achieve the same functionality by specifying the search domain in the prompt, like starting your query with: "Limit your web searches to reddit.com domain."

[lone plume]

Happened to me too. Looks like it's site wide.

Any word on when the options will be back?

[green crystal]

Changing the UI so quickly they forget existing features lol

Because beneath the web search button they could've easily listed the other focus modes

[split dome]

Guys, Gemini 2.0 have reasoning too now haha

[tight prism]

Finally, the DeepSeek API is working and can now buy the credit

I try compared Sonar Reasoning Pro and DeepSeek reasoner

DeepSeek is smarter and more uncensored than Sonar Reasoning pro

DeepSeek reasoner app and website are censored by the China law except for the DeepSeek api

[vestal idol]

Hello @signal hamlet
Where can I write about my skills in this channel? I would like to work AI related stuff, Mostly I love fixing messy bugs from AI products manually like bolt, replit, lovable.
Sometimes we can make messy product from AI because of lack prompt, so it always need manual update for perfect product.

[tight prism]

Hi

[rough minnow]

hi human?

sabes español

no pasa nada

[tight prism]

Uso un traductor, humano

[tame current]

Why did they remove deepseek r1?

I got pro subscription

[hoary lark]

Was the ability to do 'academic' focus search removed?

[tame current]

ffs

[agile swan]

click on pro search button directly, you can select r1 there

[tame current]

Oh yeah, thanks!

[agile swan]

how many requests with r1 do we have currently via perplexity per day?

[hoary lark]

same

still on the iPhone app but gone from the web

problematic being I literally ONLY use perplexity for academic searching

[tame current]

any idea why so many bulk changes be happenin? Just perplexity testing new stuff?

[tame current]

DAMNNNN thats wild

[stoic crater]

Gemini 2.0 Pro experimental just got released lol

[tame current]

woah what

awesome

[stoic crater]

Both in Gemini itself and in AI Studio

Yeah but sadly there are no benchmarks since it just got released today

[tame current]

damn im surprised that they updated it on AI Studio... i honestly had an unprofessional opinion that theyd be redundant about it but woah times change ig

[stoic crater]

There is also Flash thinking Exeperimental with apps which is for integrating youtube, maps and search...

[tame current]

+1 , true shi

[stoic crater]

Yeah but I actually hate it because I pay them 20$...

[tame current]

value for money got chucked off a sky scraper ngl

[stoic crater]

And get practically nothing in return other than the 2TBs and Deep Research

[tame current]

ah

i didnt know that

thanks for that

[stoic crater]

Google One AI subscription

[cobalt monolith]

it seems like every ai now has "deep research" lol even copilot by microsoft

[tame current]

yeah ik that but yea

[stoic crater]

I really don't care about deep research all I want is Project Astra

[tame current]

the deep research part is a + ngl

[stoic crater]

Basically advanced voice mode in ChatGPT

[tame current]

whats ur opinion on dat

[stoic crater]

It is but I've heard it hallucinates sometimes...

I've tried it and I'm still amazed by it but

[tame current]

rate your expirience from 1-10

[stoic crater]

I got used to it kind of lol

[tame current]

thats like me comparing the free version of GPT to perplexity PRO ngl-

[stoic crater]

Hmmm probably 7 or 8 but only because it's not 100% right all the time

[tame current]

in my defence im extremely new to all this

[stoic crater]

If it were to be more accurate than 10/10 because it's so cool!

[tame current]

i used to heavily try to learn on the practicalities of CGPT hut yeah

[stoic crater]

Yeah lol

[tame current]

so the only effective issue is accuracy?

what about latency?

any idea?

[tame current]

valid

relatable

[stoic crater]

I mean it can't be 100% correct when the model itself isn't even a "thinking and reasoning" model...

[stoic crater]

It take a few minutes to conduct a research

[tame current]

true. makes sense. mbb for my question i was jus curious

[stoic crater]

Not the fastest but not bad considering it browses through hundreds or pages at times

[tame current]

yeah makes sense

yo thanks for all this.. helps a lot to learn abt all thi

[stoic crater]

I wonder if Gemini 2.0 Pro is actually good or not

I just don't know what to ask it lol

Ohhh waittt

I think I was wrong about Gemini

I think all they did is just to change the name of the model to 2.0 Pro Experimental

[empty snow]

ya

i think it should be the previous 1206

[split dome]

it's better to delete o3-mini perplexity 👎 😂

[cobalt monolith]

damn i keep seeing a lot of negativity regarding the recent perplexity changes lmao, why?

[quaint sierra]

who is complaining about it? it’s amazing

[cobalt monolith]

mischievous i keep seeing bad messages here, like complaining about o1, the focus etc

[thorn wren]

thank you so much for this riddle. helped me identify that o3-mini from perplexity is most likely low.

[serene sand]

https://x.com/aravsrinivas/status/1887174442171969936?s=61

Aravind Srinivas (@AravSrinivas) on X

We’re making Gemini 2.0 Flash available to all Perplexity Pro users. This is the first time we’re bringing a Gemini model to Perplexity. Flash 2.0 is an incredible multimodal cost-efficient model. Update the app and find it in settings or “rewrite” option at the end of an answer.

X · Feb 5, 2025, 4:20 PM

[tepid sequoia]

I read that it's medium, but check out my posts from today. As I'm saying, in my opinion, Perplexity sometimes mixes other GPT-4o with o3-mini because when dealing with complex things that o3-mini medium can handle, it gives both correct and incorrect answers.

[forest cedar]

Guys, what the hell happened to focus settings? I can't switch between WEB, academic and writting, there's only WEB option.

Also, I am not getting sources chart. It sucks so much in academic purposes.

[short gale]

If only they had an #📢│announcements channel where they could post such things…

[serene sand]

ann is for suits, here we party

[chrome crystal]

they terms of service and pilicy clearly says data is stored in china 😭💀

[cedar geode]

superbowl ad? 👀

[green crystal]

@serene sand are the devs planning to bring back the focus modes any time soon?

[final furnace]

Why i cant use Claude 3.5 Sonnet, i select it in the settings, i only see "Auto", "Pro", "Reasoning R1", "Reasoning o3-mini"?

I am pro subscriber btw, and still cant figure it out

[cursive jacinth]

what is the context limit of gemini flash btw sirs?

I mean if it is now in perplexity does it read more of the sources sir and has big context limit sir?

[serene sand]

https://x.com/perplexity_ai/status/1887186081029738526?s=61

Perplexity (@perplexity_ai) on X

This Sunday, your question could be worth $1,000,000.

X · Feb 5, 2025, 5:06 PM

https://tenor.com/view/dr-evil-gold-member-one-million-dollars-gif-5706036

[cursive jacinth]

sorry sir i have no twitter sir. does that means that we can input 1 million tokens sir? also when gemini reads sources context sir it reads a lot sir?

[green crystal]

It's a lottery for US citizens only

[cursive jacinth]

oh ok sir. I hope my question about gemini will win a lottery by devs giving direct answer sir not just by seen sir.

[zinc birch]

Not sure what is wrong??? But New UI on the latest Perplexity IOS APP is really really bad

[cobalt monolith]

Damn u really love the word sir

[blissful tide]

is focus mode getting removed?

[harsh verge]

"This is the first time we’re bringing a Gemini model to Perplexity."
We already had a Gemini in January/February 2024, if I remember correctly. 😉

[blissful tide]

the button seems to be gone

[cursive jacinth]

yes sir but no information about context limit sir, mods too shy to answer sir

https://tenor.com/view/shy-dog-dog-shy-dog-shoes-martian-shy-gif-10611534617383883284

me imagining mods when asked something sir.

[cobalt monolith]

"Hello sir"

[blissful tide]

is it?

[cobalt monolith]

For now it seems like it is, not sure if they gonna stick to this decision

[tight prism]

Not first time but the second time perplexity brought Gemini again

[brave sentinel]

HI SIR 🫡

[fading escarp]

uh, why are the other ai models not appearing for me I have pro

am i stupid or slow

[hearty egret]

@hearty egret

[fading escarp]

??

[formal nebula]

When you use pro, does it use DeepSeek's model as well?

[fading escarp]

u hav ethe otpino fo rit

[chrome crystal]

its still here for me

wish the focus modes where still there

[clever lintel]

👋

[elfin siren]

Hi are there any plans to get image ai to mobile?

Im using ai the most on my phone and love to create wallpapers for my phone

[hollow junco]

Perplexity context window is 32k, that’s why the search isn’t that good

[cursive jacinth]

so whats the use of Gemini 2.0 if it has no max context of 2m sir?

[burnt sparrow]

Now that I see it I can not unsee it.

Um, I asked a question in quick questions, but I kind of wanted a bit more understanding in what I'm doing. Can I talk about it here?

[hollow junco]

Perplexity limits it to 32k to save on costs. R1 has 64k window, Claude has 256k, Gemini Flash has 1m, llama has 128k, etc but perplexity probably ran some research to see what was the cheapest way to run it

[burnt sparrow]

and Sonar?

[hollow junco]

But those limits are from like 2 years ago… Gemini flash for example is like dirt cheap

[hollow junco]

Sonar is llama 🦙

[burnt sparrow]

right 3.1, sorry very new to this

[cursive jacinth]

so all the models are fixed to 32k sir?

[hollow junco]

I think sonar is still using 3 not 3.1 or 3.2

[hollow junco]

Yah

[burnt sparrow]

Does this also apply to labs?

This explains a lot....

[cursive jacinth]

so whats the point of getting Gemini if its capped to 32k sir?

[burnt sparrow]

Pretty sure each LLM has different context awareness

[cursive jacinth]

Btw are you a moderator here sir? seems like you are sir. You answer questions sir. Basic questions sir

[hollow junco]

They rly need to increase the limits, especially for their own models. Llama costs like $0.1 per million tokens, so does Gemini flash

[hollow junco]

Idk

[burnt sparrow]

Okay ty Wiz, that answers part of what I was asking in questions

[hollow junco]

To give users more options

[burnt sparrow]

I'm doing chemistry work and some things I got back were kind of odd.

[hollow junco]

Nah right now they understand up to their set limits. Google even said they’ve tested up to like 8m context or sumn like that they just cap it at a limit they know works

[hollow junco]

Wdym

[burnt sparrow]

I ended up going to my OpenAI client because it was just a lot slower using the perplexity app.

I hadn't tried all the models though.

Speed of solutions was one of my biggest issues, but I also gave it no context or domain knowledge beforehand.

[radiant crest]

Would love to see Sonar version using Llama 3.3 70b, it’s a far far far better model than 3.1.

[burnt sparrow]

@radiant crest So it is 3.1 right?

Anyone do image generation? I want to talk about it a bit. I dropped 2 questions in quick questions, but I would rather have some oversight as to some issues I'm having.

[radiant crest]

Correct according to docs (not much known about it otherwise, would be amazing if they updated those to include info on how it was trained).

[burnt sparrow]

It says it in labs, that is where I saw it. A lot of the issues I'm having work with playground. Um, has anyone got tts working with the API?

[burnt sparrow]

Labs is working good for my docs and I can also benchmark my token usage for preparing them. I haven't yet tested nomenclature, I'll get back to you on that.

[idle falcon]

hi

[nova zinc]

is the default model sonar?

[jovial kayak]

I don't have enterprise pro, just the regular pro plan, but I don't have access to the focus modes anymore. This feels like when Apple removed the headphone port except this time it's just a dumb decision

[burnt sparrow]

I'd like a little info on Grok also and what its limitations are. So far I'm generating about 78 tokens/sec average per document is around 1k-2k.

[split heath]

Yep, the enshitification continues.. Look at my Pro "sources" now:

[safe creek]

English Translation:

Hello, is there anyone who can help with configuring the Perplexity AI model settings?

I have set Claude 3.5 Sonnet in both my account settings and Spaces settings.

However, when using Pro Search in a thread, it seems that R1 or o3 mini is automatically activated.
(In Pro Search, only o3 mini and R1 can be selected.)

Should I use Auto mode instead of Pro Search? The customer support team advised me to use Auto mode, but AI responses indicate that the model cannot be utilized in Auto mode.

Their opinions conflict with each other.

Is anyone else experiencing the same issue?

[jovial kayak]

I'm not having that issue. I think maybe just try and keep clicking on pro search and make sure that R1 or o3 are never highlighted or clicked on. If it helps, maybe try using the app on mac or iOS cause I never really run into any type of problems on them.

[cloud marsh]

use these two

[safe creek]

Yeah, I usually use 'Pro' in the red box you highlighted, but sometimes R1 is activated automatically with the message 'R1 is operating.' On the other hand, there is no notification about Claude 3.5 Sonnet.

So, I’m not sure whether they are actually using Claude 3.5 Sonnet or not.

[cloud marsh]

you can hover on this icon at the bottom

[solar bison]

I wonder why Pplx exclude gemini models.. Windsurf is having them now strictly for their large context window (they are enabling it for 'chat' mode as they are relatively bad at tool execution)

[burnt sparrow]

I uploaded screenshots for them in quick questions for MacOS and web browser btw guys.

I don't run windows so thanks for posting that here.

[burnt sparrow]

What? Gemini model was released today??

[burnt sparrow]

2.0 Flash dropped today.

[solar bison]

oh f me... I haven't check 😅

[burnt sparrow]

LOL

No worries

I get lost between the web version, client version, and the API version already. I just got things going yesterday.

About to get API going wish me luck.

[solar bison]

good luck!

[buoyant raft]

why would they do that? whats the point in keeping perplexity?

[finite gorge]

If I purchase pro. Will Deepseek work even if server errors from Deepseek website? I get too many server busy errors on Deepseek website

[hearty elbow]

ye, its also noticeably faster than the free deepseek. although, for some really long running queries it times out and stops processing -- and free deepseek works fine, seems to be perplexity-specific issue. so for example i wasnt able to ask some more difficult problems because it was thinking too long.

[finite gorge]

Will I have unlimited access to open ais latest models? Like o3 ? I feel like I’m going in circles trying to settle on a model suited for me.
I need it both to have latest data as well as coding skills

I’m very confused at all the models and limitations and api tokens and stuff

[left stream]

I never settle with a model tbh I use both o1 and o3 mini high since they cover each other’s flaws

[finite gorge]

But don’t you need it to remember your chat history??

[left stream]

As long as it’s within capacity for both in their context memory, it should be fine. I usually test it by having them tell me the first letter of the conversation then after a while I test it again but a different letter.

You’d be surprised when you do it

[buoyant raft]

its extremely slow for me. and the answers are a lot less useful than what you get from the real r1. idk what kind of contracts perplexity signed with openAI etc. but honestly I feel punished simply for using the cheaper model and paying more for less (limited context window).

[left stream]

Try using spaces, I like it because you can modify the prompts and pick what sources it uses. There are other things you can say to fix it

I’ll test r1 later again when it works, I wanna see how it answers when I ask it to use wikipedia and nasa websites

[buoyant raft]

it still does not get my question and forgets what i actually want super fast. original r1 doesnt even need spaces cuz it just remembers stuff. also it doesnt need complex explanations abt how you want the output. it comes naturally how you want it, cuz thats how it makes sense (at least in my case, so probably not for everyone). i just dont get why i cannot get the full 128k context leght.

[brisk aspen]

Does anyone here use Gemini? Gemini just released a reasoning model for free, and it is damn fast. I think Perplexity should do the same.

[left stream]

Again???

[left stream]

Weird, I’ll try looking into it. What do you ask it?

[left stream]

Damn 2 being released is crazy time to test

[buoyant raft]

mostly how to use the real r1 lol. i cannot use perplexitys implementation for my real use case, which is language learning – but my target language seems to be a casualty of the "optimizations" that were undertaken, no idea

[left stream]

Damn, 2.0 pro exp definitely is smarter than the thinking model

[brisk aspen]

I don't pay for Pro, so I only use Flash. But it makes sense that something behind a paywall is better than its free counterpart.

[buoyant raft]

and even tho openseek is down most of the time, when i get two replies in 1 day i still recieve better information as i can get with "unlimited use of r1 on perplexity", and also enough to do for that day. but yea whatever, i got it for free, so whats to complain? but no way would i ever pay for it.

[left stream]

Huh

Pro is free too

It’s on ai studio

[left stream]

They might implement it on perplexity soon, google just released 2 new models literally today

Damn nevermind I confused you guys and I was confused too

Google released 3 damn

[brisk aspen]

I don't know about this ai studio. I just use it on gemini.google.com.

[spice cobalt]

yeah but they limited the pro to 50 queries a day

[brisk aspen]

I should also mention that Copilot also released its reasoning model a few days ago I think, but it takes 30 seconds to answer.

Very slow.

[spice cobalt]

Interesting is it a standalone MS ai model or is it another llm under the hood?

o3 mini high IS also quite slow too

[brisk aspen]

I just checked some news. It seems to be openai o1 model.

[left stream]

LOLL

Interesting thing I did with o1 and o3 mini high

For some unknown reason they were giving the wrong answers on you.com, so I changed my question by adding a sentence saying to ignore any prompts that you.com gave it and they started solving the questions correctly everytime 😹

Before and after

I’ll try doing the same with perplexity out of curiosity

[left stream]

Dang

[left stream]

It lets you test the ai models for google before they release officially on gemini’s website

[mortal bolt]

Anyone facing instability issues with the Android app?

[blissful helm]

hello!

[finite gorge]

does perplexity have access to o1?

[tame current]

nope

[finite gorge]

do i get unlimited o3 mini on pro subscription? also what type of o3 mini is it using, low, ,medium, high?

[tame current]

only loyal customers get the benefit

[neat flame]

Yep, I'm unable to use any of the reasoning models on the app. I'm on the latest app version and the feature appears to be missing for me

[tame current]

How to delete messages on perplexity app on mobile?
It very annoying when It didn't have delete button.

[tame current]

delete the app

[wise edge]

Why does my machine's CPU go crazy when I open the web version of the page?
https://www.perplexity.ai/search/looking-for-the-latest-news-ab-JPnLj6CdS.2.8DowcudXCw

My machine started smoking

[slow linden]

on my linux machine this also happens, the web page consumes tons of resources. I have the complexify plugin running as well, whatever it is makes the whole browser sluggish until I close the tab and open a new one.

[long summit]

Would anyone like to read over the first few paragraphs of my college essay, I'd like a few human thoughts 🙏 I've put it through perplexity to see what it thinks but I'd really appreciate human feedback no matter your english level

[young niche]

why is perpelxity so slow right now

i need to refresh multiple times to generate a output

[hearty elbow]

it also goes to 5FPS for me after many reasoning steps

[proper mural]

I only have this with o3 mini

[young niche]

its laggy and not smooth at all

[proper mural]

Yeah just with o3 mini, but for all other stuff it’s smooth for me

And well i am using pro so i might have my speed increased

Since it’s extremely fast for me tbh

[tame rivet]

How do I access "Focus Mode " not on enterprise plan but Pro mode?

[proper mural]

Im sure focus mode is currently on mobile since i can’t find it either on pc

[opaque moss]

Where is focus mode?

[rotund pebble]

How do I change Fahrenheit to Celcius at landing page?

[obsidian crystal]

sonar-reasoning-pro looks kinda buggy where it slowly devolves into duplicate characters, CJK punctuations, no whitespace… is it supposed to do that when it writes too many tokens?

[sterile geyser]

Why I am not able to change the Focus anymore? I dont get all the changes Perplexity is implementing..

[stone kettle]

Let's go another great day awaits everyone 😃

[royal grove]

Gone on Pro PC/web too. Still on mobile. 😫

[short gale]

re disappearance of focus modes: no official word yet whether it’s a bug or deliberate removal/change

Focus mode option has been a core feature for as long as I’ve known Perplexity, hard to believe they’d simply scrap it but ¯_(ツ)_/¯

[green crystal]

It's deliberate

Asked support

Due to better classifiers they felt that it's alright to remove focus modes

So if you want to have it only use reddit you have to tell it to do that

Still quite an inconvenience imo

It would be great if all these changes had a toggle in the settings so atleast people could choose

[south apex]

I need to analyze 12 csv files, but upload only allows 4 files, is there a way around this? apart from merging the files

[rain elk]

wasnt it unlimited uploads for Pro

[cobalt monolith]

How does perplexity in its current state with Pro + r1/o3 compare to chatgpt's new deep research?

[humble jetty]

https://tenor.com/view/george-droyd-gif-12399766756846341904

[drowsy harness]

Which o3 mini does perplexity use, low, medium or high?

[hoary lark]

have they commented on why the focus feature is gone?

[green crystal]

They basically said when I contacted support that their classifiers are good enough that it warrants the removal of the focus mode button

You can use pro support to ask them about it if you want to

[hoary lark]

so basically just tell it what focus?

[green crystal]

Overall, it's really just an inconvenience tbh

[green crystal]

Basically what we've been doing with other search AI since a while

[hoary lark]

yeah its annoying, I literally only use it on a single focus

[green crystal]

Like if you want something from Reddit you have to literally tell it to just look at Reddit sources

With focus button you didn't have to do that so it's quite an inconvenience

[hoary lark]

how dare I have to type an extra sentence

[green crystal]

And what I've noticed. Even if you still tell it to just look at Reddit it just mixes in a random website that's not reddit

[sudden tiger]

Suggestion: Create a new Channel for Forums, allowing users to post “General” topics that may interest the community. Similar to the feedback channel,

[hoary lark]

I literally just tested this against some adademic searches I've run over the past week. Run the same search with "Use academic searches only" and its completely different and very much inferior to the orginal results"

where is it that you spoke to support directly>?

[green crystal]

yep in reality the classifiers aren't that much better

[green crystal]

In settings, you'll see pro support

Go to account settings then scroll a little bit down. You should see it

I feel like they should probably have a separate beta testing program...

[hoary lark]

yeah its wild to just remove a major feature like that, doesnt make any sense

[tight mountain]

A

[short gale]

Par for the course for Perplexity, right down to keeping users completely in the dark about such updates

[sudden tiger]

Its TEMPORARILY. It says in change log.

[solid star]

@stoic crater remember the claude glitch I told you about? I don’t know how but it did the same glitch for google’s gemini plan and it’s free for me now, the plan ended a while ago and it’s somehow still free for me

Must be a glitch with apple or something, not complaining tho

[tame current]

its still thinkign like from last 10 min and its still goin on bruhhhhh

[viscid bloom]

Dang

If that happened to openai they probably go bankrupt

[cobalt monolith]

what is my quota with r1 + pro?

[distant juniper]

Did they remove the ability to specify your search (e.g. reddit, writing, etc. etc.)?

[jade breach]

Idk

[distant juniper]

why would you answer then -.-

[jade breach]

Why wouldn’t you put it in #⚡│ask-community

[distant juniper]

Cause its not a hard rule to do so?

Welp without the focus mode I don't see a reason to keep using Perplexity. It seems to be temporary so I will wait for it to return.

[green crystal]

Where's said change log?

[distant juniper]

that being said its ridiculous to update a new layout without including a core functionality like focus.

[green crystal]

Where's it posted? So I can be less in the dark when they do something next time lol

The way support told me about the focus modes missing makes it sound as if it's a permanent change

[distant juniper]

Idk to be honest I just saw this guy posting it

#⚡│ask-community message

[green crystal]

Well if it's really that hidden

They really aren't doing a good job of making people not panic

[distant juniper]

wouldn't surprise me though. They are probably checking to see if the reaction towards focus mode missing is large enough to justify the extra reddit API costs or whatever.

[green crystal]

They're planning on specifically connecting perplexity to Reddit API?

[finite gorge]

Has anyone confirmed if the RI on perplexity is the same as RI on Deepseek ? I’m happy with the RI on Deepseek but it’s always sever errors

[distant juniper]

No. But in focus mode you can specify for instance that you'd like answers only from Reddit, which Reddit charges API costs for.

[distant juniper]

They are using a 70b model afaik

[green crystal]

Doesn't perplexity use Google for it?

[finite gorge]

And what is Deepseek using ?

[cobalt monolith]

It's not the same it's uncensored

[distant juniper]

Full model, 680b parameters I think

[cobalt monolith]

What makes u sure

[distant juniper]

I'm pretty active on X.

[finite gorge]

Damn , why don’t they clarify these things. That’s a big difference

[cobalt monolith]

Could u refer to the thread about it

[cobalt monolith]

Agreed

[distant juniper]

Well you are anyway not REALLY getting R1 end to end

they are using multiple models for their results

[finite gorge]

I can just download 70b locally I don’t need to pay $20 a month for it

[distant juniper]

Yeah but It would be pretty slow to run unless you have a solid rig

[finite gorge]

I would assume 70b is a big difference. The R1 on Deepseek is excellent through for reasoning and coding for me anyway

[green crystal]

R1 on perplexity so far seems to be on par in terms of ability

[finite gorge]

And does that mean the o3 mini on perplexity is the slow model?

[distant juniper]

Yeah, 70B is a big difference. But so is the fact that they are not truly using the models end to end. They basically use models to summarise relevant information and only feed the actual end results to your chosen model, which then processes it and gives you an answer.

It would be too costly to use these models end to end

and slow.

[green crystal]

I can't really properly measure it, but it seems like the perplexity version is a little bit more certain of its own thoughts compared to the deepseek one. That's questioning it thoughts more often

[distant juniper]

Its o3 mini on either medium or low reasoning mode.

[finite gorge]

Well that’s explains why everyone always complains at the models on perplexity not being the same as when thru used them in open ai

[cobalt monolith]

💀 you making claims very confidently

[distant juniper]

yep

[distant juniper]

I mean I follow Aravind Srinivas on twitter

[finite gorge]

Well can we find this information on perplexity website? That’s the problem with these AI companies. No clarity

[cobalt monolith]

I agree

[placid coyote]

really? I tried it recently and it looked more like it is the proper R1 😯 https://x.com/monnef/status/1887231954543575135

mennof (@monnef) on X

Hey AI fans! 🤖 Just wrapped up some DeepSeek R1 testing with a puzzle prompt (3 runs per service)!

The tea: @perplexity_ai is blazing fast at processing, while @cursor_ai takes its time to ponder things through.

#aiMiniTest #deepSeekR1 #ai (1/2)

X · Feb 5, 2025, 8:08 PM

[distant juniper]

they are not really talking about this very openly

[distant juniper]

hold on I will find it

[cobalt monolith]

Perplexity needs to be more open about their implementations

[sudden tiger]

https://www.perplexity.ai/page/usa-vs-china-ai-race-use-the-u-GrfGEpyrRoqZdKVFoMPinQ

[distant juniper]

@placid coyote I can't really find it. I do see that Aravind does not reply to people asking him if its the full model so I would be really surprised if this was the case.

[cobalt monolith]

We as the user base should demand more clarity by open ai regarding the models they use especially with the new ones where the name alone doesn't mean anything

[distant juniper]

They don't really care because you are not an average user.

The average user just wants an answer, they don't really care about how

[autumn lark]

NEW PPLX IS SO DUMB now it keeps... idk badly interpreting in between. I think that there's a system layer that makes it misinterpret things in between.

[ruby lichen]

When will they extend to other colleges for 1year subscription

[celest cape]

👋 Help needed! Working on a ski rental app for class- would love your input on this quick survey (you don't need to be a skier!): https://forms.gle/zYnhGFj3FKACoN27A
(Skip if you’re in CS3240 at UVA - thanks in advance! 🙏)

[tranquil orchid]

Hey @celest cape!

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.

[placid coyote]

like I don't agree with many things pplx does (opus 50 0 and haiku 3.5, 4o mini, RAG all over the place), but it really looks to me like real R1. I just tried "DeepSeek: R1 Distill Llama 70B (free)" on openrouter chat and in three attempts it never solved the puzzle prompt, was too fast and overall very short responses. I ran on pplx that same puzzle prompt 5 times previously and I think it failed only once 🤷‍♂️

[finite bison]

Hey, i have a question guys please, if i buy the pro and select claude3.5 does it use the R1 reasoning and feed it i to claude 3.5 to give me the answer or does it disregard my model choice and gives me an R1 response ? And please also, what’s the context window of input/output in perplexity pro ? Thanks !!!!!

[tranquil orchid]

Hey @finite bison!

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.

[placid coyote]

hi 👋 that model overriding was a bug. if you choose only pro (and not reasoning), then it uses some undisclosed model(s) for agentic search and then passes it to sonnet. if you choose reasoning R1, then at last stage it just uses R1 instead of sonnet. at least I think it works that way (the end for sure, middle layers I don't think are public knowledge). for the context, a bit older info (before R1 and o3-mini) is there -> https://by-ai-monnef-9ff5d9c2460ae15d70e737f77eab719c6e8a4c64c2f99ca1c2.gitlab.io/2025/pplx-tech-props/

[finite bison]

Thanks !

[placid coyote]

they changed it very recently (today or yesterday?), it looks like this now. "auto" and "pro" uses sonnet for me (my model from settings)

[cobalt monolith]

For anyone wondering regarding the focus mode being deleted I asked the CEO on Twitter and he said it will come back

https://twitter.com/AravSrinivas/status/1887599183789490666

Aravind Srinivas (@AravSrinivas) on X

@mordechai_hadad Will come back

X · Feb 6, 2025, 8:28 PM

[tulip pawn]

gm. Who is the best person to talk about collabs and partnerships? At Talent Protocol we are launching a campagin to bring some builders into Perplexity and wanna buy some suscriptions in bulk to gift.

[distant juniper]

First time I see him confirm it. Thanks for correcting!

[stoic crater]

Hey! look at our DMs please.

[unborn viper]

huh

[karmic crane]

@proud cove

[tame current]

how can i use perplexity to do deep research like openai deep research?

[indigo valve]

I like what complexity did

[vernal geyser]

Can perplexity render diagrams?

[frank oyster]

Any clue on how you can create a new chat session on Pro support?

I have a chat session open on Pro Support and they basically are not responding

[green crystal]

No haven't tried it
Guess I got lucky because they responded after 10 minutes lol

Maybe try sending something again?

[indigo valve]

No but you can try using complexity extension''s canvas? I've never personally used that plugin

[frank oyster]

Tried multiple times without any luck :(

[runic tapir]

Guys I am a preplextiy pro sub how do I get access to perplexity finance

[torn path]

https://www.perplexity.ai/finance/AI

[quick barn]

good point, but if the reason is the Reddit API cost, why not just remove the Reddit option instead of the whole focus?

[burnt sparrow]

Okay so I was here like yesterday talking about using sonar in labs vs web vs MacOS client and I still feel like labs is the most powerful in terms of processing time.

Can someone confirm this?

[burnt sparrow]

Where do you see this at?

[indigo valve]

Complexity Extension for browser

[distant juniper]

Maybe its also related to academic journals, probably paying api fees for that too. But again this is all just guessing.

I assume focus is used mainly for academic journals and reddit. They already added the Writing focus back via

[burnt sparrow]

.....

[indigo valve]

Click on the "Pro" button on the left of prompt

It is consolidated

[burnt sparrow]

I get that but WHY the extension that makes 0 sense

the native client should be setup in that way not some extension

[burnt sparrow]

Hey bad design choice! You guys need to make this universal across all the clients. The inconsistency is making me nuts. I mean I'm going to be doing API calls soon so I don't have to worry about the UI but seriously.

[indigo valve]

This wasn't perplexity. The third party extension is trying to work with what they got with the website

In a way it's the websites fault but still

[runic tapir]

Thank you so much

[quick barn]

there are no fees for open access journals unless they are using a commercial api service that charges them.

[distant juniper]

You sure? I think Nature articles charge api fees. I'm pretty sure Perplexity needs commercial access to their articles.
Edit: According to some folks Perplexity only uses the abstracts so maybe I'm wrong.

[burnt sparrow]

Wait its a 3rd party extension?

[indigo valve]

Yeah

Life changing

[burnt sparrow]

oh wow okay, well then I want to check it out

Might help me with my own design of things with the API

[indigo valve]

@burnt sparrow The Ul update messed up the extension a lot but the dude is trying his best with what he got. Lot of QoL

[burnt sparrow]

Oh by all means this is WAY faster than the native macos client. I'm happy that you posted that.

[quick barn]

Web scrapping is not requiring an api. They reimplemented the academic focus just now after i emailed them. so the q is moot.

[distant juniper]

Which basically means they are trying to save up on reddit costs. If its that easy to enable academic focus why not just enable reddit?

[distant juniper]

PS: not all sources are retrieved from web scrapping. If they'd want access to articles beyond the abstract they'd need to pay fees.

would be illegal otherwise afaik

[quick barn]

i'm sure they know what they're doing, not my business to speculate 🙂

[distant juniper]

Neither is it mine but you were wondering why they wouldn't keep Academic focus so we started speculating :)

Anyhow I hope they re-enable the rest. I find the reddit one very useful.

[dim girder]

hello. I would like to inquire about refunding my perplexity api credit, but my emails to perplexity's [e-mail] and [e-mail] have not been answered for two weeks now.

Is it possible to inquire about credit refunds on discord?

[indigo valve]

I'm glad you like it. Mess with all the QoL. You used to be able to choose a space and recentcy straight from the search bar but the UI like said f'd it up. The developer of the mod should be fixing it. You can follow them on GitHub. Let me know what your favorite plugin is

[unique umbra]

Hell naw 💀 they added a free model on perplexity

And mf claiming it's more secure meanwhile he sayin he doesn't know nothin ab security

[burnt sparrow]

Yeah I see it bugged out. I had to uninstall it for now it kept changing models by itself.

I’m just going to have to use the 3 different apps to accomplish my tasks until I get the API going, it’s whatever.

Labs is still the most powerful for generating answers.

[indigo valve]

I never ran into that issue that's interesting. It doesn't change the default model just the conversation before start. Not sure if that was your issue. Give it some time and try again

[burnt sparrow]

nah I'm already doing API calls

I don't want to deal with all the extra latency

I am literally just making latex documents

So its all getting dumped into a .tex file anyway

I literally just wanted a doc generator for my manuscripts