#š¬āgeneral
1 messages Ā· Page 174 of 1
LETāS GO
Ppxl has even less value for $200 compared to OpenAI which is already stretching things
I wonder how o3 pro will even things out or not. They need to add more things to chatgpt pro 
10 deep researches per day for $20 would be pretty good
I wouldn't mind 5 or 10 deep research per day. Still pretty generous.
Wonder how many o3/opus prompts it would take to make the 20 a month plan a loss for them
Exactly š¤
taking into account all the token limits they bake in for responses
And plus
Otherwise people are going to leave
input tokens aren't thattt much
i don't have thinking mod ?
It's same as selecting any reasoning model.
I wonder if we will ever see another proprietary model from PPLX. Almost feels like advancements/the economics of building a new LLM has passed them by so theyāre focusing on 3rd party integrations
itās been pretty dry on that front for news
#š¬āgeneral message O3 but sounds too good to be true.
i don't have opus
Wdym
Another thing from that article, a memory feature might be coming for PPLX
O3 on perplexity sounds too good to be true.
Look at the screenshot. Zoom in.
True but it be op if they release everything on that image
If they release opus š
Provided model quality is retained.
People complain about it sometimes.
And this #š¬āgeneral message
Would be a major perplexity comeback.
Or downfall... considering economically
Most start ups run in loss in their early phases.
True
But opus and o3 at once is like winning a lottery and maybech together
Excellent, that it continues to increase quality, improving the service is essential.
Though they will rate limit it for sure
Man, 20 dollars is enough for a service of this type.
If you want a higher level, go and spend $200 a month
Be thankful that we have a service like Perplexity.ai
Yeah that I am buddy, m talking about would they increase the pricing for it, if they added both of them together
Yup
I'll stick with Perplexity, but if you want more context and stuff, well, you have to spend more money.
Yeah me using enterprise too so yeah gotta stick with it either way š
Excellent, take advantage.
This will be sweet
Also image gen made me laugh look @distant aurora it mistyped headache but still made me laugh š
Is that made with ChatGPT?
What up guys?
Made a terrible mistake at work and ended up liable for 3k USD
I dont think so
Where is o3 and claude 4 Opus ??????????
Lmao. Accurate.
The t is for turbulent.
Perplexity image gen
Also
By next year weāll have 25%
what was ur mistake ?
Looks like @aakashdahal_12117 was hacked, he is spamming fishing steam links
Signing a contract without checking in the offer was any good
I never understood this. What's the percentage for?
Opus 4 passes the decimal subtraction.
Sonnet 4 doesn't.
He probably sold his account.
Dude was acting weird the last time I saw him.
according to early testers the new mode is called Labs
2.5 Pro should also be added to the list
it's a good fit as well
Want to become early tester
In the new lab mode, currently for early testers
everyone does lol
Ah...
@cobalt vale
Rivals o3
š
Coming soon
All that overthinking and still can't give accurate answer to a math problem.
@cobalt vale you see this, right? #š¬āgeneral message
All that just to answer 18.
20 minutes of thinking.
No way
I don't trust this shit.
I don't really trust artificial analysis anyway
but does the original R1 pass
Wdym?
It always did.
I see a Claude killer.
Likely trained on Gemini.
Think you found the weirdest AI response?
Prove itāand score a hoodie!
Full announcement & prizes:
#š°āserver-news message
- Drop a screenshot + thread link in
https://discord.com/channels/1047197230748151888/1373806851358986320 - Vote for your faves with

Submit by June 1, vote by June 8.
probably claude as well
@dry kestrel @short narwhal
Very likely.
Yeah I checked out some benchmarks yesterday
looks good
on paper at least
Deepseek overfitting arc begins
@left stream @valid charm you seeing this?
Year
oh true
why...
There's this and there's vals ai
can't wait for someone to update twice in a month and end that
yeah its probably due to their servers being overloaded
you will prob get faster results using other providers
So⦠how can I have access to comet?
sign up to the waitlist
nothing else you can do
So yeah, got in the waitlist like 3 months ago and nothing
NEW DeepSeek????
Same.
Yeah.
Yeah, Iām not luck with that shit š© (sorry, just a vent)
Why not name it R2?
Only they know
I have a feeling r2 got delayed cuz of Trump's restrictions regarding ai hardware sales.
Hey, what happened to other models it's showing like this kindly let me know you guys are also seeing the same ?
Opus is not showing up on benchmarks yet right?
Looks like a visual bug to me, does restarting the app work?
It's there in the same image.
mehh
Force stop
Like are you guys also seeing same screen?
Working Mate, thanks for help š
Nice
Wouldn't I need Chinese host for the update?
Well, one day Iāll get comet š«
Not sure if us hosted will have it so soon.
Until there I think it is a legend
You might as well see an actual comet. Or a competitor.
There is a company hosting it already
The updated version?
yeah
Nice
how you getting labs option
What was your prompt?
The model doesn't know what it is, pretty normal behaviour
See provider status and make a load-balanced request to DeepSeek: R1 0528 - May 28th update to the original DeepSeek R1 Performance on par with OpenAI o1, but open-sourced and with fully open reasoning tokens. It's 671B parameters in size, with 37B active in an inference pass.
Fully open-source model.
@cobalt vale @dry kestrel looks fucked on you.com https://you.com/search?q=Length+of+AB&cid=c1_a5628f5a-0353-4538-8853-68d423707036&tbm=youchat
chat.minimax.io better.
Wait
They got rid of r1
Sad.
@cobalt vale @tame current check this out
wow
u tried it ?
Okay I tried it again. Gave me correct answer on second try. https://you.com/search?q=There's+a+square+edfb+of+6cm+inside+a+right+angled+triangle+abc+(b+being+the+right+angle)+with+hypotenuse+ac+%3D+20cm+passing+through+d.+Ab+passes+through+e.+Find+length+of+ab&cid=c1_4c108ed1-a028-4c38-82de-c4529bfed812&tbm=youchat
Nope. Just found it.
Fr.
Imagine browser saying: I ain't reading allat. But I'm happy for you. Or sorry that happened.
Project Feature is rolling out ?
my type of browser
Manus in Perplexity š
So grok is your type.
Grok does that if you upload too many documents.
what agent did you use
I use perplexity as browser, lmao.
Chatgpt after that.
And I still need browser for downloading movies or prawn.
OMG ITS HERE
Finally
50 for what? Projects or deepsearch? And what's special about it?
Labs
Where are the new models?
not out yet
this is not deeper research though
just a way to compile data into new forms
Old R1 on pplx
the research portion appears to be identical
Are pro perks region based?
Yep, US only
K, thx
Turns out turning off web search and asking labs to code just makes it identical to a regular query
Wooow! This looks fantastic guys!
https://x.com/perplexity_ai/status/1928141072011776088?s=46&t=_Dl0DSU0W7tiyPYnlKjlhQ
I'm trying it right now and it looks awesome!
It doesn't seems to be able to edit the same file though. I asked to create a random excel file and then edit it by only adding one more device to the comparison and it made a new whole table
new R1 loves to add <html lang="en">
coincidentally models such as claude and gemini do the same
Black box ai made Deepseek R1 paid.
Think you found the weirdest AI response?
Prove itāand score a hoodie!
Full announcement & prizes:
#š°āserver-news message
- Drop a screenshot + thread link in
https://discord.com/channels/1047197230748151888/1373806851358986320 - Vote for your faves with

Submit by June 1, vote by June 8.
Rip.
chat.minimax.io removed it.
Atleast monica still offers it for free.
And Merlin
@cobalt vale @dry kestrel so it generated the correct answer in the thought process but somehow ditched it.
I could bet you money this is them testing features of Comet
guys how is perplex labs??
Comet is new āwen GPT-5ā, aināt ever gunna happen and if it does itāll be a shadow of what it was hyped to be
how is labs
yeaš
wasnt it said to be in may?
we almost over may now
they gonna release it may 31st 23:59:59.999 trust
they better
acting like carti atp
so they're gonna release straight ass instead?
May 2077
huh so itās time based
What Iām saying is that itās a time based limit
So whatever it canāt do in 10 minutes, what happens?
is it always 10 minutes? Thought it would be an estimate
Let me run another
50 per month?!
It's an estimate, so it estimates that it will take that amount of time.
I am telling you right now, my query has been going for 15 minutes.
why do you need more?
cuz trying it out and getting some good or not, you'll exhaust it pretty quick no?
I'm having mine generate 9 images using GPT-1 Image, which usually takes a very long time for one. I will share once it is done.
i used the labs thing to make an app it was riddled with bugs, half the features it "implemented" didnt work, it made a ui for the stuff it put in but the ui doesnt do anything
at least it made it tho
so its still booty cheeks
better off using claude
I can 100% see this being the case for coding purposes. It generates code, then creates the "App", but that code can still have bugs, errors, etc.
It would be pretty neat if Perplexity made an IDE
Rumored to come with this new lab feature (seems to be that ānew deep research optionā that testingcatalog previewed this morning). So there should be a model selector for labs soon, along with Opus 4 and o3
I don't know how editing that app works, but I doubt it will work well first-shot.
but the fact it made a ui for the features that i requested and they didnt function at all is a bit insane
not even frontend is crazy
https://www.perplexity.ai/search/create-an-app-that-can-play-di-4bHkF4rfQzauhJ74d8nj.g?0=c specifically this is what i made
the "media controls" thing it "implemented" didnt work at all
which is like
not very good for an ai that had as much time as it wanted to make this
still kind of worked but like mehhhhhh
slightly dissapointed at the result
Idk man this is working for me perfectly š
The only thing not working is the "Open Display Window"
Does labs have a file upload limit to start the task? Hope it doesnt...
It works for me
Probably something on my end then
thats kind of like an essential thing that should be working
This is what I had generated, it's not an app, just to see what it would be capable of in the image generation department: https://www.perplexity.ai/search/i-d-like-to-create-a-3d-modeli-XxJu9YYmRwG5q9XgEsSSiQ?0=d
ok that's an actually good result
but image generation would probably be significantly easier to do as opposed to coding (which will likely be the use case for some people...)
This is very true, I feel as though more people will use it for the coding use-case, especially since it's more interesting.
love the new labs feature but it seems to generate graphs and charts with eplipses⦠obscuring the information its trying to present
Im pro now
Perplexity Labs that was released today? I'm using it for more finance stuff, haven't vibe coded with Perplexity yet
holy the labs feature is so cool
in perplexity do you actually get the models e.g. claude 4 sonnet or do they nerf them or is the quality the same can someone explain how are they doing that for just 20 bucks
context window is less
how much
No. That's perplexity's newsstand or something like that.
š¬ 1.2Kāš 0āš 26Kāš 0
Common sense doesnāt exist in England?
#comedyreels #foryou #comedyreel
Ohh
@green crystal since when is chatgpt such a disappointment?
š
@cobalt vale @minor shard @valid charm @left stream
perplexity the goat
Think you found the weirdest AI response?
Prove itāand score a hoodie!
Full announcement & prizes:
#š°āserver-news message
- Drop a screenshot + thread link in
https://discord.com/channels/1047197230748151888/1373806851358986320 - Vote for your faves with

Submit by June 1, vote by June 8.
capped at 32k
whats that in real world use
@dry kestrel perplexity free search doing better than o4 mini search on chatgpt.
I havenāt really had any issues with it
both input and output is counted as a part of this token, paste some text into this to see for yourself
if you reach this context window limit, the model will stop retaining information and would hallucinate a whole lot more
for example if you give it a text that is 64k tokens, it will not be able to read the rest of the 32k tokens so if you ask for a summary, it will start making shit up
interesting
is that using sonar
Whatever the free account uses by default.
This was pro search.
It's labelled pro as you can see.
it could use either sonar pro or another model
or the free sonar
The free one isn't labelled pro.
I didn't use my 3 trials.
Kept pro disabled.
you are contradicting yourself
you just said it was a pro search
Second screenshot is pro. First screenshot is free.
I have 2 accounts.
Ok
A pro and a free.
Pro searches are labelled pro.
I'll just share links.
Yeah. You mentioned claude. I didn't have to.
thought for 1 minute š³
o3 is the thinker behind r1
It does it without Claude in it too
Besides, you used O3.
I used o4 mini.
I'll try it second time too. Just in case.
Okay, it did it correctly the third time. https://chatgpt.com/share/6838c3cf-fea4-8013-a8bd-7cea682d4c14
Yeah itās not the best for analysis
O3 is fantastic imo
Itās thinking can be very thorough
Also, is it just me or can I not enable thinking and web search simultaneously? Like it automatically decides whether or not it wants to search the web irrespective of whether or not I enable it.
Do you have plus because i cannot choose the model as i use chat gpt free
Even for non reasoning prompts it searches on it's own even if I don't enable it. Not like I mind it. I'm just asking if others are experiencing this too.
Yes
I was able to enable both but usually I donāt need to cause it almost always searches the internet
Undefeated
Oh ur saying you donāt want it to. Oh idk then. Yeah
Yeah itās always gonna search haha
How does opus 4 cost half as much as O3 on the way in? š
Or is that hallucination?
Oh. Nvm. Half again.
Bad wording lol
It meant 1.5x.
Yeah the wording didnāt make sense
Yeah got it. My bad.
How possible i can open the ZIP file from Perplexity Lab please .. if i open that URL it redict me to usually site where is nothing
As in it didn't work as you expected?
Yeah.
It's a complex task I use to test deep research/multi step solvers.
Only google, open ai, and grok have done it successfully so far.
With their paid models or free ones
And grok didn't do it in one go. I had to kick it around quite a bit.
Free actually.
It do be like that sometimes
I thought Labs only works for pro users as per the pplx launch page
But saw a picture here saying free users get 3 queries per day
Conflicting
Yes. I meant free of Google, chatgpt and grok.
Got you
I have a month of free perplexity pro.
Yeah especially because it's just launched
Yes
It can make presentations.
Yeahh
Reasoning is still very off though.
šļø
it's 50/month???
Graphs and stuff
š¤®
o
I would have been happy with 50/week
It's meh https://rebrand.ly/3eukxq6
They'll prob end up tweaking the numbers
It's not deeper research
the source gathering part is identical to research
Yeah.
@minor shard you have a O3 search to waste?
Or I'll ask joe.
I know right. Pretty cool.
hi
At least this better not be the new Deep Research High that's coming "very soon" 21 days ago
Sonar reasoning pro seems pretty fast for a reasoning model.
it's just R1
hosted by Perplexity
That explains it.
So is perplexity gonna make a R1 1777 or something?
Based on the new R1?
By the time that's done R2 might have released
Erm 1776 stands for the year that America became independent
1783 it is.
not per day??????
just 50?
This 50 per month vs O3 100/week.
comparing apples to oranges
What should I use it for?
Why is bro so defensive š
dawg what š
no ones forcing you to use it
is that wrong? š
just dont use it if u dont find a reason to
Hey cloudy, did you get that new phone?
But I want to know how to use it in case an opportunity arises in the future.
2 months is crazy
Developing apps. Sounds like a good one.
He might as well travel and bring it with him within 2 weeks.
He restored his account, alliluah
Did you read last days news about ussr continuation? Legal ussr still exist by the name russia federation
appreciate the work has been done on lab. did a usual routine using lab instead. much appreciate the work done. wanna explore how to use it more properly. to me 50 per month is not a problem, i actually prefer a even more deeper research over my imagination than just increasing more queries.
Labs has some amazing promise. I have a complex medical condition that transpired from a procedure complication back in November. Seemingly being able to pull more data and analyze it and format it better is great.
One of my annoyances has been the inability to copy a table and they extend longer than the screen so you canāt screenshot them either.
This is seemingly from how I understand it fixed with labs seeming ability to format better into a document. I could be wrong and Iāll know for sure once my query is done.
I do really appreciate the addition of estimated time remaining. For non instant searches I think this would be. Powerful to include even outside of labs. Sometimes I get annoyed when I have no idea how long something will take and it keeps going. Even if the estimated time changes up or down as the search progresses
Also is labs the new name for comet or is comet still coming and just not ready yet?
Comet is still coming, Labs is just a feature they recently added to Perplexity.
can someone explain what projects do?
Omg Labs is even BETTER than I thought. We get a downloadable CSV file!! This is something I actually desperately needed
When do you think Comet will drop? I've been on the waiting list forever
Perplexity Labs can help you complete a variety of work and personal projects. From creating a marketing campaign, to analyzing your business finances, to meal planning for the week, Labs are where you turn your best ideas into reality.
yeah just checked out the labs feature myself. looks amazing. But any idea what are its rate limits?
for pro users?
3 weeks(?) ago aravind said that max 5 weeks, so soon hopefully š
50 per month including follow ups
Fingers crossed
Follow ups (with labs turned on the thread)
hmmm, that seems very low as of now :(
i think i already used up 10-15 because i was experimenting with it and found it very very helpful
hopefully yeah.
That's why im very cautious and decided to build a completely thought out prompt before starting to use labs
yeah, imma start doing that now for the remaining prompts i got
Follow-ups meaning follow up labs correct not just follow ups to a lab with a say perplexity research
Anyone here have any idea for the context tokens for labs? (Should be 32k as usual?)
Yes
You can do searches without counting to labs
You've to switch labs on
also deep research is unlimited right? or does that also have any limits?
It's also frustrating how they don't show you the count until it's about to finish
Yeah, has the same limits as pro search
I used it in an already pre established thread that has a bunch of info in it already very helpful there
DeepSeek R1: 25 per day
Deep Research: Unlimited for Pro
This it what it says
That would be a nice feature to add a count even if they want to hide it in settings for aesthetic purposes
Great to know that we can use it as a follow up
deepseek r1 is one of the output models for pro search right? sooo, should be equal to the pro searches per day limit?
Yup but also not sure if we're offered deep seek as a model
the r1 1776 is offered as a reasoning model
Think you found the weirdest AI response?
Prove itāand score a hoodie!
Full announcement & prizes:
#š°āserver-news message
- Drop a screenshot + thread link in
https://discord.com/channels/1047197230748151888/1373806851358986320 - Vote for your faves with

Submit by June 1, vote by June 8.
hmmm, okay correct me if i am wrong but this is what i understand.
-
Pro searches: 300/day limit. When I use any AI model with search enabled, it only counts against this Pro search quota, not the individual model limits.
-
Model usage without search: After hitting the Pro search limit, I can still chat with the models without search functionality - this has its own separate 300/day limit.
-
Deep Research: Another separate 300/day limit for this feature.
So if I'm doing the math right, that's potentially 900 total queries per day (300 + 300 + 300) across all features combined.
Am I understanding the quota system correctly here?
and labs is also a plus with 50 per month from what i understand
hello I am having a problem with the perplexity student program.
I got enough referrals to get the sticker pack but when I add the sticker pack and apply the discount code it says invalid. Please help
I can confirm only on the labs one as im only quite sure of that, but @short narwhal might know for sure on the others.
should be 600 a day for both pro search and research afaik
Should be on the same limit
This is what pplx tells me
combined or separate?
Combined
okay but what if i turn off browsing and continue talking to models like normal? is that considered a pro search or does that have a separate limit?
Hit the thumbs down on this one, seems to be using old sources or smth...
The table makes no sense because how can there be a count on pro searches but none on deep research
yeah i have asked perplexity about the limits, it always shows me different results lol
deepseek r1 is not 25 a day
same thing
Again pplx hallucinating mb
alright
Same limit
Dont ask pplx about its own usage limits
it doesnt know
Ok
Anyone pls?
also what is pplx? are you just refering to perplexity in short or is it the API?
Yes short
ok
i suggest drop an email from your student email id or try the system contact option
how to do the system contact
from the contact button found in "my account" at your homepage
yess, research has a similiar one
I assume itās opus based on the CEO tweeting previously about a feature in perplexity being capable of making apps and featuring opus in it
š„
Anything with any sort of depth needs reasoning for search
Else it can hallucinate badly
Im not seeing one when im researching on android device
I don't think android has live activities like iOS
Maybe it's a dynamic island thing
yep
Sorry, I missed you comment. Yes itsis for Sonar API.
Perplexity labs my beloved
Inb4 deep research high is the same as labs but less graphs and images/s
No worries at all! Since itās for the Sonar API, Iād really recommend posting your question on our Developer Forum ā thatās the main place where the API team tracks and responds to feedback and support requests!
$300 per month sub called perplexity elite coming soon?

the research in labs is identical to old deep research though
Yeah I saw the reasoning is the same
Something tells me deep research high will be capped at like 50 or 30 per month
nope they stated 50/month
And a $200-$300 perplexity elite sub will actually be a thing 
Bonus points if they call it perplexity pro max or ultra
What model powers deeper research
there's still no deeper research, this is regular research with advanced visualisations
i just blew threw like 20 queires lol
Dang
introducing...
Perplexity Pro... Pro.
Pplx pr pr pr
Thank you. So can I ask about disappeared credit in Developer's Forum? It is about billing though.
can i get free perplexity for 1 year like everyone else did when i paid for it. my year long sub just expired. after 1 year of free perplexity i will pay up front for 2 years. thank you and be well
@signal hamlet
Is the labs function a good fit for world building?
if you're european, get on Deutsche Telekom or one of their daughter companies like T-Mobile in some countries, and claim the free year offer
its legitimately the best offer i've ever seen on anything
Or just get it through Xfinity I think still.
I would highly suggest against @'ing the admins asking for free Perplexity.
That too
how can I get access to Comet?\
I literally sent screenshot with reasoning enabled right below that.
It's 300+ for anything that wasn't free.
They reduced it to 300+ a while ago.
@atomic shell
Does perplexity have nee deepseek 0528 model?
Think you found the weirdest AI response?
Prove itāand score a hoodie!
Full announcement & prizes:
#š°āserver-news message
- Drop a screenshot + thread link in
https://discord.com/channels/1047197230748151888/1373806851358986320 - Vote for your faves with

Submit by June 1, vote by June 8.
Does perplexity have new deepseek 0528 model? Can anyone pls confirm
ah i didnt look closer lol
No. Perplexity uses it's custom R1 1776.
Thanks for letting me know bro.
how is labs different than research?
havent had a chance to try it yet
bros are doing anything but releasing comet
is anyone using enterprise pro? while replying to this please tag @rigid ether and reply, because i am active at @rigid ether

Huh
I got enterprise pro
$40 a month is crazy
Luckily I got the year subscription so I donāt have to pay $40 a month
Don't you have a you.com and chatgpt sub as well lol
Yes
all im missing is the gemini ultra and pplx enterprise sub
Gemini would need lots of upgrades to convince me for the gemini ultra
Use edu for one year free
Yeah and also the problem is I dont need the 30tb
Wait 30TB for Gemini ultra?
Also 2.5 pro still has the same old issue of "sorry but Im just an LLM and cant do this"
i think so
Ikr
and yt premium
Bet I rather stick with chatgpt pro unless they release something really really really good like gemini 4 or sum
yeah same
Im just keeping my gemini pro for that 2tb
and occasional use of 2.5 pro
Arenāt you the other person who also has pro on chatgpt right?
yeah
Same
its awesome
it better
they better be releasing it soon
also waiting for grok 3.5 and bigbrain mode
True
Bigbrain mode?
that was in the demo when they anounced grok 3 i think
Ah I think I missed it tbh
Ngl that looks sweet
Its been 3 months though
Iām hoping grok 3.5 is good asf too
and a month since elon said he was releasing 3.5
Ikr I would pay but I rather pay claude tbh
Claude limits suck
Claude impressed me with claude 4 opus tbh
True
You.com somehow had unlimited claude 3 opus when claude 3 opus was limited asf on the native app
Imagine claude 4 opus
tbh they have their ways of limiting cost
claude on you.com is not useable for coding or for long outputs
š
Yeah Iāve seen how low the messages can be
Bro that google drive connector, are the answers accurate?
I won't be surprised if they annouce projects using opus
I'm pretty sure they stop long responses on purpose, Ive raised this issue multiple times to the devs and they keep saying they fixed it and its never fixed
that's guaranteed at this point from the Aravind and Testingcatalog posts
Wdym projects? Like apps, docs, etc.
Yeah o3 had to do it with answers split
like the new projects feature
thats limited to 50 a month
its kinda like manus/flowith
are you using any extensions
like complexity
if so disable it
I think Complexity has fixed this
new 4o image competitor btw
no i am not
is it availiable on desktop client
nvm i got access
...
How is it that I have 16 mutual servers with EdDi and only 3 with you
Think you found the weirdest AI response?
Prove itāand score a hoodie!
Full announcement & prizes:
#š°āserver-news message
- Drop a screenshot + thread link in
https://discord.com/channels/1047197230748151888/1373806851358986320 - Vote for your faves with

Submit by June 1, vote by June 8.
Where is 4 Opus & o3 ???? why we always have 2nd class ?
nope joined for a couple months
Theres some mutual servers thats kinda closely related
cost
o4 mini high is better than o3
okay nice info thanks
ah i see i see
dont know
i hop eso
@modern sinew how you got bug sniper role
I reported a bug
Random scrolling
oh yeah u could easily find one
i swear if perplexity has some kind of context and output tokens limit
im gonna cancel my subscription
damm i do a lot of typo
oh ya these things are frustrating
I am trying it
The 1 kind of glitch or kind of bug i noticed is, the timer sometimes moves very fast from 10 - 6minutes in a minutes and sometimes it gets very very slow
i guess its processing time
yep, this was happening to me as well
btw can you do me a favor ?
I think the timer starts at 10 minutes default
What is it?
yes then it re assigns time i guess
Yes it is 10 mins
can you vote on my wrongest answer š
pretty much everything has a context limit, only API and Google AI Studio are unrestricted
perplexity has limit?
they do, its capped at 32k
I hate that
i just came to know about that
same as chatgpt plus
is it still 200k for 2.5 Pro?
Absolute beauty locked in a room
dont know
Doubt that, would be more plausible if it was 4.1 mini or like 2.5 flash
I agree, but capping it to 32k for bigger research models makes it less detailed or sometimes for coding it just bugs up the whole code
some times deleted the whole codebase
I think deep research gets you 128k
sure
and deepseek r1
@dry kestrel can you also ?
wait what
it gives url ?
Perplexity doesn't care about coding at all though
that so cool
Done
Yes it hosts it on there backend
I pretty much forgot perplexity is not meant for heavy coding
its cool but found a bug in it
that upside progress bar doe snot count progress precisely
althoug its cool
O3 mini reasoning also 128K, i got this confirmation during march or April 2025 from Kwak
But its sad they don't maintain this information in their website
yep r1, DR, not sure about o4 mini if they kept the 128k
Am I missing something or Perplexity labs not yet available in the Mac application? On my iPad Iāve found but on my Macbook not yet š¤
Think you found the weirdest AI response?
Prove itāand score a hoodie!
Full announcement & prizes:
#š°āserver-news message
- Drop a screenshot + thread link in
https://discord.com/channels/1047197230748151888/1373806851358986320 - Vote for your faves with

Submit by June 1, vote by June 8.
Omg free perplexity for a year š® is it Christmas? š
When is perplexity lab functionality coming to mobile phone apps (iOS) ?
Already there
Guess this is only valid in the UK. I have not seen this over here in Germany, but here in Germany, we had Deutsche Telekom offering such a plan for a whole year a couple of months ago.
Thanks much. I just had to update the app! š
What about the offer via Revolute bank, they give perplexity subscription for £7 a month.
Itās coming soon to the Mac and windows apps
Awesome, thx.
Creating projects yes
labs is available in windows app
Oh so the blog is wrong :P
That's awesome wow
Do share here
thats insanely good
Damn they really did something right with labs
now if they could just please release deep research high with a high limit
genuinely nuts
Truly
this is gonna be revolutionary for my future re runs
umm....
what is this
They do Anything but increasing context tokens š¤£
strange it didnt make an app for you
Yeah that's weird
i asked for website
maybe dont use the word website explicitly
try to say interactive cheat sheet or something
hmm
Does somebody know how long it takes to retrain a model like Deepseek? im sure Perpelity is doing this again for the updated version. im just curious how long a training takes until we get the new Deepseek version in Perplexity
fr
Think you found the weirdest AI response?
Prove itāand score a hoodie!
Full announcement & prizes:
#š°āserver-news message
- Drop a screenshot + thread link in
https://discord.com/channels/1047197230748151888/1373806851358986320 - Vote for your faves with

Submit by June 1, vote by June 8.
I am planning to purchase perplexity for one year at a time, but i am worried its a good decision or bad decision
Dope
Depends on your use cases, do your research before purchasing
I want an AI which shall have custom instructions and custom files (through connectors),
Already using chatgpt teams plan,
I need another platform for double checking my answers
Spaces limits pro users to 50 files per space and custom instructions
If that's good enough for you
You can consider pplx
Yeah šš»
Claude max is expensive,
Grok is cheap, but i hate their Buggy UI
So only choice at this moment is perplexity
go for it shit is great
How can it help for software engineers? Just want to know use cases of Perplexity to software engineers
prolly not much, swes are better off with copilot or sum
perplexity has never been geared towards software development
Ok bro šš»
I mostly use perplexity to study or learn something, and usually I'm the one providing the sources, which is ironic
can i get comet browser access
Join the waitlist
i did this during my endsems, pplx is the reason i passed lol. but even then i think pplx is an excellent reasearch/learning tool, and of course a search engine replacement and it can help you code as much as stack overflow can. thats my view on things
yeah, exactly.
I have colleagues asking me things about teachers slides, and I'm like "brother, I haven't even opened them and passed" cuz I just dump them all into pplx context and learn from it ahah
anyone have download link of comet browser
Efficient
btw when I do this, I untick web and any other online sources, so pplx only uses my local context files and instructions, no grabbing random things online.
its very useful
Anyone had luck with dashboards in labs? I'm really keen to see it build one from an upload of Entra sign-in logs, showing stuff like top apps, time of day distribution, failures by location, etc and it gets the response correct but it's not visualising anything. Maybe it's limited or I'm missing something.
yeah, I really wish pplx had built in canvas/artifact, not sure why they haven't implemented that
I might have to decrease my bread so I can study more š¢
that's what I did
Yeah I gotta do 5 classes for cybersec and other certs. Iām fucked if I do work too, Iāll do part time for now š
I dropped work completely ahah, don't advise to anyone
it made the website but its not functional
Yeah Iām just gonna slightly drop it not completely on weekends though since I want some side cash too. At least college is covered tho
What website ?
Just enough to party or go to mexico
.
yeah, better safe than sorry
good for you tho
studying
Hi guys, does someone know if Imagen 4 will be added to image generation ?
Why people are not talking about claude4 in Perplexity? Did it not make any difference?
It's my default model. Claude is really great for a lot of tasks, particullary in programmation.
So you are doing programming with help of claude in Perplexity? Is it not causing you issues due to 32000 context window?
Yep ! No, not really, Claude is just doing some bug correction or feature integration, not programming for me. So yes, sometimes, it's a bit just but most of the time, it's great.
Good to hear
And anyone uses spaces feature a lot here? Which model is accurate for RAG in spaces in Perplexity?
I use it to explain things from my classes that I don't understand or like here, to study. I use Claude 4 or Grok 3 mostly but that's from my favorite answers.
ever since I got pplx I only, most exclusively, use Claude models.
they seem to just deliver better
at least for me
I think that Claude are, currently, the best LLM on the market.
And Perplexity users subscribe only for that search? Or it is also replacement of chatgpt plus subscription?
Personaly, I use Perplexity Pro as a replacement of mosts AI. It's perfect for almost anything except chat.
like, for example, just now I gave pplx like 5 small pdfs. Tried using deep research and bro didn't even acknowledge the existence of those 5 pdfs. While Claude 4 Reason devoured all 5 and delivered.
Labs are very good with files too. Don't know the model use but here, it sums my whole 300 pages PDF.
Lol it means claude search is powerful than perplexity deep research?
In labs how many files it can research?
?
Bro what do u mean by except chat?
If you want to have a discussion, pass your way. It's not really good due to UI.
Can't tell, I'm working with one file at time, they are really big
Oh i just want to stop chatgpt sub and start this, i will chat in single conversation for 30 to 50 turns for each task. Thats the reason
in my experience claude reasoning, seems to be
Yeah it's just more like follow ups instead of a conversation
Pplx essentially just disgests content for you instead of a chat
Main focus is on providing the requested information rather than a focused back and forth
In my experience
Oh completely got it what you mean
Why people are paying for this? Even though chatgpt can deliver same, may be due to many models at one place.
Think you found the weirdest AI response?
Prove itāand score a hoodie!
Full announcement & prizes:
#š°āserver-news message
- Drop a screenshot + thread link in
https://discord.com/channels/1047197230748151888/1373806851358986320 - Vote for your faves with

Submit by June 1, vote by June 8.
Domt want hoodie, we want to win free subscription
ok bot
Multiple models, restriction only on context tokens, web search, you get labs, research etc
Spaces and image gen (whenever they choose to work lol)
pplx merch and art are beautiful, beautiful aesthetics.
If only labs wasn't so restrictive it could make itself a seperate product
People would be willing to just get it
What restrictions you saw at labs
50 requests per month plus limited context tokens and response tokens
Is it same for enterprise pro subscription
I may go for enterprise pro, because i need google drive realtime context
hello helo
Hi
Check with support
Ok bro
does anyone else feel that claude 4 is worse than 3.7?
Not the thinking
But the normal
Alr Iāve been using thinking for SEO text generation. It was lightyears ahead any other model but now its terrible
Really????
Damnā¦.
Its worthless now
Before it ranked 20-25% higher in general tests than Gemini 2.5 and ChatGPT4-1, as well as Perplexty reasoning & o4-mini
Crois en toi, avance chaque jour, mĆŖme si cāest un petit pas. Les efforts dāaujourdāhui construisent les rĆ©ussites de demain. Tu es plus fort que tu ne le pensesāÆ!
now its always 30-40% worse
Just a simple test, ask models to rank the text 1-10. Claude 3.7 always hit 8+ and now same prompts but Claude 4 Thinking gives me 5-6
@vague pagoda help him when you are here.
#ā”āask-community might help.
mind sharing some extra details - is the issue with redemption or something else?
we can't assist with billing via discord - you'd need to email support@perplexity.ai for that
dms pls
is it only 50 requests per month even if you're a pro subscriber?
well that simply means the team is going to increase the limits and release a $200 a month subscription for it XD
Pourquoi l'assistant ne rƩpond pas
redƩmarrer l'application
Oui oui
How does one access Labs?
Web, windows app, Android
@dry kestrel apparently Claude 3.7 never works on you.com. Even 3.5 doesn't. And 4 doesn't support images yet.
J'ai redƩmarrer plusieurs fois et Ƨa marche pas en revanche l'assistant par dƩfaut il marche
Ceci marche normal
Okay, I ran an image analysis test.
So far only chatgpt has passed.
Grok failed miserably.
Perplexity didn't pass
Claude and Gemini failed too which was weird.
What test is that
I gave it an image and asked it to extract information from it.
Gemini 2.5 pro fails on Gemini website but passes in ai studio.
That would mean ai studio is better.
The weird part is that native Gemini fails but Gemini on perplexity and you.com pass.
And o4 mini on perplexity and you.com failed but passed natively and on t3 chat.
@dry kestrel @cobalt vale you two wanna test/verify this?
what
Not wierd, systematically done, in best interest of the AI manager.
Ich will mal etwas loswerden. Leute, es gibt so viele die kodieren, Algorithmen mittels Programme herstellen, apps herstellen usw.
Wieso wird eigentlich nirgends über die Tatsachen geschrieben, wieviele faule Tricks die AI-Anbieter drauf haben. Obwohl man doch für die pro Versionen zahlt, wird man ganz derb vom KI "angelogen", und da sind einige Tricks, die sie drauf haben.
Habe ich vom ersten Tag der AI-Nutzung 2 Wochen getestet, Perplexity ist ja gar nicht anders. Die wollen den Dienst, den sie für bezahlende Kunden eigentlich erbringen müssten, nicht tun.
Wo geholfen werden soll wird systemstisch ausgewählt vom System, allen anderen lügt die KI an. Das ganz dreist und bis zum bitteren Ende.
Was denkt ihr über diese faulen Tricks und die meiner Meinung nach hin und wieder gefährlich für Kunden werden kann, obwohl die systematische Lügen-Maschinerie an sich ist schon gefährlich.
Ausserdem kann man in Perplexity ios app verschiedene KI-Modelle selber auswƤhlen, das ist doch auch nur ein Trick, und ganz einfach zu beweisen.
Ich würde gerne eure Meinung dazu hören.
(Werde gleich mal diesen App kündigen)
I was in rage as I wrote, well, I can manage it.
I want to get something off my chest. People, there are so many people coding, creating algorithms using programs, creating apps, etc.
Why isn't anyone writing about the facts of how many lazy tricks the AI providers have up their sleeves? Although you pay for the pro versions, you are "lied to" by the AI, and there are some tricks they have up their sleeve.
I tested it for 2 weeks from the first day of using AI, Perplexity is no different. They don't want to do the service they are supposed to provide for paying customers.
The system systemically selects where help should be provided and the AI lies to everyone else. They do this brazenly and to the bitter end.
What do you think about these lazy tricks, which in my opinion can be dangerous for customers from time to time, although the systematic lying machinery is dangerous in itself.
Besides, in Perplexity ios app you can choose different AI models yourself, which is also just a trick, and very easy to prove.
I would like to hear your opinion on this.
(I'm about to cancel this app)
Translated with DeepL (https://www.deepl.com/app/?utm_source=ios&utm_medium=app&utm_campaign=share-translation)
Think you found the weirdest AI response?
Prove itāand score a hoodie!
Full announcement & prizes:
#š°āserver-news message
- Drop a screenshot + thread link in
https://discord.com/channels/1047197230748151888/1373806851358986320 - Vote for your faves with

Submit by June 1, vote by June 8.
I want to get something off my chest. People, there are so many people coding, creating algorithms using programs, creating apps, etc.
Why isn't anyone writing about the facts of how many lazy tricks the AI providers have up their sleeves? Although you pay for the pro versions, you are "lied to" by the AI, and there are some tricks they have up their sleeve.
I tested it for 2 weeks from the first day of using AI, Perplexity is no different. They don't want to do the service they are supposed to provide for paying customers.
The system systemically selects where help should be provided and the AI lies to everyone else. They do this brazenly and to the bitter end.
What do you think about these lazy tricks, which in my opinion can be dangerous for customers from time to time, although the systematic lying machinery is dangerous in itself.
Besides, in Perplexity ios app you can choose different AI models yourself, which is also just a trick, and very easy to prove.
I would like to hear your opinion on this.
(I'm about to cancel this app)
Translated with DeepL (https://www.deepl.com/app/?utm_source=ios&utm_medium=app&utm_campaign=share-translation)
What's the lie about?
guys anyone have any whatsapp userbot experinece?
The system systematically selects where help should be provided and AI lies to everyone else.
Explain this part.
Is opus out yet
I don't see it yet.
Within labs i think you can choose opus
Bet william
š
They said they were releasing Opus as part of a feature. Not by itself to my knowledge. I think its part of Perplexity Labs but I could be wrong.
I didn't see any model switcher there.
Yeah you don't choose the model.
Me coding with labs
Dang at least theres claude api
As long as you have AI, thats possible lol
Apple watch developer edition.
do you understand about lasers of this smart watches ?
I do
Kinda looks under the skin.
Different laser for different layer.
I have a xioami s4 and my skin have burned i have pictures i can post here i think a lot of smartwatches have big problems in this
Blood Pressure
Guy's
can i post a picture here ?
Yeah it can be sensitive
Sure
Damn š
Improve perplexity voice chat transcription, is soooooooooooooooo bad asf
Hmm, I could just copy the conversation. But I am not keen about show something if there is nobody complaining. Wanted just hear some other opinions.
Lana del Rey
Wtf
after 3 days using smart watch
Bro lost his hair.
yeah now its normal i stoped use
What country are you in?
I think you have sensitive skin or the device malfunctioned

