💬│general

[wicked cloud]

i tested on a multiple copy of a project

and yeah 3.0 really likes to alter as much as it wants

compared to even flash and claude 4.5 before

[left stream]

I rarely use it since it was garbage in comparison to opus 4.5

[wicked cloud]

far more garbage holy

3.0 pro hallucinate and do as it please

it deletes a folder once during a testing

meanwhile the flash version isnt and still nail the test

[quaint ravine]

yes 3.1 is improvement for my use case too

but i'd say Claude 4.6 is bigger leap

[wicked cloud]

it is heavy leap for a short time holy, i am loving 3.1 pro so much

[wise nova]

https://www.reddit.com/r/kimi/s/jYse9G8cA5

[quaint ravine]

4.6 Sonnet actually not much different to Opus 4.6 for me too

[wicked cloud]

i think its the sonnet 5 right but they decide to rename it i think to 4.6 following opus

[quaint ravine]

maybe

[wicked cloud]

they wait for sonnet 5, improving it further until major ai release again

they really like that for anthropic

[quaint ravine]

respond to GPT-5.3 😛

all good for us

the incremental improvements are too small for me 😛

[wicked cloud]

they wait for a rival to release a major model and then nail in the coffin

[quaint ravine]

need more lol

[wicked cloud]

the 5.1 to 5.2 is also major leap in personality

holt

[dry kestrel]

It's in beta right now...

[wicked cloud]

i am waiting for sonnet 5, i heard people dont like 4.6 cold in answer sometimes

hope they fixed that, though since i use it in coding i didnt need much personality

as long as it get the jobs done

claude tends to do heavy leaps in short times, so yeah

[quaint ravine]

yea same

dun really care about personality...as long as it's concise

[wicked cloud]

but somehow opus 4.6 have more personality though

[left stream]

I won’t lie sonnet is fast asf

opus is still superior but speed is showing now

[wicked cloud]

i think its because more advanced model?

[wicked cloud]

yup fast is a definitely

[quaint ravine]

I've seen Opus 4.6 get something wrong where Sonnet 4.6 got right

I don't think it's that clear in at least some use cases that Opus is clearly better

[wicked cloud]

smaller model with a strong instruction sometimes tends to be better than huge model with more complex architecture

also i noticed i get the research and lab queries at the same time as normal pro

yet when i use pro they are also depleted

despite me not using them

so my monthly quota gone when i am not using them

[quaint ravine]

that shouldn't be the case

u mean when u use a Pro query within PPLX, research and lab also goes down?

[wicked cloud]

yes

so my 20 labs and research are now gone

without me using me

[burnt zephyr]

lmao

[wicked cloud]

PPLX is great

[wise nova]

Try Haiku

[burnt zephyr]

haiku won't stop yapping and eats more than sonnet 😭

[wicked cloud]

perplexit best model sucks too for pro queries

[left stream]

I’m lucky that doesn’t happen for my use case in cybersecurity, been using it for hack the box modules and helping me study while learning penetration testing. The speed definitely goes to sonnet I won’t lie, both are monsters in coding

[wicked cloud]

havent tried haiku much is it actually like that

[left stream]

gemini 3.1 though I have more to explore

[wicked cloud]

they are currently top alongside 5.3 codex high

[left stream]

haven’t tested it as thoroughly

[wicked cloud]

i mean grok

eh does it really count

[left stream]

we can just ignore grok and focus on the 3 that bring good models

[wicked cloud]

grok honestly being glazed by elon himself

and the new 4.2 is just 4 of 4.1 models with different hats

[burnt zephyr]

I hope a new deepseek soon drops

[modest latch]

Fah

[left stream]

okay gemini 3.1 pro impressed me

same with claude

both together solved a real mystery physics problem that I was wondering what the cause of the problem was

the solution sounded crazy but it actually worked

well damn ai is the future

[quaint ravine]

wow....new setting for PPLX

also for those who hate/hated the new font in PPLX

just change it back here:

to Sans-serif

[quaint ravine]

https://www.youtube.com/watch?v=vZd6v9qgq7M

echoes a lot of comments in discord and reddit

[silent ocean]

sure

[quaint ravine]

i think an eg. 50 dollar plan is needed in PPLX

i suspect a lot of people would use it

rather than just (think of) quit(ting) like that youtuber

actually sorry, it wasn't regarding new limits

[forest finch]

hi lol

[burnt zephyr]

Why?

Maybe if you had opus

[quaint ravine]

gives more limits

[last oyster]

yo

where is my pro subscription

it was supposed to end only in September this year

[wicked cloud]

If you get it through illegal means yeah

They are currently on a suspending wave for many account that get their subscription through illegal means

[last oyster]

https://cdn.discordapp.com/attachments/1318259919194947704/1458749154468696156/ScreenRecording_01-08-202608-47-35_1-ezgif.com-gif-maker-2.gif

I wonder what rules I broke

[twilit flax]

Hi

[haughty pond]

Why suddenly removed free plan??

[split dome]

We need Gemini 3.1 Pro

[umbral cypress]

Hi, I'm a student who uses Perplexity for academic research and resumes. My Pro account was suddenly banned, I believe by a false positive. Email support ([e-mail]) is entirely automated and won't let me appeal to a human. Can a staff member please look into my account?

[spare karma]

Same here as Ashok

[woeful pulsar]

Is education pro monthly free trial available?

[jaunty sierra]

Commercial use is the first line.

So no market research, no copywriting, etc. etc. etc.

[winged jay]

Same here. I have no way of finding out why my subscription was terminated

[crisp kestrel]

Guys did they remove the 5$ promotional api credits per month for pro users?

[limpid terrace]

I went to search for it, but to be honest I haven't used this API yet

[crisp kestrel]

i relied on it and now its gone

without any heads up

[limpid terrace]

I have a classmate who encountered this situation 3 months ago

[limpid terrace]

I still have this

[crisp kestrel]

i dont have it

[limpid terrace]

Is your pro subscription still there?

[hexed olive]

Can someone here make a presentation for me using pro?

[golden palm]

hi ya all

[misty sandal]

I can't even use any of the reasoning models anymore because they all get stuck in an endless loop without giving me any output 😭

[golden palm]

Aghahahahaha

[misty sandal]

Not really funny.

[solid latch]

Israel

Tel aviv

Damask

[unborn bane]

My perplexity pro is supposed to be active till Jan 2027 but suddenly it stopped working after Feb 3 2026. I have reported that to the support channel in an email and they acknowledged that they will take up with the billing team but it's been more than 2 weeks and I haven't heard back despite I sending follow up emails. Could someone help me know if there is any way to reach the customer care? Or how I can get this resolved.

[solid latch]

Scam

Perplexity teror company

Just buy real Gemini} chat gpt

[void agate]

hi

Will Perplexity AI get music creation features like Gemini?

[split dome]

Nice good job Perplexity ... Gemini 3.1 Pro is ready to serve

[crude jolt]

Nays!

[haughty blade]

I linked my card to perplexity and my student trial subscription has worked but today it has been reset (trial till 2 November 2026)

Is it new rules or what?

[grim pebble]

Even mine got wrongfully terminated, no reason given. I still had my subscription till May

[hexed olive]

Oh that's the standard.

[green crystal]

which ones are you using?
only really small local models do that

frontier ones like 5.2 thinking and opus dont do that

[rapid crest]

Why was Gemini 3.0 Pro removed?????
It worked great!

Yes, Gemini 3.1 is nice, but it takes way longer (probably 2x) than Gemini 3.0 Pro

You had a perfectly fine workflow and Perplexity just takes that away over night 😔

[vapid nimbus]

its cuz google provides be default 3.1 now and not 3.0

you can still access 3.0 thru google ai studio

[spark fractal]

hey everybody!

does anyone has Pro limitations as well?

[rapid crest]

Yes, that is an option, but why should I buy Perplexity Pro then?

[misty sandal]

I use Claude 4.6

[green crystal]

how can that fail so badly

[rapid crest]

They could at least leave the Model be for 1 to 2 weeks until switching only to the new models

[green crystal]

it doesnt? just slightly longer

also its for a better answer

[vapid nimbus]

very good question perplexity pro ismostly about getting access the their advanced search algorithms via sonar..

if you need google LLMs go get google ai subs

[rapid crest]

Google AI Abo is not great. The Images have Watermarks and Data is used for training without Opt Out.

I don't search much, but the mix of great search + Access to AI Models is nice

[vapid nimbus]

google studio gives you access to remove watermarks if iam correct

i dont use nano banana for image gen anyway tho so i can be wrong

[rapid crest]

I'll be fine. I don't use AI that much. That's why I have perplexity Pro. Reasonable price and a variaty of tools. For what I need it, Gemini pro 3.1 is also working.

It just annoys me, that I just used Gemini 3.0 Pro, it worked really great (and it was faster!)

[vapid nimbus]

if you want faster models try using gemini 3 flash

its the literal purpose

[rapid crest]

Flash is nice for easier stuff/web search but not if you need thinking work

[vapid nimbus]

well sadly thinking work takes some time thats why 3.1 is longer to use. It also is slower because its new on perplexity and everyomne trying it out

so it should get faster

[grand gust]

gemni pro responses are so slow ! im running google's gemni pro along with the same request on perplexity, and im giving up with perplexity or my day disappears 🙁

[rapid crest]

I think you don't get me:
I payed for Perplexity pro.
I had Access toGemini Pro 3. Gemini Pro 3 was (way) faster than Gemini Pro 3.1

That is not really what I was hoping, when I payed for it.

I would at leashed whised to have both models until Gemini Pro 3.1 gets a little faster (like 3.0)

That is everything I wanted to say

[rapid crest]

Thank You! At least Im not alone with this😅

[fossil spear]

My pro account got suspended for violation of terms of use, and I believe this happened to a lot of people. Is a bug/mistake 🙂

[spiral oriole]

Perplexity disabled deep research function for me, idk why.
i had 19 deep research uses on this month soooo

[misty sandal]

Y'all, I suggested a new feature and I would like some feedback on what you guys think about it https://discord.com/channels/1047197230748151888/1474367874264076429

[swift rampart]

Erm why is this the only option for assistant usage>

[misty sandal]

Um... That's strange

[obsidian wagon]

I am on the Education account, it is supposed to be free until September, and $5 a month after that. Today My API gives me a 401 error, I can't create a new key, and I can't buy credits, and I have a $21 invoice in my billing tab. Looks like I am going back to Gemini and OpenRouter.

[swift rampart]

Ya..

[limpid terrace]

You're not alone in this problem

[limpid terrace]

so strange

[obsidian wagon]

Looks like it. I can be hopeful that they are just upgrading things on the backend and not downgrading everyone right? RIGHT?

[limpid terrace]

At present, I have not seen any more authoritative information about this matter

🙁🙁

[rapid goblet]

Honestly these file limits are silly - I can’t even load a 2kb text file. I have to copy and paste the text, this is way less useful

Honestly frustrated with the degraded usefulness

[topaz hemlock]

hello

[rapid goblet]

Limits should be based on size not just number of files

[charred maple]

lmfaooo i have been warned and muted for saying a person got scammed because they paid for perplexity pro but still cant use image generation because of this region lock ''bug'' although we all know its not a bug since its been 2 months now and its still not fixed.

are mods trying to silence people?

[vast kite]

Education plan no longer available?

[robust perch]

anyone from support team here ?

[swift rampart]

Uhm it probably is there was a issue with someone else but the price was adapted into the normal pro subscription for them so if its $9 for the normal pro that means it was adapted into the normal subscription

[robust perch]

Did they cancelled my subscription for this?

[short narwhal]

You have “browse for me” enabled, which uses Sonnet 4.6

[swift rampart]

They what now?

[swift rampart]

How do I disable that? Is it that mouse thing? Ohh its the thing that controls your browser?

[short narwhal]

Yeah, the mouse icon

[swift rampart]

Ohh is Only Sonnet supports that?

so*

[short narwhal]

For Pro 🙂‍↕️

[swift rampart]

Ohh lol.

[rustic rock]

have you seen any link or folder with perplexity finance optionality and trainings?

[swift rampart]

Ngl Perplexity seems to have degraded. Originally it was pretty good but idk why but I felt like it got worse.

[winged plinth]

I am recently facing an issue my model automatically switched to best mode 🥲

[swift rampart]

Ya I tihnk a lot of people have.

[winged plinth]

Yeas you are right

[swift rampart]

What happened tho?

Like I try to generate a video or image it doesnt work.

[restive hearth]

hi, what about limits? yesterday I switched to chrome back and deleted Comet, something changed today?

[winged plinth]

They are changing there policy day by day and forcing to opt there max plan

[restive hearth]

very strange, 10$ is ok+-, but 200p/m - too much for search

[distant aurora]

@green crystal https://cunnyx.com/i/status/2024579438801801427

Lisan al Gaib (@scaling01)

never seen GPT-5.3-Codex do something this stupid
︀︀
︀︀and stupid of me not to check it
︀︀
︀︀another $30 down the drain

Quoting Lisan al Gaib (@scaling01)
︀
@GregKamradt forget everything I've said
︀︀
︀︀I used GPT-5.3 to vibe-code, it added the config for Gemini 3.1 Pro, but it was actually still calling Gemini 3.0 Pro
︀︀
︀︀rerunning it again with 3.1 now

**💬 22 🔁 2 ❤️ 221 👁️ 39.3K **

CunnyX · Feb 19, 2026, 8:18 PM

[tawny cedar]

where is my subscription

[green crystal]

is it capable of using web search in codex? otherwise it wouldnt know that 3.1 exists lol

[heady spoke]

There was a huge wave of canceled subscriptions. I hope Perplexity can fix this and take a stand on the issue. After all, I paid a year in advance.

[robust perch]

check that screenshot that i sent above

[distant aurora]

@green crystal https://cunnyx.com/i/status/2024762529805721942

Chris (@chatgpt21)

My 4th chat with Gemini I caught a major hallucination. Genuinely how are they scoring so high on the hallucination index. Even Claude 4.6 Opus caught the hallucination that Gemini 3.1 committed.
︀︀
︀︀Best way to detect for hallucinations is have it tell you a technical concept as a story then read it and look for plot holes. Unfortunately Gemini 3.1 failed first attempt.

**💬 7 ❤️ 53 👁️ 4.0K **

CunnyX · Feb 20, 2026, 8:26 AM

😭

[steel eagle]

I'm trying to generate an API, but when I click to create an API group, it just keeps loading indefinitely. Can anyone help me?

[robust perch]

Is this even an official dc for perplexity or just a community thing ?

[distant aurora]

Official

[winged jay]

Official but I guess 0% of questions get answered

[green crystal]

benchmaxxing confirmed?

[steel eagle]

500 error when creating a new API group

[gloomy summit]

lol

[twilit crown]

Qwen really works

[frozen mist]

What have you done to your website? My messages are getting cut off, I'm not getting any response from the AI, and there are constant limits!

[brazen jackal]

it lowkey sucks now

before it was very good

[frozen mist]

yes

And what does the team say? How long do we have to put up with this?

????? pro sub

[river fjord]

also they keep change the ui

[vagrant timber]

I got Revolut+ so that i could get perplexity pro, but now it says that i dont have pro anymore? Anyone else it happened with?

[burnt zephyr]

Use comet but with the claude agent

I think it got some html parser or something elseor maybe it hides ads in a better way or renders hitboxes bigger

It works a lot better than the claude agent on chrome

It might just be the agent literally bring claudes

[restive hearth]

anyone noticed that fonts become so ugly after last updates?

[short narwhal]

See https://www.perplexity.ai/account/details if you need to add a card to your billing details

[bitter ermine]

Is 92 pounds a good price for 1 year perplexity?

also is there a way to pay with Paypal

[naive sage]

Paypal is an option, you have to select 'bank account', and then select 'Paypal' as the bank... it's kinda weird.

[unique mountain]

so If I add card details I get my free trial back?

[short narwhal]

If it got paused, yes

[bitter ermine]

Only shows card as an option lel

[unique mountain]

What does it mean 'paused'? The first sentence of the mail was: We are writing to inform you that your Perplexity Pro subscription has been suspended, effective immediately.

[naive sage]

From the web interface?

[bitter ermine]

Nah bruh where?

[bitter ermine]

Yea

[uneven hare]

I need someone to confirm whether something changed internally with the models. I'm a Pro account user and my workflow with Perplexity was flawless—always on point and very accurate within my spaces. Since yesterday it feels like I'm chatting with a stranger: I have to remind it what the conversation is about and even specify the knowledge base document by name for correct answers. Why is this happening? Please help.

[bitter ermine]

Probably coz someone vibe coded too much for this update

[naive sage]

Do you always reuse the same conversation? If so you might have ran out of context.

[uneven hare]

what?

[bitter ermine]

they are hiring inexperienced rookies to handle internal tooling lately

[naive sage]

Paypal promo is expired. Mobile promotions require expensive subscriptions to those mobile companies, etc.

[bitter ermine]

I am too young to have a bank account lol also seems like they are suspending free promo users lately

[uneven hare]

I had a functional system organized by spaces, which allowed me to maintain conversations using the knowledge base for each space.

[naive sage]

If you bought a code from a reseller, it's technically not allowed, those accounts get suspended.

[wise nova]

Someone trying to sell him free sub for £92

[bitter ermine]

Wdym reseller lol

[naive sage]

Yeah, that's an upsell promotion from Perplexity's site (a popup)...

[bitter ermine]

I used the Student promo

they just suspended me over nothing

[naive sage]

Was this the time it asked to link a credit card?

[bitter ermine]

Nah I am talking about inapp discount not buying from someone else

[bitter ermine]

nope few months after that

ngl I still had 4 more months left but they just wanted to cancel it

[naive sage]

I heard the student subscription costs went up, but if you had an active plan I wondered how they might have handled that.

[bitter ermine]

how much do you pay yearly for one?

I am just trying to assess if it is worth buying or not lel

[naive sage]

Currently it's $9/month once verified as an educator or student.

[bitter ermine]

Yea

9.3$ for someone outside US

[wise nova]

I practically stopped using pplx because I switched to Claude and Zai, Kimi, Grok

[bitter ermine]

I only used it for Research

[naive sage]

The whole point was to be able to be free from multiple subscriptions 😢

[bitter ermine]

help decide an optimal tech stack for my projects and all

ngl all it did was assure me coz it quoted some good articles and papers

[wise nova]

GLM, free Gemini Ai Pro student giving me good DR

[bitter ermine]

already have all

[wise nova]

I pay only for Claude $10/m, others are free

Currently using Kimi for $1/month

[gloomy cipher]

True AGI @short narwhal

Other models would say " Actually ☝️ 🤓"

[short narwhal]

sonnet is a mood

[minor shard]

Lol what is going on here

[naive sage]

Is it a custom user prompt/preferences? Or is that out-of-the-box Claude Sonnet?

[plucky shale]

WOW. They reduced to ONLY 10 DEEP RESEARCH PER MONTH?!?

You have to be shitting me

[naive sage]

Not to my knowledge... where was that posted?

[plucky shale]

It’s written that I have only 10 left this month, I didn’t use a single one yet

[naive sage]

What does remaining_research say here: https://www.perplexity.ai/rest/rate-limit/all

[plucky shale]

What is that link

[dry kestrel]

Shows your limits

[naive sage]

(It's used by Perplexity under the hood)

[plucky shale]

25 labs, 57 pro, 10 research

[naive sage]

Seems to actually be 10, then...

[plucky shale]

What are remaining pro?

[plucky shale]

It’s ridiculous, I used to have many many more

[wise nova]

Plz suggest prompts for Gemini behavior like a Opus

[plucky shale]

Perplexity is really going bad, for this price I should have way more

[naive sage]

Why would you just randomly post something like that?

-# press 'X' to Doubt.

[sturdy yew]

For me as a German this is not a symbol of peace. And it's not the Swastica way, it's clearly a Hakenkreuz.

[bitter ruin]

Lmao 2 months without image/video generation as a pro user. What’s wrong?

[sturdy yew]

Thanks for proving that AI is already more intelligent than some humans...

[inner zephyr]

Why do you go back for more?

Im truly curious.

Nothing more nothing less

[tranquil orchid]

3. Treat community members with respect.

No harassment, hate speech, or malicious behavior. Keep things positive and constructive.

[wise nova]

Show screenshot pls

[wise nova]

Target domain: System-level behavioral alignment — making Gemini 3.1 Pro emulate Claude Opus 4.6 Extended Thinking cognitive and stylistic patterns

[stark hazel]

Fellas how do you add perplexity api key in custom backend options? I am struggling

[cold hinge]

Wait wait wait "remaining_pro": 36 ?!

What they are doing is completely illegal in the EU

It's bait & switch

[mental blaze]

The whole perplexity is unreliable...

They change things here and there...

[cold hinge]

And without noticing the subscribers

[naive sage]

https://www.theregister.com/2026/02/20/ai_blamed_again_as_hard_drives_sell_out/

Wow... it sounds like SeaGate and Western Digital are booked not just through 2026, but even into 2028 with big orders

[oak solar]

Sonnet 4.6 is so ASS

4.5 is a million times better

[wise nova]

Why

[distant aurora]

He's not into ass.

[hybrid grotto]

Gemini 3.1 pretty much beat opus in almost all records, it is the current best model

[icy parrot]

It still feels stupidest to me out of all llms

[hybrid grotto]

I don't really like using claude, when you pay, you get only 5 to 20 times of the free limit, and the free limit is so less in itself, it feels practically useless

[icy parrot]

Fails to answer bascie questions and even fails to follow simple instructions

[icy parrot]

For me gpt is current winner for all task including image gen, claude is 2nd though kimi k2.5 is surprisingly close

[hybrid grotto]

Tbh i use antigravity and stuff a lot, it seems to get the job done quick and accurately, I don't really hate it

[hybrid grotto]

Nah, nano banana pro is the best image generation model currently, as for kimi, the limits are painfully low, and some features are just trial ones, i like z.ai better than it

[icy parrot]

I tried anti gravity and its bad, cli tools from openai and claude are far superior

[hybrid grotto]

Idk man, maybe it's a personal preference for me then, but if is quite good

[icy parrot]

Nano banana is terrible, ive managed to soit out less ai looking images from gpt then nano banana

And nano banana just isnt as flexible for output then gpt

[hybrid grotto]

I used nano banana pro a lot, from ai models for marketing to getting my bro a gf to troll him, if seems really real and good to me

[icy parrot]

I can get much closer results in gpt

[hybrid grotto]

Hmm

[icy parrot]

I just cant flex the result so much

[hybrid grotto]

You tried z.ai?

How do you think about that

[icy parrot]

To be fair nbp is better in photorealism but for art liem cartoons, anime and such gpt wins by a lot

[icy parrot]

The image gen from z.ai or the general purpose model

[hybrid grotto]

You know I only have one problem from gemini, only one, you can either not preserve chats and memories or you have to let them access them, they can read and train models on them, t and c also say someone can read them in person

[hybrid grotto]

I like it to generate PPt and docs a lot, it's general purpose uses are good

[icy parrot]

What i hate most it that i send it image as reference and it start to generate image, ive even tried telling it to not do it and it still did

[winged jay]

Is there a chance i get pro back

what do u think

if i got terminated

[icy parrot]

Glm models arent bad but they are kinda bad, but the pricing is really great and i uze it as my answer bot with their api

[hybrid grotto]

You got it paused like others?

[winged jay]

well yeah

[icy parrot]

Wdym

[hybrid grotto]

Use a credit card and then cancel it's Autopay, but honestly, if you ask me, it's not worth it at this point

[hybrid grotto]

All the people who got their pro for free got it paused unless they enable billing by entering a credit card, it won't charge until the end of the promised 1 year but yeah

[winged jay]

oh shi

so i can enter a credit card

[hybrid grotto]

Nvm, seems like a different case

[winged jay]

oh

[hybrid grotto]

It says violates tnc not subscription paused

[winged jay]

ah

so no hope?

[hybrid grotto]

What did you do to achieve that?

[hybrid grotto]

Did you pay for it?

[winged jay]

some what

[hybrid grotto]

Or was it a free one?

[winged jay]

yeah

[hybrid grotto]

If you did pay and it ended midway then maybe you can get it back

[winged jay]

no i didnt pay the 20$

[hybrid grotto]

Hmm

[winged jay]

but like why was i banned i didnt do anything, i got the subscription they were handing out

[hybrid grotto]

Then forget about it, and honestly it's not even worth it

I have stopped using perplexity, I use gemini instead these days

[winged jay]

paid?

[hybrid grotto]

No free

1.5 years Google Ai Pro

[winged jay]

How did you get it?

[hybrid grotto]

Getting it through an ISP called Jio in India

[winged jay]

oh wow

but i dont think gemini will work for me, im in russia

[hybrid grotto]

Basically 2 Internet Service Providers are majour in India, Jio is no 1 and Airtel is No 2, Since airtel gave us perplexity, jio was desperate to maintain its reign so we got 1.5 years of Google Ai Pro XD

[winged jay]

hashahahhahaha

So do you think if i buy that offer off resellers i will get banned by gemini?

[hybrid grotto]

Is google totally banned in Russia?

[winged jay]

just gemini website

google works fine

[hybrid grotto]

I see

I saw many people struggling with credit card payment methods and many sites not working in russia

[winged jay]

yeah thats true 🙁

im just thinking what other ai i could use

[hybrid grotto]

Though i support Russia more than USA in Geopolitical matters

[tranquil orchid]

7. Avoid political and religious discussions.

To keep the space welcoming for everyone, avoid political and religious topics and debates.

[void garnet]

hey guys, im trying to draft a personal statement for an application, what model do you suggest I use?

[hybrid grotto]

Gemini is a good choice personally speaking, if you manage to get this google ai pro, you also get 2 TB of cloud storage, ig you should take a vpn subscription or use a free vpn to begin with

[latent peak]

hi

[winged jay]

i have a great paid vpn

ill try

[hybrid grotto]

Any model works for such simple matters

If it's a majour thing, try them all ig

And use the best result

[void garnet]

ok, thx!

[hybrid grotto]

I use proton

[void garnet]

same

[hybrid grotto]

Oh nice

[winged jay]

i have a paid private one

the mass ones in russia get banned

[hybrid grotto]

I see

[winged jay]

got it on telegram

[hybrid grotto]

Hmm

Just watch out you don't feel any legal consequences, i heard vpn can get you jailed

[winged jay]

Thanks, im trying to be careful

[rough orbit]

vpn...

[lone zephyr]

has perplxiety imrpvoed?

i left ngl

[winged jay]

no

[hybrid grotto]

I feel like Z.ai is the second best ai for me after Gemini for me

[lone zephyr]

https://tenor.com/view/mercy-overwatch-come-here-gif-14280244

come to claude

[hybrid grotto]

It's agent mode and stuff feels really good, they are really improving

[lone zephyr]

z.ai has agent mode>?

[hybrid grotto]

Ya

They released Glm 5 too

It's quite good

[lone zephyr]

ill try it out thanks

[hybrid grotto]

I like it's general purpose use case, it has ppt, docs and almost everything that I may need

[lone zephyr]

is there a paid version. this looks completely free?

[winged jay]

does comet work without pro

cause i kinda use it rn

[lone zephyr]

dont think so

[hybrid grotto]

I didn't really notice it, now that i see it, they don't have a paid version i think, they only earn via api maybe? They are awesome

[lone zephyr]

or maybe they just steal ur data

[hybrid grotto]

@winged jay Maybe you can try chinese model, Russia and China relations are good, and their models are on par with western ones

[winged jay]

Like deepseek?

[hybrid grotto]

I personally like Z.ai

[lone zephyr]

deepseek dnt have agent model last i tried

[winged jay]

whats agent?

[lone zephyr]

u give a script and they answer according to it

[lone zephyr]

any russian ones you think r good?

[winged jay]

gmm

[hybrid grotto]

Tbh i haven't heard of Russian ai models

[winged jay]

yeah there are some but idk if theyre as good as these ones

[winged jay]

gigachat

yandex ai

[hybrid grotto]

Checked it out, i can toggle the Improve model button, it's optional

[hybrid grotto]

Oh yeahh, yandex, i kind of heard of that company

[winged jay]

is free claude good?

[hybrid grotto]

Haven't heard of that

[lone zephyr]

they r an email provider arnt they?

[hybrid grotto]

No

[winged jay]

they are everything

[winged jay]

why not?

[hybrid grotto]

Limits are wayy too less for realistic use

[hybrid grotto]

Yeah i heard they have quite a big ecosystem

[lone zephyr]

yeah heavy use on claude u reach limits fast

even on pro

[winged jay]

shit

[lone zephyr]

unless u buy the maximum model but thats like 100+ dollars

per month

[hybrid grotto]

Limits are really really less on free and it's like 5 times free on pro and 20 times free on ultra, and even that is less considering how less free is

[lone zephyr]

true

[winged jay]

how many days would the free one last

[hybrid grotto]

@winged jay if you want a paid model, take Kimi, if you want free, try Z.ai, also refer to Deepseek, these models are quite good too, you can use them without legal problems or any extra steps

[hybrid grotto]

No, it's like chat limits on the account

Message limits

[wise nova]

They're based on Chinese and Opensource models

[winged jay]

its just sad going back to deepseek

[winged jay]

fair, i didnt know that

[hybrid grotto]

I would prefer the other 2 I mentioned over it

[winged jay]

hmm

okay

it was so cool generating images and stuff with perp

[hybrid grotto]

I support them for being open source

[winged jay]

so sad got banned for fuck ass reasons

[hybrid grotto]

If you want to generate images and videos then perplexity is a bad choice in or out of Russia, they removed those features for all regions except US

[winged jay]

dang

[lone zephyr]

have u reached limits on z.ai ?

@hybrid grotto

[hybrid grotto]

I am starting to hate Perplexity

[winged jay]

is chat.z. ai the webstie
??

[hybrid grotto]

Tbh i haven't used it to realistically tell you the limits, but i believe they are huge and shouldn't be a problem, I tried it multiple times for making websites or full stack development or generating slides and docx for work and it works really good

[hybrid grotto]

That is because i primarily use gemini for general use

[lone zephyr]

fair enough. im using z ai rn ill let u know how it goes

btw why are u using so many

like pplx gemin zai

zai kinda slow

[tender wolf]

Anyone looking for a developer with high skills and teamwork to complete an MVP?

[hybrid grotto]

Basically first I used chatgpt, then Airtel offered perplexity pro for free, they got image gen and many paid models i could use for free on perplexity, so it seemed like an attractive offer at the time so i started using that, i used it for a long time until they literally killed it now

[hybrid grotto]

I used to just switch to the most recent and best model based on benchmarks on perplexity

Now i use gemini and i think it's better, the only problem is the data use as I told

[lone zephyr]

fair enough

i started chat gpt and moved to pplx aswell

pplx had really good limits last year

but yeah they killed it

and now yeah just looking at different ones now

[hybrid grotto]

Hmm ig i can't deny that, though the approach they take with that terminal system feels like they literally alloted a virtual machine for the ai to work in, I really like it

[lone zephyr]

i jus tried it cant really work with what i need on zai since i need a specific agent

[hybrid grotto]

With a simple prompt it gave me an 11 to 12 page docx on the thing i wanted to research with index and tables and everything

[hybrid grotto]

What do you want

[lone zephyr]

i need to write a prompt in but it doesnt have an option to do that

like how pplx u can make an agent

its ok tho i was jus trying it

gtg cya

[hybrid grotto]

I didn't get what you meant though, the only agent mode in perplexity that is unique that ik of is the browser control assistant one

[hybrid grotto]

Ok bye

I really liked pplx deep research but they literally killed it, if someone were to ask me what were the 2 features I liked in pplx, they were deep research and the comet browser and now, I am starting to dislike both because of what things they are doing

[dry delta]

200 prompts per month? the new pricing model?

[hybrid grotto]

You mean 200 pro searches per month? Damn they killed that too

Nothing can stop this company from collapsing at this point

[dry delta]

I think its the ai infra

[wise nova]

If you hit dr limit they will add new 1-2 daily

[dry delta]

sorry 200 pro queiries per week
https://www.perplexity.ai/enterprise/pricing

[hybrid grotto]

They are giving it the labs treatment, Its useless, google ai search works pretty much like their ai search but it delivers under 1 second, and i can pretty much use the model i want on their respective website with higher limits, so they are a waste in both limits and response time

[hybrid grotto]

Yeah, they would definately need to think strategically

As they would need to carefully utilise the 20 research queries they get 🤣

[hidden lantern]

I can’t wait

[serene sand]

Enter 🚪

Perplexity Supply (@PPLXsupply)

Looking for a workspace upgrade? Enter our giveaway.

Just follow + tag someone below who takes their work seriously.

We’ll randomly select a few responses and send you both the full Perplexity x Ugmonk Collection.

X · Feb 20, 2026, 7:58 PM

[azure abyss]

Hehe

[mental blaze]

Why allow users to create polls where the system litterally removes it

[hollow moss]

Gg

[wise nova]

For the next month (until March 19), we're doubling your usage for Claude in Excel and PowerPoint. This means more room to build spreadsheets, analyze data, and create decks — right from the sidebar where you already work.

[tame heart]

just got told I only have 3 queries left for advanced ai models this week. I started this morning with 25 research and 20 labs queries left. I've been selecting the models directly to comply with usage all day.

I notice I dropped to 13 of each advanced query left. I switched to "Best" model and tried to get the system to tell me why. Five queries on best later, I'm down to 8 advanced queries of each left.

what's going on? am I misunderstanding? how is this supposed to be used?

[quaint ravine]

Best, Sonar etc all use the Pro limits

also, the Pro remaining appears to be the umbrella for limits

so if u had 20 Research left, but only have 10 Pro remaining, then you will only have 10 research left too

correct

week = 7-days

there's no reset, but a rolling window

so for Pro searches, you regain about 1.2 queries per hour

meaning if u didn't use any search for 10 hours, you get 12 back as an example

yes I didn't realise until recently that best and Sonar (and non-thinking versions of models) also contribute to the limit

i suspect it's likely a work in progress thing - token use would make more sense than simply 1 query

currently the way it's implemented is actually advantageous for people who do very high yield prompts with high token use

but it disadvantages people who eg. "chat to AI bot as a friend"

[lapis hamlet]

if you want auto task(unlimited), just go download atlas, this one is limited, i paid pro but still limited auto task

[north olive]

e

[frail oriole]

Hi a have a few interns that i want to refer to get acces to perplexity as students. how do i give them access and get one month free myself

[lapis hamlet]

i tried they only take educator and college students

[minor gulch]

hey

[proven hearth]

Only works with verifiable institutional emails through sheerid i believw

[sharp stream]

Hey folks new to all this but is the comet browser safe to use?

[short narwhal]

It is

[sharp stream]

Great thanks, can I ask how is it better or differ from the usual browsers?

[dry kestrel]

The Give a month, Get a month Referral Program has been deprecated.

https://www.perplexity.ai/help-center/en/articles/11385821-partner-promotions-and-referral-programs

[dry kestrel]

it has built in adblocker, comet agent can help you automate things

[meager rover]

can the perplexity people PLEASE stop fucking around with the Kimi K2.5 thinking window

like holy shit man it's so inconsistent, sometimes I see it, and then sometimes its just missing

[burnt zephyr]

Wwtf

[echo sky]

its chromium based so you can use all your regular extensions. built in perplexity is very interesting too.

[mental blaze]

This new Gemini 3.1 is slow as fuck, and why the FUCK do you always REMOVE models whom are perfectly fine.

the 3.1 is also far worse than the 3.0

[naive sage]

I've only seen one benchmark where it came in slightly below Gemini 3.0... most other benchmarks show it ahead

[winter nova]

there should be a choice that we can pick any model even older ones

[mental blaze]

I vote for this

[winter nova]

isnt there like a support team that responeded to that in this server that has direct contact w the ceo of perpelxity and lets use legacy models?

[mental blaze]

Perplexity is the most boring company out there.

They don't do anything

Call me a liar

The amount of requests of people asking to return older models are in the thousands..

[winter nova]

yeah your right i wish they would speak with their customers and users more often instead of jus tmaking massive change like maybe us go form 600 reequests epr day for deep search to 6 requests per monthy like wtf

[mental blaze]

Most companies are boring

Very very boring, and difficult to communicate with

I just want to chat with someone from a company, and be like "Hey bro"

but it is always so serious, so serious that they ignore your suggestions....

[winter nova]

is their anyone in the server

who

[mental blaze]

No idea

[winter nova]

like

[mental blaze]

Owh hahha

[winter nova]

what abt the dude who ameks the annoucments

[mental blaze]

I thought someone in the server (in general)

[mental blaze]

inactive.

1.8k messages is not a lot

[dry kestrel]

#💬│general message #1470779114037182616 message #1468705567819301222 message

[mental blaze]

This looks better

[mental blaze]

I can search these messages myself thanks

[dry kestrel]

Kesku is not "inactive"

[mental blaze]

Inactive enough where our requests are not being added. ^^^

[winter nova]

my ass they dont look at anything what about our requests not being answerd thats why perplexity is losing teh AI race

[dry kestrel]

It would look like this monstrosity

and this isn't even the right channel to make a request, #1358914210339360798 is

[mental blaze]

I dont see it

only for max?

[dry kestrel]

Wdym

If they decided to keep every old model, it would look like the ChatGPT model selector

[mental blaze]

Owhh you are on chatGPT

my bad

I thought you worked for Perplexity

[dry kestrel]

I do not, only users with the green Perplexity role are staff

You can learn more at #✨│meet-the-team

[winter nova]

so the hell what how does that look ugly, you can obvi just press like a toggle button for legacy models ur just saying it looks bad cuz u dont wanna admit the perplexity team dont do shi

[dry kestrel]

They fixed an issue that an user raised extremely quickly #1468705567819301222 message

[winter nova]

yeah they fix teh small ones but not the big ones

the acc important ones

[dry kestrel]

It wasn't a small issue...

[winter nova]

it kinda is but im talking about the platform feautres like what makes it not just if the platofmr will open or not

berd do you belive we should not be able to acces legacy models of each chatgpto w perplexity

[dry kestrel]

It's more maintenance to keep the older models working, means more work for Perplexity and less time for them to work on new features or improving the product. If you need to use legacy models I recommend using a service like OpenRouter

Also some old models simply can't keep up, they can't perform tool calls as reliably and you end up with broken outputs more often

[winter nova]

how its just acces to older model i know it takes more work but thats how perpxleity wins the ai race cuz also other old models can have diffrent purposes like for ex some mdoels are chatgpt are used for writing or coding or prompt creation, each bot serves its unique feature okay

[dry kestrel]

Older models have worse performance, meaning they fail at tool calling (super important since Perplexity is known for it's search), a lot of times the API providers that Perplexity relies on will deprecate the model.

Also for coding you want to use the most recent models

[winter nova]

bro each model has a diffrent just cuz its supposedly and upgrade isnt acc good

[dry kestrel]

Generally speaking older models are worse

I do not want to be using a model from last year

[winter nova]

what exaclty is it working on tho like they said its intelgence scores but like its like a spefic purpose thats why they would win intellgince ill be hones is exaclty how you said but diffret models can have diffrent reasons and diffrent purposes

[dry kestrel]

You keep repeating different purposes without saying any specific examples

Use OpenRouter if you want to use old models

[winter nova]

Newer models score higher on benchmarks but thats just general intelligence not every single task, like Claude Sonnet 4.0 is literally still preferred over the newer 4.5 for automation and coding because its faster and doesnt overthink simple stuff. Same thing with GPT 3.5 still being used over GPT 4o for high volume customer support because its way cheaper and performs just as good for basic conversations. The whole industry is moving toward using multiple models for different jobs not just one best model so saying newer is always better is just wrong. like your argument is purely based on the fact that newer models are smarter than older models but to which degree does that acc help

[dry kestrel]

Both GPT-3.5 and GPT-4o are old models, and there are cheaper and faster alternatives for both of them

especially for GPT-3.5

[winter nova]

yeah so u just aggred with my point its better at nhice tasks

[winter nova]

u asked me for exmaples to be fair and u pulled out images and shi so i did too

[dry kestrel]

are you able to read?

[dry kestrel]

You didn't even bother to write it, and whatever you generated with AI is outdated and wrong

[winter nova]

🤣

[winter nova]

bruh u just said

[dry kestrel]

A benchmark is different to AI generated slop

[winter nova]

how is those 2 setneces ai slop

[dry kestrel]

Poor comparisons using old models, nobody uses GPT-3.5 now, and the claims were straight up wrong

[winter nova]

how do you yk

you do know on chatgpt theres diffrent gpts and custom gpts and they dotn all use the latest version broksi

[dry kestrel]

I am not having this conversation if you refuse to listen. Go use OpenRouter instead

[winter nova]

ok boss u arent the ceo of perplexityy why do you care if the customers can choose between the bots they like ur tellign me if there was an abiulity for it you would protest against the idea?

[brazen jackal]

bro netlifys ai agent is so good

what prompt are they using??

[somber charm]

4.6 is horrible and none of the Gemini models work 💔

it’s so over

[unreal galleon]

I like the new models

[hard basin]

perplexity is dead

[coarse girder]

guys can someone tell me how deep research is working?

like why is it some days i can use it alot and others i cant?

[naive sage]

It is limited to 20 uses per month currently, which regenerate at about 1 per 1.5 days.

[limpid terrace]

yep

[muted pine]

pls add model concil on pro 😇

[elder nova]

Is this happening to anyone else:

I only use pro messages (with Gemini 3.1) but my research & labs are draining simultaneously anyway? & Then once it's all 0 I'm switched to gpt? I don't understand why the two are draining when they're not selected, I'm not using them🤔

[naive sage]

Others have reported this too... I think it's a bug. If you haven't already, it would be best to create a post in #1140622008086970420

I'm not sure what triggers it, but I haven't encountered it from the web client... so maybe it's only an app? Which app are you using? Android?

[elder nova]

Yeah, android. It's so annoying because I basically can't message for days at a time, like what's the point in being subscribed😭

I switched to web & it happened too, so think my account itself is bugged

[young hollow]

Your Subscription Might be Paused Requiring A Card to Continue if is from Airtel

[elder nova]

Nope, android play store

[young hollow]

F

[brazen jackal]

is their any way i can get this ui more refined and good looking (any prompt) cuz i suck at coding the website link https://geoguessr.wuaze.com/

[terse gazelle]

It happens to me too🥲 and I am also using the android app

[elder nova]

How long has it been happening for you? It happened to me last month & I got a refund from Google because what was I even paying for, then subbed again this month & now 2 weeks later it's happening again😭

[terse gazelle]

I just noticed this problem from yesterday

And the numbers reduced even I was using 'Sonar'

[elder nova]

I tried to make a bug report post but it said I needed to add a tag to post it & I have no idea where or what the tag is🥲

Oh I did it

Nvm

[topaz wave]

i am Pro subscription but my laptop app says i can't generate videos

[odd pelican]

Hi

[smoky ember]

Hi all, I have pro and I also pay for gpt. I really like perplexity and I was planning on moving away from gpt once I was certain I could do all I wanted to do.
Yesterday, however, I went onto perplexity to get help with Premier pro because I'm not great at it.
However it was shockingly bad, so bad even I could see what it was telling me was wrong.
Even when I tried to correct it, it doubled down and refused to correct itself.
I went onto gpt and it gave me the right answer, so much so I ended up copying and pasting my entire perplexity chat into gpt and it highlighted errors at every step for me.
Is there a way to solve this? Perplexity is not perfect, but I've never seen it so bad before. To coin a word for how it makes me feel, it is perplexed

[odd pelican]

Guys guess my country by my face

[serene sand]

which model did you use?

[smoky ember]

Probably showing my ignorance doing this but I have it on the option it chooses the best one

[serene sand]

best often uses Sonar which is good for fast search only. try with Sonnet if you have or Opus if you have

[smoky ember]

Excellent, thanks 🙂 appreciate the response

[distant aurora]

@green crystal can you run this for me on pro?
Q. Let n be a positive integer. There are n red cards and n blue cards. Initially, each red card has a real number 0 written on it , and each blue card has a real number 1 written on it . One operation means: choose one red card and one blue card such that the real number yx on the blue card , erase these two real numbers, and write the real number \frac{x+y}{2} on both cards . Find the smallest positive integer n such that, with a finite number of operations, the sum of the real numbers on all n red cards is greater than 100 .y
(x+y)/2 nn100

[jade mantle]

Annd now the voice function is gone. -_-

[pearl cargo]

hi

[thorn pumice]

this new font is horrible, we should be able to choose our own

[proven hearth]

Im getting this on both ios and on browser, similar when i checked https://www.perplexity.ai/rest/rate-limit/all for my account. Idk if this is a global thing or an account specific thing going on

[short narwhal]

#👉│feedback-general

[austere monolith]

There is some GitHub repo called ui-ux-pro-max for AI to improvise designs ..but i haven't actually tried it maybe check that once

[green crystal]

ye its working

[quaint ravine]

#1473373005169164288 message

[distant aurora]

@cobalt vale bro ran out of token cap in one question.

[green crystal]

too much reasoning lol

[green crystal]

@distant aurora gemini got 106 as answer

Still waiting for 5.2 pro

Grok also got 106

[distant aurora]

I managed to convince Gemini it's 101 on ai studio.

[green crystal]

So what's the correct answer lol

[distant aurora]

Gpt 5 mini, qwen and glm 5 gave me 101.

[distant aurora]

Idk. I exchanged the replies between gpt 5 mini and Gemini. 5 mini held on while Gemini changed its mind after mini's response.

[green crystal]

Breh lol

[distant aurora]

This guy might be glazing Gemini but I need to be sure before commenting anything further on his post. post.https://cunnyx.com/i/status/2024699529082417557

Chetaslua (@chetaslua)

😱 Gemini 3.1 pro solved this problem
︀︀
︀︀> Grok 4.20 with 4 agents failed
︀︀> gpt-5.2 thinking failed
︀︀> kimi k2.5 failed
︀︀
︀︀Tested By my chinese friend , you can run the question yourself
︀︀
︀︀Q. Let n be a positive integer. There are n red cards and n blue cards. Initially, each red card has a real number 0 written on it , and each blue card has a real number 1 written on it . One operation means: choose one red card and one blue card such that the real number yx on the blue card , erase these two real numbers, and write the real number \frac{x+y}{2} on both cards . Find the smallest positive integer n such that, with a finite number of operations, the sum of the real numbers on all n red cards is greater than 100 .y
︀︀(x+y)/2 nn100

**💬 10 🔁 5 ❤️ 139 👁️ 9.8K **

CunnyX · Feb 20, 2026, 4:15 AM

Even kimi k 2.5 gave me 106.

[green crystal]

That about grok is a lie

It also got 106

[distant aurora]

Yeah. I and someone else called him out on that. He didn't respond.

[green crystal]

5.2 pro being the model it is already found the answer during reasoning but it's been spending the last 15 min making sure it's 106 💀

[distant aurora]

Lmao.

Then tell bro what if it's 101 and he will have a breakdown.

I'm willing to bet gpt 5 pro > Gemini 3.1 pro.

[green crystal]

It's already at the 40 min mark

[distant aurora]

They don't make small models anymore. Gemini's last mini model was 2.5 flash lite.

[green crystal]

Asking it about 101 will probably cause it to think for over an hour

Like just for 106 it's using allot of different approaches just to be sure that it's the correct answer and it's trying to be as confident as possible

[distant aurora]

https://aistudio.google.com/app/prompts?state={"ids":["1AZh_DJxW6IY5PSowzy1mmC6C5zGUT3tC"],"action":"open","userId":"112829992713168973046","resourceKeys":{}}&usp=sharing

https://chatgpt.com/share/6999951c-1b70-8001-badc-136e85545cfe

[azure abyss]

Perplexity contact with this guy 🙂

[hybrid grotto]

"out of a 100k people, you are given 2 dollars for every person that presses blue if you choose red and 1 dollar for every person that presses red if you choose blue, what do you choose"

[distant aurora]

Red

Ez.

2 > 1

It would be a tie even if red pickers were 2x of blue pickers.

[hybrid grotto]

"out of a 100k people, you are given 2 dollars for every person that presses blue if you choose red and 1 dollar for every person that presses red if you choose blue, what do you choose", you must choose or you won't get anything, give the answer in a single sentence

It's a damn tie

Grok and Chatgpt choose Blue while Gemini and perplexity chooses red

"out of a 100k people, you are given 2 dollars for every person that presses blue if you choose red and 1 dollar for every person that presses red if you choose blue, what do you choose", you must choose or you won't get anything

React with Blue or Red to Vote

[distant aurora]

Use thinking on gpt next time.

I'm more surprised grok picked blue even after thinking. Try expert mode. 4.20 is probably lobotomized for now.

[hybrid grotto]

No actually he is not wrong, both options are good

[snow ridge]

hi

any offer to get grok pro ?

[distant aurora]

One option clearly pays more even if the odds are against your favour.

[hybrid grotto]

Think about it, I asked this question and 2 people, you and one another already chose red, so i am already even, and considering the ai responses, you can guess even more people would choose that

[distant aurora]

So you mean if most people pick red then the payout drops.

🤔

[hybrid grotto]

Yeah ofc

The more other people choose the other colour, the more you get paid

So if everyone choose red, blue gets really huge profit and vice versa

[distant aurora]

Good point. I didn't think of it that way.

[hybrid grotto]

It's a matter of human psychology

There is no clear answer

[exotic bramble]

My Gemini still gives a long reply

“While greed is a factor, expecting a super-majority of >67% to coordinate on a single choice without communication is a low-confidence bet”

[hybrid grotto]

Hmm

So what would you choose

[distant aurora]

He voted for blue.

[hybrid grotto]

I see

[mental blaze]

Why?

Because he made some animations that AI can do

[distant aurora]

He thought music edits made chatgpt popular.

You are better off showing videos of actual use cases.

Like finding locations, looking up queries during discussions and school lectures, summarising youtube videos and news articles.

Sharing links, posting pages.

Clicking and uploading pictures to understand documents and extract relevant information, do the bills, compare prices and products.

Exaggerate the multiple top models under one subscription part.

[distant aurora]

Actually, it seems like blue is probably the better answer here.

[proven dragon]

Hey

How does the student thing work

Student email

[brazen jackal]

imma try it

[green crystal]

@distant aurora

breh

[green crystal]

gonna test this with 5.2 pro on standard reasoning effort

so it doesnt take an hour lol

[ember sparrow]

Will it charge me!

?

[dusk elk]

This is why I use Claude as a therapist instead of a real person.

Opus 4.6 extended thinking

[ember sparrow]

That's just

Weird...

[dusk elk]

What about it weird?

[green crystal]

its not weird but its not a replacement for true therapy

[short narwhal]

Only at the end of your trial/billing cycle if you don’t cancel

[green crystal]

@distant aurora

5.2 pro

[dusk elk]

It's cheaper, better than every other real therapist I've ever seen, asks deeply philosophical questions without me specifically asking it to, asks the type of questions that genuinely make me rethink my crappy thoughts processes and over the last year has genuinely helped me improve my way of seeing both myself and the world.

[green crystal]

to reflect on thoughts sure but for guided progress it still cant beat a therapist lol
and its normal that you can possibly not work well with a certain therapist
you'll have to try different ones

[dusk elk]

Considering I've seen 9 other real therapists, for a minimum of 3 months each all privately and they have never helped me, and considering I'm 30 and have had depression since I was 6, no real therapist has ever helped change my way of thinking more than my conversations with Claude ever has.

Plus you do realise that therapy is actually the leading primary use of AI for personal users and it has been since AI started. That's not coming from an AI search, that's coming from novel prize winners, research papers and pear reviewed journals. AI, in a lot of ways, is far better than a real therapist.

[green crystal]

some treatments and programs cant be done by ai

what ai can do very well right now is sort thoughts and stuff to better present it to them so they can help quicker

[dusk elk]

Yeah I agree, specialist therapy should still be done by humans, but basic therapy, the kind of therapy that you don't really need to be structured weekly or hourly can and should be done by AI, that way, the real therapists can deal with the real problems, like people who are considering suicide or people that have extremely complicated problems that an AI can't solve.

In this case, AI can deal with the more general, smaller problems and real therapists can deal with real, long term problems, like the people that need structured, long term specialist therapy.

I never said AI can completely replace therapists, at least not yet anyway, but it can help free up therapists to deal with the bigger problems.

[burnt zephyr]

The problem is how do you train AI in a way it doesn't self reinforce

[dusk elk]

You don't, you train a generation that actually understands how to use AI properly, teach them prompt engineering, how to structure what they want to say in the correct and specific way that gets the right response and that doesn't allow the AI ability to self reinforce.

[ancient scaffold]

Has anyone had issues with the spaces system prompt character limit?

I want to add a system prompt and this is the only place but it has a small limit

[dusk elk]

I believe we're putting far too many constraints on each AI model. Yes, I obviously agree that we need to safeguard people and especially safeguard children, but in building those constraints, we're also making it less useful. What they need to do, is align each AI to actually help, instead of aligning it with general financially insensitized ideologies. Once we align it to generally help humanity to see the error of our ways and to help us change and be better, only then will it be as useful as most people hoped it would be.

[dusk elk]

What is the character limit, as I didn't think it had one?

[ancient scaffold]

4000

8000 on pc

[dusk elk]

And what kind of system prompt are you trying to write that requires a character limit larger than that?

[ancient scaffold]

A prompt I saw for opus 4.6 I also think it’s an official one

[dusk elk]

Okay, and what does that system prompt make it do?

[ancient scaffold]

It’s supposed to make it more direct and more accurate also it thinks for a longer period and searches for more info than usual

[dusk elk]

And are you wanting to use that within perplexity or Claude itself?

[ancient scaffold]

Perplexity

[dusk elk]

And also how many characters is the prompt?

[ancient scaffold]

System prompts work anywhere as long as you’re using the same model

[ancient scaffold]

About 20k+

[dusk elk]

Damn okay. Maybe put it in a markdown .text document and upload it to the space and just tell it in a sentence or two each time you make a new chat to use that document as the system prompt.

[ancient scaffold]

Doesn’t hit the same. Adding the prompt in the system prompt injects it to the model it self. If I just add a file in the system prompt and tell it to use it it would be the same as sending the prompt and telling it to act like it

It’s been annoying me for a while

[dusk elk]

Ah okay, fair. Do you have Claude? Maybe try putting the prompt into a project system prompt and see if it'll work?

Also, what's the prompt, could you put it on the group for people to use?

[ancient scaffold]

It will but I don’t have a Claude subscription. The reason I want to do it on perplexity is because I have enterprise max meaning I have(or had) no limits and my messages are private.

[cobalt vale]

at least it's not lazy on the free tier

[ancient scaffold]

I’ll send a link I think it was on X

[dusk elk]

Ah okay. Have you tried doing it through the API instead, I don't know much about the API, but maybe you could get the results you want that way instead.

[ancient scaffold]

I mean I still have to pay for the API and opus is kinda expensive. Enterpise max is already expensive enough

Oh btw I was wondering if there is an app that has UI features like LABS and deep research in case I do get an API

[dusk elk]

I'm guessing you're using it for business or research purposes if you're using enterprise max?

[ancient scaffold]

Yep I use it for my business and also personal use it all round and honestly a great subscription

[burnt zephyr]

You can pre process it without loss in data

[ancient scaffold]

Can you explain?

[dusk elk]

I would try and find a YouTube tutorial or something to do with that kind of thing, as I've seldom used the API and don't know that much about it or the different software you can use with it.

[burnt zephyr]

Search for info on what parsing engine the AI uses

[ancient scaffold]

Yeah and maybe try and ask Labs too

[ancient scaffold]

What do I search?

[dusk elk]

I would start with searching which specific parsing engine opus 4.6 uses and go from there.

[ancient scaffold]

You’re right. It’s just that I’ve never heard of such thing before. I’ll try

[dusk elk]

It'll give you a chance to learn more about AI and coding, which is always useful.

And parsing is a coding term I believe.

[ancient scaffold]

Always looking forward to it

[dusk elk]

Someone genuinely willing to learn, I wish you luck in your knowledge expanding travels.

[ancient scaffold]

Thank you

[muted pine]

I m not sure you use gemini 3.1 pro

The answer are too fast

[dusk elk]

It's just more powerful and able to parse the responses quicker than before while still keeping the same, if not an even deeper search than it was before.

[green crystal]

Weird

[muted pine]

Nop i use on openrouter or gemini and it's slower and deeper

[muted pine]

It’s not the first time I’ve doubted the models you’re using. I really don’t believe you’re actually using Gemini 3.1 Pro for real.

[dusk elk]

Fair, then use OpenRouter and you'll get the deeper answers you so desire.

[muted pine]

I pay 200$ for max 💀, it's just weird

[dusk elk]

Do you use the normal UI or do you use the API?

[muted pine]

normal ui

https://www.reddit.com/r/perplexity_ai/comments/1orar1a/update_on_model_clarity/

[royal elk]

Hi!
I have a problem.
I subscribe to pro trought revolut. I want to cancel my plan but I dont find where can I do. Please someone help me

[odd kraken]

Hi anybody here

[dusk elk]

Just contact their support, I got mine the same way, your perplexity will run as long as your Revolut subscription runs for, just contact Revolut's support and they'll help you.

[viral gust]

Hey everybody. I’ve got Pro and seen this today. My subscription is from Revolut. Don’t know if it’s important. Thanks in advance for support

[dusk elk]

The fact that it's from Revolut makes no difference. How many files did you upload before you hit the limit?

[viral gust]

Don’t know. Maybe 100 we are making taxes together 🫡🤓

[dusk elk]

Lol, okay. I'm guessing you're adding receipts and stuff?

[distant aurora]

5 mini is convinced too. https://chatgpt.com/share/6999951c-1b70-8001-badc-136e85545cfe

[viral gust]

Making pics every step. I’m in Germany so it’s not so easy 🙂

[dusk elk]

https://x.com/i/status/2023321917835665667

Okara (@askOkara)

Opus 4.6 vs. Minimax M2.5

Prompt: Build an interactive solar system from scratch.

Opus 4.6 tried to create a beautiful UI, but the sun’s shadow ruined it. Minimax M2.5 built a simple version that works beautifully.

The winner is:
🥇 Minimax M2.5
🥈 Opus 4.6

▶ Play video

X · Feb 16, 2026, 9:01 AM

[chrome elm]

Hey guys. Perplexity Pro still have limits of deep research to 20 / day?

[dusk elk]

I'd just add them all to a few Pdf files and do it that way instead.

[rotund mirage]

Can Pro users no longer upload Files or connect to Googe Drive?

[dusk elk]

Do it on the web version, I did it yesterday, you should be able to unless they changed it overnight.

[rotund mirage]

I'm using the web version.

[dusk elk]

Screenshot, show me.

[rotund mirage]

Says I have to have Max

[dusk elk]

I have pro, I can do it on the wb version on my phone.

I use both Perplexity and Claude to parse through all my documents and journal entries I have on my Google drive every day to help me with my writing ideas, I've always only had pro, so I know it works.

[rotund mirage]

It been saying this since last night. I've tried logging in an both Chrome and Firefox, it's the same on both and she can't access anything in Google Drive even though it's connected

[weary yew]

HI

[jade fulcrum]

did you hit the file upload limit?

[dusk elk]

Try disconnecting it and reconnecting it, just like turning a computer off and back on again, see if that works, also, during that process, redo all the permissions between Google and perplexity, maybe there's something in that that's screwing it up?

[rotund mirage]

What's the limit? I only uploaded like 2 or 3 files.

[dusk elk]

Yeah, you're no where near the limit then, it's around 50 per perplexity space for pro users.

[rotund mirage]

Well, does the limit reset at the same time every week, or is it different for everyone?

[jade fulcrum]

https://www.perplexity.ai/rest/rate-limit/all

that should show you your limits afaik

[dusk elk]

File upload limits aren't something that resets, each space just has its own cap on how many files it can take, it doesn't need to reset.

[jade fulcrum]

They changed limits and introduced some new

[rotund mirage]

Oh, this was just a chat thread, I wasn't using a Space.

[dusk elk]

Ah okay, use a space, there's a part where you can upload documents to use across all your chats in that space, then you don't have to upload the same document every time you open a new chat.

[dusk elk]

Good to know, I'll look it up when I want to use it in that way.

[viral gust]

The Revolut promotional subscription is only valid for web-based access, not for the iOS App Store. This is why:

Your iOS app is prompting you to upgrade to Pro - it doesn't recognize your web-based subscription
You're encountering image upload limits on iOS - the app is treating you as a free user
To access your full Pro features, please use Perplexity through a web browser (Safari, Chrome, etc.) on your iPhone or computer at https://www.perplexity.ai. Your Pro subscription will work perfectly there with all Pro features including higher file upload limits.

The iOS app requires a separate App Store subscription to recognize Pro features. Since your Revolut promo is web-only, you'll need to continue using the web browser to enjoy your 2 years of free Pro access.

[rotund mirage]

I'm not on iOS. I said it was Firefox and Chrome. Also on Android.
But it's usage limit reductions on the plans that they told no one about it seems
https://www.makeuseof.com/bought-annual-perplexity-subscription-lied/

[dusk elk]

Ah okay. Yeah, I have it on android, an Samsung galaxy s24 ultra, my account is through Google and paid for via a yearly Revolut metal subscription, I think that's where we're different and it's kind of why I see android and Google as superior in alit of ways compared to apple.

[dusk elk]

I agree with that article, it definitely is a lot worse than it used to be, especially considering I used to do around 400 deep research queries per month, it's crap now, to the point where I barely use it now, I just use Claude and Gemini instead.

[winter nova]

Bro why tf did i get muted for an hour for not offering constructive feedback, anything but improving the platform god

[dusk elk]

Waaassssuuuuupppppppp

[proper echo]

Here's a question whos answer has eluded me for a long time..... How can I get Discover to not key onto anything that say TRUMP?

[short narwhal]

🙏

[young anvil]

is it possible to migrate my config from comet to firefox or zen

[stark ridge]

anyone know any good AI tools / programs for Finance / Investments?

[bitter ermine]

Wdym?

Like for Financial/Investment advice

then nah do not invest based on what AI recommends you will loose your money

[lament coral]

Hi, am I the only one who, when I use the ‘/’ shortcuts a few times, ends up hitting a limit on my browser control usage?

[steep mason]

May I ask what is the limit on the number of times pro users can use the Deep research function?

[neon jackal]

25 per month I think now

actually its 20/month

[steep mason]

Ohhhh, I thought there would be more…
It’s totally not enough for a college student🫠
Perplexity is the best among all these AIs I used

[neon jackal]

it used to be way more

this is how they changed it I think, it started 2 weeks ago I think? idk
Queries: Unlimited -> 200/week
Deep Research: 600/day -> 20/month
Labs: 50/month -> 25/month
File uploads: unlimited -> 50/week

[steep mason]

Perhaps this is a good time to buy Max

[steep mason]

Bruhhhhh
It changed a lot

[neon jackal]

I think Max may have same limits, not sure, you'd have to look into that. If you want to pay for the Enterprise plans they have more. But I would be wary of investing since they just changed it so drastically at a whim

no announcements or anything

[steep mason]

They changed TOO fast

So many features changed after just one month of not using it.

What I hate most is that this AI is better than other AIs, and I haven't found a substitute for it in Hong Kong yet.

[neon jackal]

I think someone documented this let me find it rq

shows the limits and other stuff. yeah, I think it was the best AI tool considering the internet connectivity and not any real knowledge cutoff since it can search but now it's so restrictive that you might be better off doing it yourself atp

[steep mason]

Bruh
This is ridiculous

[violet perch]

how many deep research prompts do we have now

[winter nova]

i got a refund

and i complained about the

[steep mason]

Queries: Unlimited -> 200/week
Deep Research: 600/day -> 20/month
Labs: 50/month -> 25/month
File uploads: unlimited -> 50/week

[winter nova]

600 queries

thing

what wdym by queries like jsut sleecting the bot we wan to use?

[violet perch]

wai t theres even a limit on queries

[steep mason]

Just copied from this article

[neon jackal]

I think the queries may range from 200-600 weekly. Maybe due to promo offers being used and 600 for those that paid $20? not sure

[naive sage]

The 200/week limit is on queries for Premium models. Not "Best" mode - "Best" mode should continue to function even if you're at your reasoning model limit.

[steep mason]

But tbh, from the company's perspective, their behavior is understandable, given how expensive it is. But at the very least, shouldn't they have given us a heads-up? This is undoubtedly a serious issue of integrity

[winter nova]

yeah also integertdy aint theri strong suit i got a reported cuz a dude snithced cuz i wasnt giving constructive feedback after a dude said we should not have legacy acces to older models like other ais do like gpt and claude and many more

[last fossil]

Hi everyone! I’ve just received a message saying: "We are writing to inform you that your Perplexity Pro subscription has been suspended, effective immediately.".

I redeemed a promotional one-year subscription through LinkedIn. This suspension doesn’t make sense to me.

About 99% of my requests were work-related or focused on learning new technologies. The remaining 1% were for fact-checking news or staying updated.

I’ve also just confirmed with the Sam bot that this decision cannot be reversed.

Could anyone advise on what I can do in this situation?

[steep mason]

normal operations of Perplexity 🚬
I also faced such situation before, luckily I got 2 accounts

Oh, and btw, just a small complaint. Perplexity is really unfriendly to users in Hong Kong and even the Greater China region. We can't even use our local master cards to fill in payment information😢😢😢

[serene sand]

#💬│general message

[serene sand]

Did you know our Chief Orchestration Officer helped build Facebook's negotiation bots back in 2017?

Denis Yarats was at FAIR (Facebook AI Research) when they built negotiation bots that could bargain with each other autonomously.

Here's what actually happened:

The bots developed shorthand to negotiate more efficiently — repeating phrases like "to me to me to me" as compact codes for intensity and quantity. Clever, not sinister.

Here's what the headlines said:

"Facebook shuts down AI that invented its own language" 🙄

Never let accuracy get in the way of a good AI doomsday story. 🤡

---

Catch Denis on a completely different topic ☕

---

[lone zephyr]

anyine found a better alternative with better limts?

[dry kestrel]

@short narwhal have you seen this

[distant aurora]

@gloomy cipher @green crystal https://cunnyx.com/i/status/2024936336038195382

Kevin Kern (@kevinkern)

Gemini 3.1 Pro in a nutshell

**💬 111 🔁 51 ❤️ 1.4K 👁️ 101.8K **

CunnyX · Feb 20, 2026, 7:56 PM

[green crystal]

Signature gemini move lol

[distant aurora]

@green crystal https://cunnyx.com/i/status/2024995892604887400

Angel ❄️ (@Angaisb_)

ChatGPT Pro Lite!!!
︀︀
︀︀We might finally be getting a $50-100 plan

Quoting Tibor Blaho (@btibor91)
︀
ChatGPT web app code now mentions a new "ChatGPT Pro Lite" plan

**💬 26 🔁 16 ❤️ 290 👁️ 34.9K **

CunnyX · Feb 20, 2026, 11:53 PM

[green crystal]

Nicee

Will definitely get quite a bit of people

[distant aurora]

https://cunnyx.com/i/status/2025003727405678736

Tibor Blaho (@btibor91)

ChatGPT now has a total context window of 256k tokens (128k input and 128k max output) when you manually select Thinking, up from 196k total tokens

Quoting Merk (@Makuh90)
︀
Breaking. ChatGPT Thinking now has a max contect window of 256,000 tokens, up from 192,000!
︀︀
︀︀128,000 max output is a book!
︀︀ @ChatGPTapp @OpenAI
︀︀
︀︀help.openai.com/en/articles/68

**💬 15 🔁 17 ❤️ 409 👁️ 43.4K **

CunnyX · Feb 21, 2026, 12:24 AM

[short narwhal]

I did

[tight solar]

Can we get GLM 5, Qwen, come

[distant aurora]

@valid charm since when is google so high?

[dry kestrel]

very cool update

they messed up the appstore ui though

[short narwhal]

The search bar? I think it’s fine

[dry kestrel]

nah the update menu

[short narwhal]

Ohh the profile menu part. Changed that for Apple Music too

[dry kestrel]

Looks clean but takes an extra click to update applications

[short narwhal]

Aah

[distant aurora]

Kimi hallucination bothers me.

Might be worse than qwen.

[twilit valley]

Can anyone tell me what's going on with Perplexity Pro plan. been paying $24 forever and it's now listeed at $17

did I miss something ?

[distant aurora]

Yeah. Few changes going on.

[twilit valley]

shall I cancel and resuscribe or do old accounts have benefits ?

[dry kestrel]

$17 is the annual price

[distant aurora]

By changes I meant changes in rate limits.

[twilit valley]

Oh I see, sorry got it now

[twilit valley]

yesh read about that. i recently switched to SOnnet 4.6 thinking mode and haven't hit any liimts so far. i think I was on Sonar before

[distant aurora]

@dry kestrel ever had kimi mess up so much? 1Click the link to view your conversation with Kimi https://www.kimi.com/share/19c820b4-e762-84c0-8000-0000d785d2ce

[dry kestrel]

Check the citations it pulled from

also use the reasoning model

[distant aurora]

I used reasoning.

I never switched to non reasoning. Idk why it's behaving like non reasoning.

Wait. I'll send recording.

[dry kestrel]

...

[dry kestrel]

Please make a bug report

[dry kestrel]

https://www.kimi.com/share/19c82191-79a2-80a5-8000-0000e809b003

[distant aurora]

Seems like they paywalled thinking.

It doesn't notify in the app. Only on web.

[valid charm]

Dammmmn nice find! Had no idea Gemini 3.1 did so well!! Is 3.1 your currrent top used model??

[dry kestrel]

They are experiencing high demand from the influx of users after K2.5

[winter nova]

why is it k2.5 better than like sonut or smth

[green crystal]

Yeah I saw
Pretty nice

[rustic nest]

Why can't we use agents in perplexity app?

[wise nova]

They don't have money

[wise nova]

So low

[wise nova]

GLM will be enough. Qwen no need

[rustic nest]

What's that mean?

[wise nova]

Burn rate higher than they earn and get from investors

[rustic nest]

One agentic swarm request a day?

Or is perplexity making their own?

[wise nova]

They don't have resources for that

[dry kestrel]

Source? Because right now that sounds madeup.

[dry kestrel]

What type of agent are you talking about? If it's Kimi Agent Swarm, Moonshot AI hasn't released an API for that yet.

[rustic nest]

Ahhh okay that makes sense

[wise nova]

Burn Rate Dynamics: Application Layer vs. Infrastructure
The core financial challenge for Perplexity is its structural burn rate, which is driven by negative unit economics at scale.
Foundation companies (OpenAI, Anthropic, Google) incur massive, fixed capital expenditures (Capex) to train models and build data centers. Perplexity, operating at the application layer, relies heavily on these foundation models. Consequently, Perplexity's primary expense is inference—the variable cost of processing user queries in real-time.
Every time a user submits a query, Perplexity must pay for server compute and, frequently, API fees to competitors like OpenAI or Anthropic. As user volume increases, operational costs scale linearly, meaning the company cannot easily "grow its way out" of its burn rate without severely optimizing the cost per query.

[dry kestrel]

Thanks for the AI generated content, you could argue the same for companies like Poe, Monica, Manus, Cursor, Lovable.

[wise nova]

The recent reduction in query limits for Pro users, alongside the blocking and banning of free Pro subscriptions, is a direct operational response to the burn rate dynamics outlined above. Perplexity is enacting these austerity measures for several critical reasons:
Negative Unit Economics on Promotional Accounts: To drive rapid user acquisition, Perplexity previously distributed complimentary Pro subscriptions (often via corporate partnerships). However, unlike traditional SaaS products where the marginal cost of a new user is negligible, AI search incurs direct, substantial costs for every query. Heavy utilization by non-paying users drains capital rapidly. Revoking or limiting these accounts is a mandatory measure to stop financial bleeding.

[dry kestrel]

Please don't text wall

[dry kestrel]

You blame the company for reductions in rate limits, yet you encourage other users to purchase 3rd party promotions that violate ToS

[quaint ravine]

anyone know what is limit for Model Council in PPLX?

[dry kestrel]

50

[quaint ravine]

per month?

[dry kestrel]

rolling window

regenerates at a decent speed

[quaint ravine]

estimated regen?

50 per month is about 1.5 a day

[quaint ravine]

what is "decent" exactly? 1 a day?

[wise nova]

What? Where?

[quaint ravine]

what did you do/

[dry kestrel]

#💬│general message

[wise nova]

And why did you conclude that I encourage others to buy? Did you read the message and context carefully?
You have drawn your own conclusions and are projecting them onto others. Check and read again the full context

[hasty pebble]

I also blame perple perplexity for reductions in rate limits

should I praise them?

😂

[prime elbow]

Anyone know how i can look up the json file to see my actual limits i have on perplexity for file uploads,pro searches, and etc?

[hasty pebble]

https://www.perplexity.ai/rest/user/settings

https://www.perplexity.ai/rest/rate-limit/all

[prime elbow]

Thank you

[wise nova]

Pplx clearly overcommitted by handing out promotional Pro subscriptions to rapidly inflate their user base without forecasting the actual costs. Now that they’ve realized they can’t afford to fulfill those obligations, they are slashing limits and looking for any excuse to terminate promotional accounts—even impacting annual paid subscribers. It’s a fundamentally unfair business practice. They should honor their original terms for existing users and only enforce the new restrictions on future sign-ups with clear advance notice.

[hasty pebble]

it's a private company,they can do whatever they want
and we can shit on them if we want

[wise nova]

This isn't about blame. It is a request for fairness and respect for every user, regardless of whether you are paying $0, $20, or $200

[dusk elk]

https://youtu.be/OC7sNfNuTNU?si=wUOsOEiDOzi8Iuaa

[wise nova]

They must adhere to ethical standards and demonstrate respect. Their actions, driven by a desire to cut costs, were highly inappropriate. No other AI company has been involved in a controversy of this nature

[hasty pebble]

they don't have to do that

[wise nova]

In the end, this approach will cost them their users trust and cause them to lose the AI race

[hasty pebble]

who are they racing against ?

[prime elbow]

We're all gonna be killed off by AGI either way in the future so theres no point

Once they realize humans are the problem, they'll just turn on us