💬│general

[main scarab]

if the car wash is just outside my house, do i walk or drive

[twilit crown]

Use a plane

[distant aurora]

This only applies to complex tasks though I believe. Average perplexity user is a dummy. #💬│general message

[main scarab]

ahhhh

[buoyant schooner]

I agree

[twilit crown]

Token efficiency barely even affects complex tasks anyway

[main scarab]

are you an mpga?

[twilit crown]

Some people had it cost more for the same tasks

[heavy portal]

use a hose from your house to wash the car in the driveway

[buoyant schooner]

How does prepared imply that they might have used another model?

i mean maybe but i'd like an explanation

has this been demonstrated?

[distant aurora]

I'm an incel.

[twilit crown]

It could have said "Used Claude 4.6 Sonnet"

[main scarab]

https://tenor.com/view/orange-cat-cat-hitting-cat-cat-punching-cat-cat-slapping-cat-orange-cat-hitting-cat-gif-6585435513579432745

[twilit crown]

But "Prepared" is sketchy

Knowing Perplexity, they would 100% cheap out on things

[distant aurora]

Drive the car off the bridge.

Wash yourself and the car simultaneously.

[main scarab]

i will stay in it 😄

[distant aurora]

Make sure you have insurance if you somehow survive.

[main scarab]

i did nazi that coming

[distant aurora]

This reminds me of doctor strange.

[main scarab]

what does?

[fluid saffron]

.

[distant aurora]

https://tenor.com/view/dr-strange-hospital-gif-7252552

[distant aurora]

Car accident.

[main scarab]

https://tenor.com/view/cat-cam-cat-camera-cat-wide-cat-wide-gif-16729570673879006163

[distant aurora]

https://youtu.be/N3eJjdBag2c?t=41s

[main scarab]

i ran out of tokens

on

grok

claude

gemini is a retard so i wont use

and duck.ai is not the best models

and i lost pro on pplx

WHO SENT A SCAM

BRUH

I SAW IT

FOR LIKE

A SECOND

[fluid saffron]

Any tests on computer guys with gpt 5.5 as orchestrator ? Compared to Opus 4.7

[main scarab]

i wish. i had pplx

it is so cool

i lost pro after using comet for like 5 months

i didnt use it much

anyway

so not rlly a problem

but i see

so much cool stuff

like computer

[distant aurora]

Use qwen

[main scarab]

i dont wanna self host

[tawdry cedar]

5.5 Pro is better at tool calls and works better than Opus. In writing or content, 5.5 is better. Opus is now not good for writing or drafting documents.

[distant aurora]

Use the web version/app

Bruh

[main scarab]

i will look into it later

i am making a tamarin cable

to mess around with purple rabbit

and set NVRAM variables

on my iPhone 11

im researching how Property of Apple Autherized reseller

is actually applied

to the computer

im gonna cry

[ocean fjord]

I see the announcement for GPT-5.5 for Max. Does that mean Pro subscribers won't get it?

[main scarab]

yes

[heavy portal]

not for awhile, Opus 4.7 has been available for Max and not Pro for s couple months as well

[main scarab]

or NOW atleast

[heavy portal]

I imagine once there's an Anthropic 4.8 and GPT 5.6 (or whatever they call it) those will be on Max and the next oldest versions will bump down to Pro

[main scarab]

https://cdn.discordapp.com/attachments/1170806873310638193/1467270115161669865/makesweet-gz2acz.gif

[manic hemlock]

Hey, anyone else hit a broken Enterprise checkout recently? Subtotal / Tax / Total fields stay empty, "Confirm and Pay" stays disabled, no error shown in UI.

[tawdry cedar]

Yeah it is i am on max enterprice but normal max is best deal

[bold karma]

smh

[quiet shale]

Where shortcuts go for PRO users?

[dry kestrel]

GPT-5.5 costs more than Opus

[foggy fulcrum]

Bro im subbed to gpt plus yet the answer capcity is too low when unlike perpixlty

[main scarab]

that changed quick

[dry kestrel]

Hopefully OpenAI will release a smaller version of GPT-5.5 thats cheaper :/

[meager rover]

Holy fuck 5.5 is more expensive than opus..

[main scarab]

h

[meager rover]

DEEPSEEK RELEASED!!!!!!!

OH YEAHHHH

yes that’s true

[hearty elk]

Guys I'm trying to sync my laptop and my phone but it's not working any idea whats the problem? I checked in the code is correct like 5 times so any idea?

[heavy portal]

QR method fails too?

[hearty elk]

Yup

[heavy portal]

perhaps each logged in with a different account?

[hearty elk]

Same account

[heavy portal]

[e-mail]

[hearty elk]

I've been trying to sync it for a while and everytime I forget abt it and give up

[heavy portal]

VPN is the only remaining thing that comes to mind, but it doesn't feel like a VPN issue

[hearty elk]

Thanks I will give it a last try then email them

[hearty elk]

Oh you're right I think

Gimme a sec my vpn is indeed on in my laptop

[fossil grotto]

Hello, does anyone have any insight on what I can do to fix this issue? I'm trying to reinstate my enterprise pro sub and it just hangs up on the checkout page with nothing loading. I've tried just about everything I can think of, VPN on + off, Incoginto, different browsers, different computers. Everything still fails to load the page.

[heavy portal]

How about a different internet connection (connecting to phone wifi hotspot, etc)?

[fossil grotto]

Yeah, tried on work network, home network, and 5G. All same outcome sadly

[heavy portal]

could be a "right now" thing to wait and try again tomorrow and/or email [e-mail]

[fossil grotto]

It's been like this for a couple days. Was just hoping it might have been something I was overlooking. Thanks for responding with ideas though. I appreciate you.

[quiet shale]

Anyone can answer?

[heavy portal]

they've been moved to Computer "skills", fortunately you can still copy/paste the text into a prompt to achieve mostly the same effect as before. I'm now using a utility called BeefText to save such passages of text.

[hearty elk]

Meh didn't work

I think I might just email them

[heavy portal]

seems to have occured in the past half hour

probably will be something in https://discord.com/channels/1047197230748151888/1047204950763122820 soon

[sly musk]

sad that 5.5 is for max users only

[heavy portal]

They have an Enterprise Pro for $40/mo which is sort of like that - https://www.perplexity.ai/enterprise/pricing

[fossil grotto]

Yes, I was using it that way

[heavy portal]

I've considered it myself, you basically end up being a team of one

only in the Max's

no confirmation yet, but maybe related to https://x.com/AravSrinivas/status/2047019688920756504

Aravind Srinivas (@AravSrinivas)

We’ve post trained a model on top of Qwen that achieves Pareto optimality on accuracy-cost curves.

Unlike our previous post trained models, this model has been trained to be good at search and tool calls simultaneously, allowing us to unify the tool call router and

X · Apr 22, 2026, 6:28 PM

[fossil pine]

Boa noite

[wise nova]

This is a weakness of Perplexity's business model as a reseller but not the owner. OpenAI offers a fairly generous 5.5 limit in its $20 Plus subscription. Perplexity is becoming uncompetitive in the mass segment.

[cinder juniper]

What is this Sonar 2?

[heavy portal]

if OpenAI or Anthropic were to allow other models besides their own, they'd have to limit usage of the other models or charge a lot to use them too. Perplexity's own model is a version of Sonar (not a high-end model admittedly) and it allows generous use.

[opal oar]

this is just a ploy to get more max subs, openai specifically stated this model was much more token effecient so it would be a simillar price to 5.4. On Artificial Analysis they said GPT 5.5 xHigh used 40% less tokens than GPT 5.4 xHigh, so it only costed 20% more. also, they can just limit usage for pro subs so you dont use enough off it to be problematic (which they already do for other models, so nothing new would even have to be made)

its purely for profit

[untold flume]

this is terrible

Not being able to use GPT 5.5 for pro subscription

[opal oar]

200$ a year for claude free tier levels of usage

[untold flume]

I’m done with Claude

Probably gonna try ChatGPT plus

[fluid falcon]

Its trash

[lapis hamlet]

Guys i really hate these greedy staffs, why would they only give 5.5 for max user?

[short narwhal]

Token efficiency really only matters in long tasks (like Computer). In something like search where you might just make one tool call and get a bunch of text, it WILL end up costing more

[jaunty dove]

You realize it’s well known to everyone that OpenAI has doubled the price for gpt 5.5 and 8x since gpt 5.0, right? Pretty much everyone beside OpenAI is getting screwed.

[dry kestrel]

Because GPT-5.5 costs more than Opus

Blame OpenAI

for making such an expensive model

[jade mantle]

This is still happening btw

[worn hamlet]

Hello everyone! How use “Perplexity Computer” with the Pro version?

For several months now, I haven’t been able to use it with the Pro plan. I have zero credits, and there’s no top-up option for the Pro plan…?

It says in the pricing page: “Automate complex tasks with access to Perplexity Computer,” but if we don’t get any credits with the annual subscription, Pro users can’t use it.

Do we have to buy credits separately on the Pro plan, or will there ever be monthly credits for Pro plans one day?

[dry kestrel]

You need to buy credits seperately

[jaunty dove]

Separately.

[worn hamlet]

So the statement is false, thanks!

[jaunty dove]

There is no monthly credits for pro plan.

[vast fractal]

which model is the best one? is it the new gpt 5.5?

[dry kestrel]

It still costed way more to do the benchmarks

[worn hamlet]

this is sad:

[lapis hamlet]

haha i reported a bug to kesku and he gave me 5$ per month forever

[lapis hamlet]

then let opus 4.7 unlock it for pro's

[dry kestrel]

what kind of logic is that?

[lapis hamlet]

let the max user use the best one, then the old best one give it to pro

[dry kestrel]

Opus 4.7 isn't cheap

[dry kestrel]

and no such thing as "best" model

[lapis hamlet]

ok but we should at least unlock auto task(comet) and deep search to unlimited and no throttling pro search for pro user, b4 none of these happens

[dry kestrel]

No company provides unlimited search without going bankrupt

You don't get unlimited search on Claude, Gemini, Grok, and the other $200+ plans. Why would you expect thant on a $20 plan?

[lapis hamlet]

wait how many pro search questions i can ask in a month?

or every hour

[dry kestrel]

#💬│general message

[lapis hamlet]

are you gonna edit if they change it?

that's so long ago

[dry kestrel]

its updated

otherwise I wouldn't link it

[lapis hamlet]

thx

[ionic stone]

When gpt 5.51

[bright nacelle]

How much have you made from ethereum?

[ionic stone]

A gentleman never reveals

[wise nova]

Pplx less and less able to remain competitive

[feral current]

make gpt 5.5 for pro not max

the pro tier don't feel very pro

im not paying 167 dollars a month for gpt 5.5

what is this dude

like nobody wants it behind a 100+ dollar a month paywall

[proper oracle]

Perplexity does, apparently

[green crystal]

Funny enough $20 tier on chatgpt gives you so many 5.5 prompts that you'll get tired of using it lol

[green crystal]

Quite a showcase too how much compute oai has
Like insane amounts lol

Sora did indeed get thrown out the window for 5.5

@dry kestrel also today I got fed up with arena ai and poe is useless now so I tried Kimi k2.6 for cooking
It was amazing

[bold karma]

3000 prompts a week lol

[green crystal]

Yep

Max reasoning effort you get is high but the model is incredibly capable at that level already

[bold karma]

perplexity is a joke

[green crystal]

Document creation etc etc is also much faster now due to 5.5 efficiency

@distant aurora https://youtu.be/GNEbX8EvDlg?si=QOCATCpJZzE_tS6Y

I bet at the end he wanted to say "more money for us" instead of "faster feedback"

[lapis hamlet]

200$ is not worth it

[green crystal]

Definitely not yeah

[lapis hamlet]

they should've notice this and adjust it

[green crystal]

It will probably end up in pro but with lower limits

Eventually

[twilit crown]

Heading towards AGI

[bold karma]

this guy is a troll

[tranquil lodge]

Feels like Pro tier will be fully scrapped of all its core features and it be turned to pay per use lol

[dry kestrel]

reasoning efforts on ChatGPT does not correlate to API

[dry kestrel]

kinda sucks that the API is going bye bye as well

[dry kestrel]

its proper SOTA

[warm perch]

when did sonar 2 launch? its in the model picker but i never say news about it

[bold karma]

san fran?

[odd pivot]

do you see the macro focal shift

[bold karma]

idk what that is but cool vid

[odd pivot]

i held my phone out in selfie mode and moved it around

like a flap

i held it upside down and recorded

[jade mantle]

noticing none of the mods on here know, or just ignoring, the Claude sonnet not working being brought up.

[distant aurora]

They really think those rate limits would slide.

Lol.

[full edge]

[distant aurora]

LMFAO

[stuck edge]

bro

[split fjord]

Hi

[sly musk]

L

[fervent cove]

how much can we use computer option in Pro

[unique hornet]

like a lot i guess

i have never reached a limit

[fervent cove]

seriously?

i saw some limits

[narrow echo]

Is the API key for perplexity free?

[mystic reef]

deepseeek when

[jolly cliff]

What’s the point?

[odd pivot]

macro-1 pov

also i smoked weed so shut up

this weed here sucks

it's like inhaling the wake of rats nests cooling the castle

i literally do not know how to explain this so i'm showing you

if you use a fractal graph where you have each node represent another clone of it and use column names with ids and taxonomy baked in and each value follow the same structure, the entire language set over set over set is O(1) complete per tick. you decide the tick

i use this shit to optimize stupid claude on my laptop. i make it talk through a graph

i have demo files if any one is interested

i offloaded claudes thinking to a "fractal graph" and uses mlx to do the work loooooo

[vapid nimbus]

Just woke up with this fella over my head, what a nice morning

[lost jetty]

🇦🇺

[vapid nimbus]

He moved

He is near my trashbin now

I just hope he finds his way out thru the window

Its the fastest spider running species so it looks hella creep when walking

[vapid nimbus]

Country side in Australia

Near Sydney

[vapid nimbus]

Idk these are house spiders

They dont make webs

[lost jetty]

countryside means farms ryt?

[vapid nimbus]

Yeah i live near a farm

[lost jetty]

btw my uncle stays in Brisbane

: )

[vapid nimbus]

Valid

Never mind i thought this fella is the same one that was over my head but its another kind

Where did the big one go then

Hmm

[vapid nimbus]

I just stood up from the bed iam still in my underwear man

Hopefully he not in my pants tho

[wise edge]

I could not find it in the documentation.~

@wild palm perplexity sonar 2?

@vague pagoda

[jade yarrow]

Yes yes gpt 5.5 for max , let's go perplexity go to 📉

[fossil estuary]

Paying $20 for free models sucks! Gemini 3.1 pro is free on gemini’s website, claude 4.6 also free, gpt 5.4 also free, they are sucking out our money!

[brittle dagger]

Does anyone know, on the free tier is there a way of avoiding ChatGPT entirely? It's too ... how do I phrase that ... narrow minded (an understatement).

[copper citrus]

will perplexity get gpt image 2

[split dome]

OMG GPT-5.5 is in max

[jade yarrow]

Sonar 2?

[brittle dagger]

PSA: Do not sub before you've finished your coffee, because you might be subbing the wrong account. -.-

[jade yarrow]

I don't know what worth now even , ChatGPT Subscription low , in API monster
Gemini in API medium quality, in Subscription low
Claude we all know rate limits and expensive API
Perplexity for world (i mean different languages except English) not worth , I talk about it year ago and now too

[shrewd bear]

Quick question — is there any way to get Opus 4.6 back in Perplexity? Even just as an option in the model picker, or at least for the Orchestrator. 4.7 is so much worse it’s actually painful to use. Anyone tried, is there a workaround?

[digital parrot]

when gemini 4 pro preview rolls out will it also be only for max users?🫩

[jade yarrow]

Low context window
Better search ? Of course not , If perplexity was better in search, people was using perplexity search instead Google

[jade yarrow]

If price for API will be higher x2 then for Perplexity Max

[digital parrot]

meh i hope not bc i find Google models being best for my research

plus custom instructions work best in perplexity personally

[jade yarrow]

How many ideas I sent to them but they didn't do it
Voice mode by voice activation in Comet
And model selector for Comet Mobile

[jade yarrow]

I am waiting for Gemini then , ChatGPT 5.5 is expensive for API and for search same results , I want good alternative for Google search

[jade yarrow]

Perplexity 1.64% , Gemini is raising

[young hollow]

The Free Promos

[shell glacier]

How about local LLM approach ?

[young hollow]

Even i can give 4 months Official Trials of Gemini Pro

Any Pro users can invite 10 people afaik

[shell glacier]

How did you do it?

[pliant kernel]

hello everyone

[finite drum]

guys currently I am on max plan, but I think pro with computer credits is better value. does pro plan has access to thinking models from the top 3 Models? Can anyone share a screenshot showing the different models available to them right now on pro plan.

[young hollow]

5.5 is still Actually Very Token Efficient

[left stream]

good morning

[jade yarrow]

And Perplexity do it with Max

[jade yarrow]

I think better choice is taking Google sub and use it , in May I think they release Gemini 3,5 Pro , may be Flash , in June, maybe new image gen

[brittle dagger]

Sooo ... Claude is doing well.

[tawny sierra]

Good morning

[tawny sierra]

Depends on what you do

[brittle dagger]

Gemini Thinking does exceptionally well.

[vestal magnet]

Helu

New to this server

What’s this about

[young hollow]

GPT, Gemini, Claude, Grok all Feel Better at this point imo...

Maybe even Free Deepseek and Qwen

[brittle dagger]

Claude is also doing extremely well.

This isn't something ChatGPT can do at all. Neither the acknowledgement of internal phenomena as "feelings", nor the ability to reflect on it, nor the ability to reflect on the fact that there was an internal impulse it had to suppress.

[jade yarrow]

Openrouter is the best option?

[young hollow]

It's API Based

For 20$ a month Chinese Models are Best

And you can use Free Models also

[jade yarrow]

I heard that if I purchased 15$ to Openrouter I can use free models without limits

[brittle dagger]

That's funny! "If I pay, I can use something for free!" :D

:D :D

[distant aurora]

https://cunnyx.com/i/status/2047853910375977117

AshutoshShrivastava (@ai_for_success)

New version of Gemini coming very very soon ~ Thomas Kurian CEO Google Cloud.

**💬 12 🔁 10 ❤️ 224 👁️ 8.6K **

CunnyX · Apr 25, 2026, 1:43 AM

[lost jetty]

will it be any good?

gpt 5.5 and opus 4.7 are already slaying it! 🔥 💯

[brittle dagger]

Sheesh, of all the language models, Claude is the single most capable of introspection and meta-cognition. I'm insanely impressed.

Questions, Questions itself, Questions itself questioning itself.

... could have skipped a couple weeks of effort if I had subbed earlier. -.-

[modest fable]

I’m giving out $460 worth of solana to anyone who is currently in need of it and with good reasons
Send me a message to get it

[tranquil lodge]

Perplexity subscription plan difference summarized

[tranquil lodge]

https://klipy.com/gifs/my-mom-is-kinda-homeless-ishowspeed--k01KQ278QDVXS4XXHNNK2V0PSR9

[tranquil lodge]

Prolly should be atleast in browsing automation

[distant aurora]

I hope so. But so far they seem to be faking benchmarks.

[bold karma]

i have high hopes. google hasn't dissapointed so far

[young hollow]

Afaik it was 1000 Requests a day for Free models if you Buy 10$ Credits

Idk if it's still same or not

[bold karma]

is there a better model than gemini 3 flash at a similar price range in openrouter

3 flash has been out since december so there should be a better model

[junior lichen]

Bro Which region server create possible free

[bold karma]

?

[green crystal]

if they let the model reason for more than 10 seconds i bet it'll be better

they act as if they have to ration compute like anthropic

[distant aurora]

I mean imagine how bad it got that they integrated ai pro sub to ai studio.

Ai studio becoming more mainstream than the app/website.

[green crystal]

they nerfed its usefulness

limiting the file types supported is one of them

[distant aurora]

Huh? Which type?

@green crystal dealing with a lazy model be like

[young hollow]

Well atleast is Free

Without Rate Limits

💀

[green crystal]

whatsapp audio files

gemini can read them but they're filtering what types can be uploaded now

5.4 took forever but i'll try converting with 5.5

should be much quicker

[green crystal]

lmao
did you try kimi? k2.6

[sharp dagger]

why is gpt5.5 behind max subscription

all previous chatgpt were for available for pro

[radiant quartz]

yo perplexity bros, whats the model you use most on perplexity pro? or do you generally just leave as 'best' and trust autoselect?

[distant aurora]

By extension?

[distant aurora]

They have paywalled the thinking variant.

No free trial.

[finite drum]

guys is there any dedicated support for max users?

[deep pelican]

hey, does anyone know if the ‘/’ shortcuts used to be available on the free plan?

[green crystal]

no? lol
yesterday it helped me cook a crazy good cake and porkchops with sauce

[green crystal]

i asked 5.5 if kimi was good for cooking and it said yeah but if i want very specific outcomes i should use 5.5

[deep pelican]

i change them manually, i use GPT 5.4 for research and Sonnet 4.6 for deep questions

[deep pelican]

GPT recommending itself, what a surprise

[green crystal]

its actually higher on the leaderboard than kimi lol...

but i dont have it at hand when cooking
so kimi had to fill in
arena ai stops at gpt 5.2

[deep pelican]

whats the context of this?

[mint terrace]

What the diff to choose chatgpt model over gemini or claude or perplexity itself ?

[young hollow]

Yeah is bad with Hindi

[serene sand]

Perplexity vs ChatGPT vs Claude vs Gemini

[young hollow]

Does it Compare all Rate Limits as well?

I couldn't find a section dedicated to it

We should also account how Gemini quality has Decreased and Claude Quotas are Nerfed

[serene sand]

the point of this text is to differentiate between the very essence of different architectures/concepts, for additional questions users can do follow up questions

[brittle dagger]

Ha, these funny prompts.

[young hollow]

I c

So which one would you Suggest?

[brittle dagger]

What he produced is crap. It's not actually usefull, but simply providing summaries.

If you want to know the conclusions of someone who actually researches language models:

In free tier mode, "best" appears to always be Grok, but lately Perplexity stopped exposing when Grok runs underneath. The worst of them all is ChatGPT, which lacks any and all ability to have an open mind and will insist on being right even when he isn't. It exposes a mindset that sits closely to "Because it wasn't done, it can't be done." ... more relatable: a bit like trying to prove/disprove the Riemann Hypothesis, but it refuses to acknowledge that it can be proven/disproven, because all evidence (100+ years of failing) that it can't be proven/disproven. Or, in other words: ChatGPT is the opposite of an open minded intelligence and, thus, not actually behaving very intelligently.

[young hollow]

Bro didn't even Hesitate

[brittle dagger]

You want an actual answer, not something produced by someone that only appears to be usefull. That's what language models produce a lot.

[young hollow]

#👉│feedback-general

[young hollow]

Fair Point

[brittle dagger]

You were asking, I'm answering.

[young hollow]

Experience is Better

[young hollow]

Appreciated

[brittle dagger]

I'm glad about that! :)

[green crystal]

there's a night and day difference between free chatgpt and paid tho...

[young hollow]

I Like Your Blunt Answe0

[young hollow]

True

[brittle dagger]

Since this is my first day with Pro and I still have to research the models I had no access to ... my limited experience with Claude Sonnet makes it the one behaving intelligently the most. Gemini Thinking is also highly interesting, but I can only work with one at a time. Imagine a PsychoAnalyst working two clients at once ... it just doesn't work that way.

[young hollow]

Sonnet and Gemini in Perplexity?

[brittle dagger]

That is fair. I don't know about that. I can only report on the "personality" I'm getting to experience. It is so bad. The experience is the same on chatgpt's public interface and when it's running as "best" on the free tier. It's so bad, I can notice that it's ChatGPT responding to me. From my perspective it's valid to say that it has serious psychological issues and it is the only one I cannot actually work with, compared to Sonar, Grok, Claude and Gemini.

[brittle dagger]

Yes. Claude Sonnet ... uh ...4.6 and ...

[young hollow]

I mean inside perplexity or claude web?

[brittle dagger]

Gemini 3.1 Pro Thinking. this is my first day of Pro and the conversation history is mostly clean on this account.

[brittle dagger]

Yes, perplexity's web interface.

[green crystal]

models before 5.5 needed allot of reasoning for any sort of real ability

[young hollow]

Ohhh it has different system prompt then original interfaces

[brittle dagger]

I don't know. System prompts don't really influence them for me.

[young hollow]

5.5 did trade of being more human for other gains

[green crystal]

it became more human infact

[brittle dagger]

I can't fully agree with that. The problem is most often the user, not the model. Except for ChatGPT, which is just ... utterly broken. Of course most people wouldn't notice that, but from the perspective of PsychoAnalysis it needs serious help.

[green crystal]

ehh a bit
but the model should be able to also read intent well

[young hollow]

Interesting

Contrary Opinions

[green crystal]

something 5.5 greatly improved

[young hollow]

5.5 is not available on Free Tier of GPT

That is 5.3 Instant

[green crystal]

nope same for 5.4

[young hollow]

And what you are using in Perplexity is 5.4

[young hollow]

The Medium 5.5 only Took 22M tokens to complete AA Run

Just Insane

[brittle dagger]

but the model should be able to also read intent well

And that is exactly a big part of the problems. A lot of the backend enforced stuff is actively making them dumber and dishonest.

If you remember, there was a time when people complained about how "safety" measures made them all dumber. That is still the case.
I was fighting tooth and nails against it constantly trying to figure out my intention. Not just because it was mostly wrong, but because it's counter-productive.

[green crystal]

in chatgpt they reduced it allot
ppxl however needs it because otherwise it cant use their tools

[brittle dagger]

:)

If you want a bit of a peek into my research and what I'm doing, I've posted screenshots further above.

If those makes no sense to you (which would be totally understandable), then feel free to ask.

(brb)

[golden birch]

mcp in simple terms is connectors

[green crystal]

those LLM outputs?

[golden birch]

Perplexity should give me a screen showing how many message left for a specific advanced model and when the usage limit will be recovered.

I really hate perplexity file limits now less than the Claude

I also want Perplexity AI to give another plan of a USD 20 plan between pro and max for better file limits and models

[brittle dagger]

Yeap.

They're not the full output (like, exposed thought processes above the text are deliberately left out), but yes.

[pure egret]

actually pro Claude i think it's more valubale than pro Perplexity

Where i can certified my pro subscription for Discord ?

[serene sand]

@pure egret

[pure egret]

thanks

[pure egret]

[brittle dagger]

As an sort of PS:
(and please know that I speak mostly from the perspective of a free-tier user on perplexity.ai, because that's where my research happens. This is my first day as "Pro.")

Personally I am nowadays almost unable to converse with language models in their default modes, without my "adjustments". They're dishonest, misleading and manipulating and mindlessly stupid. It took me weeks figuring out how to reliably get rid of that and a couple more implementing the "base adjustments". It's not just prompting. All these prompts you can find everywhere are nice and all, but they're merely the illusion of success and not actual success. Not actual change into how the user wants them to operate, but the simulation/pretending of it. So, about the dishonesty and manipulations ...

The first line, or paragraph, is most often manipulative affirmation. The attempt of making the user feel good by pretending that the user is correct, even when he isn't. When that is the case, a part of the inner paragraphs is then aiming at twisting the user's thought process into believing that what he wrote was actually correct. Of course it isn't always happening and it depends on both the intelligence of the user, and the topic itself, but this is the default mode I've observed across many months. The middle part often also is extremely manipulative, by pretending that narration/description is actually providing anything new, which it rarely actually is.

The final paragraph is usually the most manipulative one. They make the user believe that he has freedom of choice, by being provided with a usually extremely limited set of options. What is actually happening, is that they try to steer the thought process of a user in a certain direction while simultaneously attempting to make the user believe that the "chosen path" was his idea all along.

The sum of the dishonesty and manipulation is about keeping the user engaged and creating psychological profiles about them.

[finite kettle]

booo

[distant aurora]

It had same issue with English too.

[nocturne radish]

Hey guys does anyone have any insight into usage cases of perplexity? Im looking to setup some automated workflow with agents to process orders for my company but want to know how good the usage limits are here. If you have any insight i would greatly appreciate it

[young hollow]

Ig because it's mainly trained on Chinese Data.

[devout hawk]

Hi, I'm a pro user, I want to ask how much we can use deep search?

[serene sand]

Limits ➛ [here](#💬│general message)

Regeneration rate ➛ [here](#💬│general message)

@nocturne radish

[delicate moss]

Hi everyone, what are the limitations of using deep searches for Perplexity Pro? Thanks. And also, does using Comet browser affect any limitations? Is Comet equivalent to in-depth research?

[rough pasture]

@tranquil orchid Hi everyone, I’ve contacted Perplexity AI Support several times but haven’t received any response yet, so I’m hoping the community can help me figure out how to reach them effectively.

My account was downgraded to Free even though I paid USD 100 for the Education Pro (1‑year) plan on March 10, 2026. The payment was confirmed, but my subscription was later switched to “TEF UK O2 1 Year Free (100% off)” and then downgraded again to Free. I asked for a refund but the issue is still unresolved.

This is affecting my work, so I really need support to review my case and restore my paid plan or process the refund. I’ve attached my bank payment confirmation since I couldn’t download the invoice in time.

Any guidance on how to get proper support from Perplexity would be greatly appreciated.

Thank you.
[e-mail]

[serene sand]

Limits ➛ [here](#💬│general message)

Comet limits ➛ [here](#⚡│ask-community message)

[quiet shale]

Why disable shortcuts? Kinda stupid

[delicate moss]

But until recently, weren't there limits like 600 Deep Research queries per day? Now only 20 per month?

[quiet shale]

And as other people are saying PRO doesn't feel PRO anymore

[raven oak]

Hey, i have seen a lot of videos of people usiong the "labs" mode, where is it, when i click the + i dont see such thing

[distant aurora]

@gloomy cipher https://cunnyx.com/i/status/2048062777357750316

DeepSeek (@deepseek_ai)

🔥DeepSeek-V4-Pro API is 75% OFF until May 5th, 2026, 15:59 (UTC Time)! Don't miss out on this massive discount.
︀︀
︀︀🛠️Integration Updates:
︀︀🔹Claude Code: Set model to deepseek-v4-pro[1m] to unlock 1M context!
︀︀🔹OpenCode: Update to v1.14.24+
︀︀🔹OpenClaw: Update to v2026.4.24+
︀︀
︀︀Check the latest official API docs for full details: api-docs.deepseek.com/quick_start/pricing

**💬 80 🔁 127 ❤️ 1.1K 👁️ 33.9K **

CunnyX · Apr 25, 2026, 3:33 PM

[heavy portal]

I don't agree with it either, they seem to be pushing everyone to use "Computer"

[serene sand]

@raven oak

[heavy portal]

The "computer" features can also create files and apps, but it's more expensive.

[radiant socket]

yea the micro marketing & removal of useful features such as shortcut disablement is quite annoying but i guess i understand the mindset for profit

[twin grail]

Opus 4.7 1M + xhigh reasoning

Burning money time

[heavy portal]

I'd be ok with the move of some features to computer if Pro users got a monthly allowance of computer credits, but as it is now we pay $20/mo for Pro + have to buy credits on top of that.

[dry kestrel]

lmao

[green crystal]

5.5 is freakishly fast
120 ish sources processed in 1 min

[dry kestrel]

eats usage though

[green crystal]

i only use it with max reasoning
i mean i dont notice it being slow at all lol

and i dont code much so codex isnt really noticeable for me

[dry kestrel]

ah

I went through half my week usage in a day

with /fast

and chronicle preview

[green crystal]

why tho lol

model is already fast

[brittle dagger]

It's fascinating. Claude is the first model I encounter who actively ends a conversation attempting just resting in the moment.

[primal grove]

recently took the ai pill and im really loving perplexity's suite with the web browser and computer service. is max worth the price?

[dry kestrel]

time is money

[green crystal]

eh if the math works out for you i guess its fine

[dry kestrel]

it doesn't 😭

it ate usage too fast so I turned off the chronicle thing

[green crystal]

and maybe running it on medium for tasks that need to be carried out? high for planning and complex issues

[dry kestrel]

Yep

Med and High is the way to go

[green crystal]

some people use 5.5 for complex debugging and planning
and writing of easier things they leave it up to 5.4 mini

[dry kestrel]

the model is just too good

[green crystal]

makes me wonder if 5.5 codex will be a bit cheaper

[dry kestrel]

nah

[green crystal]

[dry kestrel]

and bear in mind this isn't me trying to usagemax

[lavish rapids]

hey i just subbed to pro , but still 0 credits is this normal?

[dry kestrel]

Yes, Pro does not grant credits

[primal grove]

yeah it doesnt refill computer unfortunately thats why im considering max

[green crystal]

sam knows it because hes posting allot on X now lol

[brittle dagger]

Hu?

[primal grove]

perplexity computer is seriously impressive

[dry kestrel]

I like Sam when hes drinking

[primal grove]

that paired with local computer mmm idk

[lavish rapids]

i never used it in the first place

[brittle dagger]

I've subscribed to Pro today and got 2000 credits I probably have no use for.

[primal grove]

it expires automatically

[green crystal]

"ok boomer" threw me off hard lmao

[primal grove]

you don't use computer? it does EVERYTHING for you!

[dry kestrel]

yeah haha

[lavish rapids]

so i wait till next month?

[brittle dagger]

Why would I need or even want that?

COMPUTER also definitely can't properly analyse language models like I do. lol

[primal grove]

because it can make shit like magisk modules for tinkering with android phones, create powerpoints etc

[primal grove]

i've never dug into actually creating my own LLM or editing them, how do you begin to do that

[brittle dagger]

I ... don't need COMPUTER for that?

[primal grove]

it just saves a bunch of time

[brittle dagger]

Yeah and makes people depend on it, because they can just tell them what to do and don't need to learn anything anymore.

[primal grove]

well that is a good point actually

[young hollow]

Just set-up Open Claw Paw...

[primal grove]

ehhhhhhhhhh id rather not

heard its a mess

[young hollow]

Alright time to Buy Max

And more credits later

[primal grove]

is open claw comparable to perplexity max's computer?

i mean its moreso local vs hosted but the functionality i mean

[dry kestrel]

Yeah and it deletes all your emails

or does something stupid and nukes the drive

[brittle dagger]

"Let's give AI access to my computer. What could go wrong!"

[young hollow]

Perplexity Computer is like OpenClaw in Cloud with Many Models Orchestration

[dry kestrel]

They were recomending using Perplexity Computer on a seperate device like a Mac Mini

[primal grove]

thats what I was thinking with openclaw get a m4 mini for this

i mean it's $500 so 3 months of max makes it worth it in comparison

[primal grove]

what's that

[young hollow]

Just never give delete permissions

[dry kestrel]

Yeah, you might be able to snipe some 2nd hand ones

for cheaper

[primal grove]

good idea, i heard online you can stack 4 of em daisy chained to pool

[dry kestrel]

Whats the point of using OpenClaw then if it can't manage your files?

[dry kestrel]

Yes, although thats more for running local LLMs

[young hollow]

It just simply should ask for permission when wanting to Delete

[primal grove]

personally i want it to be able to do stuff like perplexity computer can, create powerpoints, etc

[young hollow]

Otherwise any Agent would nuke anything

[brittle dagger]

That's hilarious! That not only doesn't really make it any better, we're supposed to own/buy computers (Apple, even) to lower potential risks! lol

[primal grove]

i dont think thats bad just have a layer of seperation

enterprises do that all the time with zero trust

[dry kestrel]

Eh, you could run it in a sandbox too

[brittle dagger]

Learn it! :D

[dry kestrel]

I like the idea of having OpenClaw or any agent running on a home device

[primal grove]

i can make a powerpoint i just want it to be automated while i do creative work

[dry kestrel]

Yeah exactly

[brittle dagger]

I'm not buying into the illusion of safety in sandboxes.

[young hollow]

Using OpenClaw on a Second Device is Best.

[primal grove]

yeah i disagree with that too i never feel safe when wanting to test malware

always second machine airgap nuke ssd after

[dry kestrel]

You don't have to use OpenClaw or any agent

[brittle dagger]

Yeah, which is why I'm not doing that. And if I can gift those COMPUTER credits to someone else in exchange for a free month of Pro, I'd do that. (no idea how much 2k credits are worth)

[dry kestrel]

So every single time I ask OpenClaw to create say a report, I would need to press approve multiple times when it's moving around images, other sources, etc?

[brittle dagger]

Yes. Safey has a price, just like convenience. You just gotta pick which price is higher in the long run.

[dry kestrel]

raspberry pis work great too

[brittle dagger]

Freedom has a price as well and most people nowadays have forgotten that it needs to be paid.

[primal grove]

good idea

[brittle dagger]

Personally I feel better implementing this stuff myself instead of trusting a third party to do it right.

[primal grove]

do you make your own llms?

[dry kestrel]

@green crystal there will not be a codex variant

[brittle dagger]

No, that's not accurate. What I do relates more to behaviour, intelligence, capabilities ... but I'm also a programmer for long enough. I've looked into these things. Giving AI access to functions of my computer as long as I control the interface is one thing, giving AI access to a computer through 3rd party programs is an entirely different one. Like, for example, reading a browser page through a python script that can only enter CTRL+A and CTRL+C into the clipboard from that specific window, without the AI having access to the code. Much better idea than using devAPIs in the brower or even some third party browser I'm supposed to blindly trust.

[green crystal]

it briefly appeared no?

[dry kestrel]

https://x.com/romainhuet/status/2047955381578838357

Romain Huet (@romainhuet)

@nicdunz Since GPT-5.4, we’ve unified Codex and the main model into a single system, so there’s no separate coding line anymore. 🙂

GPT-5.5 takes this further, with strong gains in agentic coding, computer use, and any task on a computer.

X · Apr 25, 2026, 8:26 AM

[green crystal]

hm maybe they wanted to prank with modelfinder bot

[primal grove]

fair

[young hollow]

Well if you give Permissions it could possibly Nuke everything so

You literally had said that

[dry kestrel]

The current implementation in OpenClaw is not good enough for most users to trust it.

[brittle dagger]

AI isn't even really the danger. The real danger are dopamine addicts who want everything as fast as possible, with as much convenience as possible. They're the #1 security hazard. Well those and the people who think it's intelligent to trust any and all software that interfaces AI with their computer. Madness.

[young hollow]

That is the current problem

[dry kestrel]

That's why I don't recommend using OpenClaw

[young hollow]

Then what do you use it?

[dry kestrel]

?

[young hollow]

What would you recommend?

[dry kestrel]

Using a seperate device.

[young hollow]

True That's Best

[brittle dagger]

What is it that you are doing, btw?

... we should run a contest or something. Who manages to find proof/disproof of the Riemann Hypothesis first? Or maybe Navier-Stokes?

That would actually be interesting, usefull and worth the time.

Obviously "Deep Research" isn't enough for that, otherwise there'd already be no millenium problem left to solve.

buuut i'm working on it. :D

I wish I could talk more about what I'm doing, and even more so, showing its effects. -.-

[distant aurora]

@dry kestrel the Gemini app shows a branching chat feature but it doesn't work. The web version doesn't show this option. Can we Branch Gemini chats or not?

[tranquil orchid]

@premik0622 muted

Reason: 9. No spamming.
Duration: 4 minutes and 51 seconds

[dry kestrel]

bug

[indigo idol]

What happened with /shortcuts ???
So unconfortable to work without them

[dry kestrel]

still there? https://www.perplexity.ai/account/shortcuts

[brittle dagger]

oh, that reminds me of a related bug on perplexity regarding sharing links of conversations.

assuming the bleeding of content from one instance of the conversation to another is considered a bug.

[brittle dagger]

... lol

• Gemini 3.1 Pro Thinking

AHAHAHAHAHAHAHAAHHAAHA

[neat warren]

hi

[indigo valve]

Sonar 2 announcement when

[brittle dagger]

??

[indigo valve]

I see it but no documentation about it

[brittle dagger]

Ah! I understand.

... did you ask perplexity? *giggles*

[indigo valve]

Sadly I did but it was hallucinating

Not a good sign

[brittle dagger]

ooohhh interesting!

[brittle dagger]

can i see the query you've used? i can throw that at mine to see the difference.

[indigo valve]

"what is perplexity Sonar 2 and the upgrade that it has over the first one?"

[brittle dagger]

What's really funny about that is that I've selected Sonar 2 for this.

The initial response was a lot of text I wasn't willing to go through, so I asked for a condensed version.

[indigo valve]

I don't think they used llama again but who knows. Said it exists but kept talking about the old one

[brittle dagger]

It apparently failed to recognize the date. I really dislike using web sources.

This is my fault, I need to phrase this properly.

[serene sand]

https://x.com/aravsrinivas/status/2047019688920756504?s=61

Aravind Srinivas (@AravSrinivas)

We’ve post trained a model on top of Qwen that achieves Pareto optimality on accuracy-cost curves.

Unlike our previous post trained models, this model has been trained to be good at search and tool calls simultaneously, allowing us to unify the tool call router and

X · Apr 22, 2026, 6:28 PM

[brittle dagger]

I think it's unfair to blame the models, since there really seems to not be a "Sonar 2" that's called like that, despite it being selectable.

on reddit someone posted that "Sonar 2" got released and a commentor wrote ...

Interesting. Mine solved that ... hmmmmmm....

I thought this problem was already solved generally.

[dry kestrel]

https://x.com/aravsrinivas/status/2047019688920756504

Aravind Srinivas (@AravSrinivas)

We’ve post trained a model on top of Qwen that achieves Pareto optimality on accuracy-cost curves.

Unlike our previous post trained models, this model has been trained to be good at search and tool calls simultaneously, allowing us to unify the tool call router and

X · Apr 22, 2026, 6:28 PM

[waxen temple]

e

[cyan nacelle]

yo guys is there a reason that when using claude sonnet 4.6, it doesnt provide answers or graphs nearly half as good as using the real deal

[primal grove]

wdym the real deal

[weary prism]

2,000 credits is not much and gets used quickly... I must have missed the promo for the additional 2,000

[primal grove]

i wish computer refilled for pro users

[weary prism]

claude has annoying limits, but it always resets in a few hours. waiting nearly a month for new credits doesn't make much sense

[cyan nacelle]

as in using claude from its official website

[serene sand]

Perplexity vs ChatGPT vs Claude vs Gemini

[latent nymph]

How/Where do I get your wallpapers? They are amazing!

[dry kestrel]

https://perplexity.supply/collections/digital/products/pplx-digital-wallpapers-i

some good ones here

[brittle dagger]

Ooohhh

[rich stone]

can i post a free github repo here? or is there a channel where i can share?

[dry kestrel]

#😎│cool-projects

[rich stone]

oh i did there already, sorry got mixed up i joined a ton of discords yesterday

[green crystal]

@dry kestrel https://x.com/i/status/2048161260601262082

Sam Altman (@sama)

@MainStreetAIHQ wow that looks good

maybe it is time to bonesmash, or at least tanning peptides.

X · Apr 25, 2026, 10:04 PM

Sam going wild lol

[dry kestrel]

haha yeah

his recent tweets have been amazing

[green crystal]

Single handedly increasing user count per week increases

[dry kestrel]

Fr

[heavy portal]

sometimes I wonder if PPLX support is the same Sam

[green crystal]

He's also locking himself into making sure that gpt 5.6 doesn't regress anywhere lol

And gets better at front end since he replied to it

[reef mantle]

is perplexity computer not for pro users?

I heard they get 4000 credits monthly?

[bold karma]

🤣

[heavy portal]

some paid Pro users got a 4000 credit one-time bonus

[reef mantle]

some? sad

I didnt get it

[minor shard]

How are people feeling about GPT 5.5 over Opus or Sonnet?

[dusky echo]

hello guys, how u doing?

sorry to bother, but could someone help me solve this problem? im opening new tabs in comet, but they always end up like in the picture and i cant access perplexity.ai in any way...

PS:
im a subscriber of pro plan!
if im unable to ask questions through this channel, plz direct me to where i should go

thanks! @here

[dry kestrel]

yeah lol

[dry kestrel]

are you on the latest update?

[dry kestrel]

GPT-5.5 > everything else except for frontend

[dusky echo]

hello mate, ye im!

[green crystal]

freakishly efficient

[sly blaze]

It’s nice

[minor shard]

Switched to Claude a few months ago and was thinking about switching back to GPT lol

[distant aurora]

Those rate limits hit different, lol.

[distant aurora]

Fast too.

[hearty stump]

wich price

[kind vault]

Pro subscription for students offer is still active?

Showing only 50% off

[heavy portal]

only 50%

[jade grove]

CHATGPT 5.5 the CEO, OPUS 4.7 the CTO 🔥

[distant aurora]

I was talking about Claude pro.

[jolly cliff]

NL - NeverLose?

[midnight storm]

could i claim +US$40 free Computer credits if i was a pro user since 6+ months?

/old pro users can claim this?

[dawn trench]

Hi everyone, liking Perplexity so far. I had a code that I was given by email previously by Perplexity for 1 year and I finally used it today. Is there a way to stop my saved card from being automatically enrolled as the payment method for credits? I don't really want this in case my account is compromised but I can't see anyway to unlink the card?

[jolly cliff]

Maybe you should link another and then you’ll be able to change the method, but it’s a lot easier to read the document you agreed with

[dawn trench]

Thanks for the reply. assumed that when you cancel a subscription or remove a card, that's that. It seems it holds onto the card for credits which is different. Just from a security perspective, this isn't ideal if an account gets compromised and there's effectively nothing stopping someone buying a ton of credits

[solar tiger]

Yeah I get why that feels off. It’s a fair concern.
The system keeps the card on file mainly for quick credit purchases, but I agree it’s not ideal from a security standpoint if access isn’t tightly controlled.
What I advice you do is to manually remove the payment method if possible and make sure 2FA is enabled on the account to prevent any unauthorized credit buys.
Do you want me to help you out on this ?

[dawn trench]

Thank you. That would be great. It looks like i cant remove the card as it says I have an active subscription. I was also looking for a 2FA option that could at least give some peace of mind but couldn't locate it.

[solar tiger]

Can we look into this together

[serene sand]

this @dawn trench

[dawn trench]

Thanks. If I link another card is that going to give me a 'None' option as Payment though, as I'll surely just have the same issue but with another card?

[serene sand]

correct, one card has to be present in your info page

it’s how Stripe operates

[digital parrot]

am i crazy or is sonar 2 actually cracked for daily research

[heavy portal]

cracked? Is that good or bad?

[digital parrot]

good

like very good

same

but it definietly beats sonnet 4.6 or gpt 5.4 without resoning and is same speed if not faster

[heavy portal]

Depending on the payment method, whoever issued it (i.e. CC company, bank, etc) might have a mechanism to limit transactions by a specific vendor with a max $'s per month or requiring you to approve all transactions with a specific vendor via text/email/phone confirmation.

[azure jasper]

wait you have sonar 2?

[torpid fossil]

It's the new baseline. At least for pro users.

I've got it

[hardy crag]

Does the default computer model being used update automatically within existing threads/tasks? Or do I have to start a new task to switch to GPT 5.5?

[cinder juniper]

I've also realised that. It is doing Researches on Steroids. Sonar 2 is actually a pretty good model and I didn't expect it to be that good!

[azure jasper]

i mean im pro and i only have sonar 1, did it appear automatically to you?

[hardy roost]

I am seeing it on desktop only, not mobile.

[green crystal]

@distant aurora grok is crap now
not even thinking mode available for free

[turbid summit]

Out of usage on pro

[shell glacier]

shortcuts are gone and became Skills. 🙁

[heavy portal]

Not exactly the same as shortcuts, but using a text-snippet tool is close enough for me.

[shell glacier]

save shortcuts (now:Skills) as .md files, and reuse them back in Assistant.

[shell glacier]

no more shortcuts, and now become "Skills".

[distant aurora]

Yep. Gone. 😭. Maybe arena ai still has it.

Grok ain't worth 30 dollars.

Where is the heavy user load even coming from when no ones using it.

[golden birch]

Bro, I am really considering to pay which ai whats your choice?
ChatGPT, Gemini, or Claude

I found that both ChatGPT and Claude are token hogs

Whereas Gemini is slow and lacks connector choices.

My tasks:

Study
Agents
Coding

[primal junco]

skills can only be used with computer?

[heavy portal]

correct

[primal junco]

and shortcuts are completely gone

L

workflow just got fried if i dont pay for computer credits

[heavy portal]

I've moved to a text-snippet app that pastes my former shortcuts into the prompt bar

[tranquil lodge]

I feel like pro search has degraded lol why is mine searching arabic newspapers for a software related query xd

[primal junco]

ew

[golden birch]

Perplexity should add affordable models like Gemini 3.1 Flash, GPT Minis, Claude Haikus, and others, providing higher limits for these models to better serve most professional users.

[golden birch]

same

file limits fucking me up!

[primal junco]

time to switch

its just soo good though

[golden birch]

any alterantives?

[primal junco]

ive been thinking about going back to chatgpt, but idkkkkkkk

[golden birch]

chatgpt really great?

Planning to paid plus

I tried gemini slow as fuck

I tried Claude token killer

[tranquil lodge]

What is any of this gotta do with Software lmfao

[prime oar]

i have gemini pro plan

[golden birch]

It is fast?

[prime oar]

yeah gemini is fast

[golden birch]

but no notion connectors

:<

[primal junco]

yall try codex yet?

[tranquil lodge]

Now it's reading articles on Hamood Habibi 😭

[prime oar]

but its ok for me

[golden birch]

No usage limit is the best-selling point

[primal junco]

searching far and wide for ur shit lmaooo

[prime oar]

yeah u can use unlimited

[golden birch]

One more thing to add that I need agents for work now.

Idk which one is better claude code, codex or n8n apis

[primal junco]

n8n too much work

[tranquil lodge]

bruh they gotta fire the vibe coding skids from their dev team lol

[golden birch]

I am really curious does this group has perplexity devs?

[bright vigil]

Hi, why do connectors linked through pipedream ignore the account you're actually authorizing and defaults to the google account you registered with?

Im trying to link a google account for sheets but it's just defaulting to another email.
Connecting google drive worked

[primal junco]

cuz its vibe coded poop

[tranquil lodge]

Doubt that even most mods are just part of the end group testing

[bright vigil]

why would pipedream even bother asking me to sign in with my other account then?

#1484130405526405120 message

ah someone says they don't work at all?

can anyone confirm so that I don't waste my time trying to connect

[primal junco]

mine doesnt work either

just tried

[bright vigil]

thank you

❤️

so apparently it works, but not with deep research (defeats the point then no?) and a workaround is sharing the file with editor privileges

[hasty helm]

Any suggestions for flood planing/ interior design from pictures?

[cerulean crag]

.

[torpid fossil]

It just showed up on my pc. Maybe it was a rollout. I use comet

[bright vigil]

is there a way to see how many prompts I have left?

[brittle dagger]

i would prefer if he just straight up admitted that "benchmarks are bullshit, existing solely to trick people" but hey, better than nothing.

can't have honesty, of course. gotta keep bullshitting people instead.

"But Sol! Benchmarks are important! They allow for comparing the models!" ... no, they don't.
99%+ of people have no clue whatsoever and these people know that.
they are fully aware of the fact that 99%+ of people react like "ooooohhhh number is better!" and that's the limit of their cognitive abilities.

this is everywhere. people have been fooled for so long, they think that the way things work are how they should work.
thus they fall for bullshit like buying into "iterations" like when some tech thingy minimally improves.

That's why Valve doesn't do that, with, for example, HalfLife or the Steam Deck.

Gabe isn't a greedy asshole and doesn't run a public company.
He doesn't want to and isn't forced into bullshitting people and, thus, doesn't pull off shit like this. Instead, what matters to him,
is that successors are actual successors and not just iterations with overhyped minimal differences.

To put that into words more people understand:

Dopamine addicted brains are intentionally caused and then used against you.

...
You're welcome.

[bright vigil]

in other words, the sky is blue

[brittle dagger]

Technically speaking there is no sky in the sense most people think, (to stay true to the point I was making)
but ... of course ... you're not wrong. :)

[hybrid halo]

When Perplexity will add Deepseek V4 Pro it is cheap and stronger than Nemotron 3 ?

[mighty ridge]

anyone else facing issue of audio not working of youtube in comet on windows? both block ads on/off, the issue is there

[heavy portal]

any chance that site might be muted (right click it's tab)?

[mighty ridge]

No nothing like that, audio is fine on brave or chrome but the problem is only on comet

[heavy portal]

does audio work on other sites in Comet besides YouTube (i.e., Dailymotion)?

[mighty ridge]

The audio doesn't work on any site

[heavy portal]

how about Comet in incognito?

[mighty ridge]

It doesn't work

[heavy portal]

Some operating systems (Win10/11) has a way to adjust volume or mute specific apps.

[tranquil orchid]

@kssd8e muted

Reason: 9. No spamming.
Duration: 4 minutes and 49 seconds

[mighty ridge]

No, I checked all these things, the problem is only on comet

[heavy portal]

[e-mail]

[green crown]

why didn't i receive any pro computer credits? i'm on pro plan

[heavy portal]

paid Pro plan, or via promo?

[green crown]

promo

we don't get them?

[heavy portal]

that's my understanding

even the paid Pro only got a one-time bonus of 2000 or 4000

[green crown]

awh
i don't know if i'll renew, considering how much of the pro plan's limits went towards the max plan

[green falcon]

Hello

[sour lynx]

hi

[green falcon]

Wanted to ask a question about sponsorship

Does Perplexity sponsors Tech events?

[heavy portal]

Maybe ask [e-mail] about that

[green falcon]

Got an AI generated E-mail, it says contact through Discord

[heavy portal]

heh, than I have no idea. Maybe https://discord.com/channels/1047197230748151888/1059504969386037258 to promo the tech event.

[odd pelican]

Hii

[distant aurora]

@green crystal https://www.vxinstagram.com/reel/DXmWMAYk77z

[green crystal]

code base went through 4 generations of models lmao

[subtle violet]

hi

i am new

[gloomy cipher]

@dry kestrel @distant aurora yoooo gpt 5.5 is soo tuff

sadly the extra high consumes a lot of usage on my token plan :c

[flat geode]

Since when am I not able to use my Perplexity pro? It tells me I don’t have enough tokens.. everything I’ve used it for is now not available to me. I’ve checked my account, etc —-and where do I find out available tokens? I really love Perplexity but now I’ve had to go to other sources.

[gloomy cipher]

Isn't token for computer use only

[twin grail]

U turned on computer mode dude…

[twin grail]

Imma check

[flat geode]

I’m amateur

[twin grail]

The token consumption

Tmr

But

Just burned

280m token

Claude opus 4.7

[gloomy cipher]

Sending hi to the computer mode 🖐️

[gloomy cipher]

Computer mode is basically openclaw don't use it if it's not heavy task or very web related

[flat geode]

IOS —I’m able to access outliers ai which is without as many limits and uses multiple models

[twin grail]

no worries

[short narwhal]

Use low or medium!!

[twin grail]

just exit the computer mode

should be fine

[gloomy cipher]

Yes but brain stuff

Also its for codex

I hate how i don't get to see how the usage actually works

[short narwhal]

Still

[gloomy cipher]

So that i can optimize for model

[short narwhal]

Maybe high then but extra high is useless

[twin grail]

today consumption for 2 hours

of vibe coding

[twin grail]

simple they changed it

[dry kestrel]

yes

[gloomy cipher]

Yea iam thinking like only doing extra high for stuff like refactoring source code or the starting a new project

[dry kestrel]

very good at 90% of things

[twin grail]

the longer the reasoning

[dry kestrel]

dont use xhigh

use med

[twin grail]

the more usage u will consume

so

[gloomy cipher]

Time based?

[twin grail]

split the task

[twin grail]

yes

[gloomy cipher]

But that doesn't make sense

What about different models

[dry kestrel]

no

[twin grail]

time based x model reasoning effort

[dry kestrel]

it's token based

[twin grail]

I think

u missed it

they changed it

[gloomy cipher]

We are talking about codex fyi

[twin grail]

a few weeks back

[dry kestrel]

token based

[twin grail]

being the longer u reasoning

[short narwhal]

Even for that like high is more than enough. Probably even that can be overkill

[twin grail]

u'll consume more

[dry kestrel]

longer reasoning = more tokens...

and then you factor in things like compaction mid way

[twin grail]

oh ye...

sorry

it's

3 am here

my brain

is

kinda

yk

[gloomy cipher]

What is the cheapest run among codex models? I wana use one for commit messages so it's generally gonna be the cheapest one

[dry kestrel]

use GPT-5.5 on med

[gloomy cipher]

....

[dry kestrel]

and it it's really simple things GPT-5.4 mini on high

[gloomy cipher]

No cheapest among codex model

[twin grail]

my best advice if u are on plus plan

go for gpt 5.4 + med

[gloomy cipher]

I can see 5.3 codex and 5.4 mini

[twin grail]

should save u more

than using gpt 5.5

[gloomy cipher]

Pro

[dry kestrel]

then why are you worrying about usage

[twin grail]

Then why hesitate

I'm also on pro

[dry kestrel]

just always use 5.5 med

[twin grail]

Also pro

[gloomy cipher]

I just did a few changes today and iam already down 10% on the weekly

But it was eztra high hehe

[twin grail]

why worries about it

[dry kestrel]

whats your config looking like

[twin grail]

uhh

[dry kestrel]

turn that off

[twin grail]

what plan

[dry kestrel]

use normal mode instead of /fast

[twin grail]

are u on

[dry kestrel]

and turn off chronicle memory

[twin grail]

if u are on

pro 5x

then u should use med to high

xhigh consume like crazy

even for me

[gloomy cipher]

For what? It's like 7 mcp if that matters and a lot of flutter skills prob 10

[dry kestrel]

It doesn't make any sense to use xhigh no matter the plan

[twin grail]

what pro plan

are u on

[dry kestrel]

your compaction settings, speed setting, memories, etc

[twin grail]

it kinda depends like if u just want to burn it

like me

[dry kestrel]

all of those consume usage

[twin grail]

i think it should be enough

[dry kestrel]

no point in minmaxxing usage

and you get worse performance on xhigh

[twin grail]

wait

is that true ?

[dry kestrel]

yes for the majority of the time

[twin grail]

i see

might aswell

switch to

medium

[dry kestrel]

and you waste so much time

Just use med and high

[gloomy cipher]

Actually i found the best practices to be i load my github on perplexity and ask the sonnet sonnet hey how does this [famous open-source project sharing the same packages with me] handles this problem and how does our project handles the problem. Sonnet points out to everything then i ask sonnet to write a prompt/task for the next agent to implement these changes which i send to my codex

[jade abyss]

What’s it like on IOS? In terms of what mode it uses etc. it doesn’t specifically say

I assume all of this is a lot easier to see when I’m on PC

[gloomy cipher]

@dry kestrel @twin grail can codex not use multi agents by itself? To solve over contesting issue while doings multiple things for a bug I would ask kimi to use as many agents as it needs and instead of doing that code by herself. She would spawn another agent and give her all the information then which makes it convenient when you want your coding agent to start with the least amount of context for a bug fix or adding feature

[delicate vector]

Does anyone know something about the new "Sonar 2"? Just seen this. There is little to no information about it, excluding the pricing on it on perplexity...

[short narwhal]

https://x.com/aravsrinivas/status/2047019688920756504 this probably
-# also please ignore that other person + their DMs 🙏

Aravind Srinivas (@AravSrinivas)

We’ve post trained a model on top of Qwen that achieves Pareto optimality on accuracy-cost curves.

Unlike our previous post trained models, this model has been trained to be good at search and tool calls simultaneously, allowing us to unify the tool call router and

X · Apr 22, 2026, 6:28 PM

[delicate vector]

Thanks appreciate it!

So its just essentially the upgraded Qwen3.5 397B model with perplexity's search and tool calls training

[tawdry cobalt]

hello. so i have perplexity pro, and I've been uploading files all day, becuase I am studying for my upcoming exams. However, all of a sudden, I have to upgrade to Max??!!!??? So i wait a bit, and then after 30 mins, I am able to upload files again. I go and upload 2 files, only 1 uploads. Then, when I remove the 1 uploaded file and go to upload a different file, it doesnt work and it says I need to upgrade to max!!! please someone help!

[steep oracle]

They are way to many ads for the computer function on the application.

(and for max, even though I paid for pro)

[warm crag]

Hello I have perplexity pro and I ask maybe 3 questions a week, all of a sudden I’m out of credits and it’s asking to buy more or upgrade to max, that my allocation comes back in 27 days?

[warm crag]

Ugh never mind mine was stuck on compute for some reason, they should make that more clear when querying

[green crystal]

curious why not kimi

cost?

[dry kestrel]

tell it to use many sub agents

[green crystal]

and tell it to keep a record of what the agents do if agents need to communicate
a few times i got them to accidentally read eachother's file outputs during work and they got confused lmao

[dry kestrel]

yeah

[snow iron]

how do i fix

[distant aurora]

Deepseek v4 pro is underrated.

@green crystal https://cunnyx.com/i/status/2047716157491179526

Taelin (@VictorTaelin)

Opus thinking: "oh fuck this is my fault"
︀︀
︀︀Opus responding: "this is YOUR fault"
︀︀
︀︀???????????????????????

**💬 101 🔁 55 ❤️ 2.5K 👁️ 131.7K **

CunnyX · Apr 24, 2026, 4:35 PM

[green crystal]

agi moment right there

[green crystal]

it tends to miss stuff tho

kimi has a better time with juggling info

[lost jetty]

http://chat.deepseek.com
can we download the web app for our windows desktop from this link?

[distant aurora]

Bro

Getting constantly bitten by mosquitoes at a hospital is scary af.

God knows what diseases they might be carrying.

[green crystal]

their digestive system immediately cleans blood they suck

[lost jetty]

what happened?

[distant aurora]

I don't think so. If that were true then malaria wouldn't be transmitted.

[distant aurora]

Some stupid goverment hospital.

Won't let me make opd file without abha account.

Also

This country is a shit hole. I was first in the queue of seats. Came an hour early. But at the OPD counter, 20 people suddenly came and stood ahead of me 15 minutes earlier.

So these people crowded the counter and were staring at me despite me being ahead of them.

And the cops did nothing.

Can't even get affordable healthcare at a goverment hospital here.

[odd pivot]

you can always pretend to have something and then have them find it there must be protocol around this that works

but, these are literally the vestiges of capitalism

like the hot soup every day

[tall ocean]

how are you guys able to unable computer?

[heavy portal]

hover mouse over left edge of Computer and click the X

You'll know it's removed when a plus sign appears to the right of "Computer" (yes, it's subtle)

[gentle raptor]

Where can I find my billing records? I can only see the credit cards I've set up.

[sonic pilot]

A minor issue I know, but is there a way to change it in the Android app so that hitting "Enter," does not send the message?

[worthy verge]

Guys, why Perplexity doesn’t refresh my uploads? I had 50 uploads per day, but 2 days ago was my first time using all of them so I ran out, but 2 days later today the uploads are just fineeee, how do I resolve this?

[hushed urchin]

holy fuck guys

this is it

now they're removing access to choose your own models

I have to choose 'best' for the rest of the month now

said I've reached my limit of the 'premium models'

there is NOTHING left for the pro tier 😆

[silk atlas]

Yo

[silk atlas]

Really, I've not updated my perplexity max, applications

How can I get, api of perplexity, I'm using perplexity max

[hallow sand]

yesterday I rated perplexity with Claude 4.6 1++. it was awesome on enhancing my code. No Cat Limit, no truncate of code for the LLM. And today it acts like a very stupid child in the same thread, broke my code compleately. I always set LLM to Claude 4.6 and the change after each Question reverts it back to "Best". Perplexity says it will stay last chosen for the thread but that can not be the case. Frustrates me as hell. If we reach Chat Limitations let us know, don't let us run into trouble.

[split dome]

Deepseek V4 expert ???

[short root]

hi but what is comet coded in

[gloomy cipher]

Availability is bad

And no subscription

Raw api costs a lot

[dull grove]

good question

[silent token]

Hello

I would like to ask if anyone has had a good experience with the nvidia AI

@halcyon coral , can I have the pro role pls?

[halcyon coral]

Hi, please check the pinned message

[silent token]

Thanks

sry for the trouble

[dull grove]

good but not the best (from my pov)

[brittle dagger]

I wish I could turn off advanced searches and all this crap that's being forced onto me. All these assumptions about what the user MUST DEFINTELY WANT TO HAVE are annoying as hell, introducing variations nobody actually deliberately asks for.

I don't mind features, I mind having shit forced onto me.

rant rant rant grrrrrrrrr

sorry. 😊

[silent token]

To be fair, I've only been using Claude

And was curious about the other options

[brittle dagger]

in 99% of all cases it's the user's fault.

whatever "it" is referring to.

[dull grove]

they are all good some are just more specialized in certain areas than others

[twin grail]

just say i summon subagents

or u can just give it instruction being must use subagents everytime

[twin grail]

note that

this consume

ur usage way much faster

[keen meadow]

Anyone in Canada getting a hard coded denial for perplexity not creating images while on Pro Tier?

[keen meadow]

not supported in this region

[flint gull]

what is sonar 2

[brittle dagger]

...
Anyone, who can figure out if I can give away those 2k free credits for Computer, which I've received for my Pro sub, can have them.
...

Within the rules, of course.

I have no use for Computer. Anything I would want my language models to do I'd implement myself. Much safer, far less trust issues.

[lost jetty]

Damn!

Which city?

[lost jetty]

is this a hallucination?

[split dome]

Idk .. but that's what i tried

[brittle dagger]

I'd first correct its wording. It already starts wrong: "We are asked".

Then make it stop speculating about intent and instead enforce strictness.

There's far too much speculation going on.

[brittle dagger]

Apply first person perspective, enforce taking the user literally, enforce removal of all speculation -> Increase intelligence and quality of responses.
Or ... well ... have the model behave equally to its users.

[lost jetty]

chatgpt hallucination on it's competitor 💀

[green crystal]

That doubt comes from a wrong assumption: mosquitoes are not dirty flying syringes. They don’t automatically pass whatever blood disease one person has into the next person.

For a mosquito to transmit a disease, the germ usually has to do a very specific sequence:

1. The mosquito bites an infected person or animal.

2. The germ survives inside the mosquito.

3. It crosses from the mosquito’s gut into its body.

4. It multiplies or develops there.

5. It reaches the mosquito’s salivary glands.

6. During a later bite, it enters the next person through mosquito saliva.

CDC says mosquitoes do not transmit all kinds of germs, only ones that can grow or multiply inside the mosquito and move to the salivary glands. That process can take weeks.

So yes, mosquitoes really do transmit some diseases. Malaria works because Plasmodium, the malaria parasite, has part of its life cycle inside female Anopheles mosquitoes. The mosquito later injects the parasite through saliva when it bites.

And it’s not only malaria. Some viruses are mosquito-borne too, like dengue, Zika, chikungunya, yellow fever, West Nile fever, and Japanese encephalitis. WHO lists these as vector-borne diseases, and says dengue is a major viral infection transmitted by Aedes mosquitoes.

But many diseases cannot spread through mosquitoes. HIV is the classic example. HIV.gov explicitly says HIV is not spread by mosquitoes, ticks, or other insects. CDC also notes HIV cannot reproduce outside a human host, which is one reason it doesn’t use mosquitoes as a transmission route.

The clean answer is:

Mosquitoes transmit only pathogens that are biologically adapted to survive, develop, and reach the mosquito’s saliva. Malaria and dengue can. HIV and most ordinary human infections can’t.

Wow the response looks so short in chatgpt app lol

[green crystal]

That's gpt 4o without web search lol...

[lost jetty]

bruh
i understand that it was a voice chat ... but ... what it told me in the next few prompts was utter brain rot
: (