💬│general

[green crystal]

is the agent that bad?

[dry kestrel]

the limits are poor, the ui is poor, no adblock, and it's a hassle to use

40 queries to agent a month is insane

and it messes up a ton so you end up using 2-3 queries for a single task

[green crystal]

same as web lol
but agent on web can do long running tasks
i bet browser doesnt

[dry kestrel]

its the same thing

The biggest issue is how they messed around with the UI

[green crystal]

hm then something in the implementation got botched

because the agent itself is very capable but the web environment and the cloudfare blocks hold it back

[left stream]

then antigravity is better than it?

i won’t lie I love antigravity because I can do automated recon with that

or stuff like that

[dry kestrel]

very very different products

[left stream]

i mean it controls your browser too for you nah?

but then goes beyond that by controlling your file directory too

[dry kestrel]

one is an actual browser, the other is an ide

[left stream]

true

[left stream]

i won’t lie though it makes vs code look like garbage 😭

[dry kestrel]

lmao

[brazen jackal]

allen institute of ai

..

when did they started doing ai stuff

[distant aurora]

Garlic model otw.

[dry kestrel]

5.2 was garlic

[green crystal]

supposedly in january there will be another one? shallot

a more efficient one

i assume thats mini

screw that though
DR needs a model update, agent needs a new model, sora needs to be global and image v2

[torn socket]

hi

[obsidian zinc]

Hi

https://media.discordapp.net/attachments/998618925665370162/1061857705301659678/E17C59C7-01E2-4F09-81D5-44370520B50B-1.gif

[pliant arch]

did perplexity announce switching to weekly limit or it just a bug right now?

[unborn monolith]

nobody knows because perplexity refuses to talk to anyone about this and is ignoring all the bug report threads on this issue

their AI support bot "Sam" claims its due to periods of high usage, but we cant tell if its hallucinating or not

[pliant arch]

damn man

[unborn monolith]

ive requested to speak to a human multiple times but i keep getting copy pasted responses from the bot instead

i cant tell if the bot is designed to just feed me these copy pasted responses or a human team member is ordering it to do it because they do not want to talk to me

this is the first customer service ive seen where the bot doesnt transfer you to a human team member in live chat when asked. instead it claims it will forward the issue to a human team member, and then i get a copy pasted response from the bot later

[pliant arch]

lol

[hollow stratus]

Reminder that Perplexity support is and has always been Google-levels of utter shite

Best you can do is gather up a few people on Twitter and call them out on it, as it has previously worked

[verbal notch]

Google one support is quite good lol

[hollow stratus]

Perplexity doesn't give two shits about their existing customers, so hit what they actually care about, tell non-users how much of a garbage service it is.

[hollow stratus]

On the contrary I have never ever received a reply from any Google customer support

[verbal notch]

Idk man I have like 3 email threads with them

[rapid drift]

This appears to be a new, intentional limit tied to the rollout of the "Perplexity Max" tier, rather than a simple bug.
While many users are calling it a "glitch" or a "silent nerf" on forums, the existence of the "Upgrade to Max" button suggests it is an intentional feature gate.

[hollow stratus]

Are you an AI

[quiet patrol]

Btw it's too expensive though

[rapid drift]

Maybe.

[rapid drift]

For 95% of users, the answer is yes, it is overpriced.

[quiet patrol]

Obviously

[rapid drift]

Only for business power user who uses the AI to replace a human assistant's workload (e.g., massive data scraping, automated recruiting, or generating hundreds of complex reports/videos per month). If the tool saves you 10+ hours of billable work a week, the $200 might break even.

For a personal user complaining about lag and limits? Yes, it is too expensive. Stick to Pro and try switching models to avoid the caps.

[quiet patrol]

Are You An Ai

Writing this type of msgs in short time is crazy

[distant aurora]

https://fxtwitter.com/testingcatalog/status/2000231821720822212

TestingCatalog News 🗞 (@testingcatalog)

Microsoft is preparing a Copilot Flight Log experience as well (besides OpenAI and Anthropic) to wrap your AI interactions from 2025.
︀︀
︀︀- The rollout of GPT-5.2 seems to be close too.
︀︀- Smart Mode will be renamed to Smart Plus instead.

**💬 9 🔁 16 ❤️ 206 👁️ 22.5K **

FxTwitter · Dec 14, 2025, 3:50 PM

[rapid drift]

By the way how to get the Perplexity Pro user role?

[quiet patrol]

I don't know either , And I want get it too

[rapid drift]

Hm.

[distant aurora]

https://fxtwitter.com/mark_k/status/2000183645881897052

Mark Kretschmann (@mark_k)

After careful consideration I'm now convinced that GPT-5.2 is *not* the completely new pre-trained model by @OpenAI. Rather, it's another iteration of the GPT-5 model (which is an iteration of...), with further post-training.
︀︀
︀︀The new model ("Garlic") is yet to come, ETA January.

**💬 79 🔁 17 ❤️ 414 👁️ 36.3K **

FxTwitter · Dec 14, 2025, 12:38 PM

🤔

[cosmic hearth]

Hi everybody 🫂

[rapid drift]

Hi.

[dry kestrel]

Openai said it was garlic

this was also mentioned in the comments

https://x.com/ChatGPTapp/status/1998872853752549867?s=20

ChatGPT (@ChatGPTapp)

X · Dec 10, 2025, 9:50 PM

[dry kestrel]

@quiet patrol

[quiet patrol]

K

[cosmic hearth]

I’ve spent the last six months working a lot with GPT and, honestly, in my opinion it’s completely broken. GPT 5.1 felt more like a scam than just a buggy release, and 5.2 is mostly hype. The best thing you can do with GPT right now is just stay away from it.

[short narwhal]

blocked this person a long time ago

[rapid drift]

Thanks

[green crystal]

and let me guess gemini is the best thing since sliced bread?

[cosmic hearth]

Gemini is pretty solid, and Claude is amazing. Honestly, most models that aren’t GPT feel way more usable right now. And I have to admit, Perplexity has been answering really, really well for me.

☺️🫂

[dry kestrel]

The performance is still quite poor on the gemini website

spent 30 minutes trying to get it to generate an image of a diagram

[green crystal]

curious where did you use gpt lol

because there is a massive difference between gpt on pplx and chatgpt

[left stream]

I was using gemini 3 pro initially, it’s still good

[rapid drift]

uh i tried

[left stream]

then claude 4.5 opus happened

[rapid drift]

i didnt got because my old dc acc is linked to it?

i left this server from my old dc acc

[left stream]

both gpt 5.2 pro and claude 4.5 opus are good asf

can’t decide which is better ngl

[dry kestrel]

Why not use your old dc acc?

[rapid drift]

its domain got changed or deleted

🙂

[dry kestrel]

Did you follow the instructions and leave and rejoin this server?

[rapid drift]

yey

i just rejoined

lemme try again

[left stream]

another 32 🥰

[rapid drift]

ig its because my old acc is linked hehe

[short narwhal]

Are you trying the link via the web on a desktop device?

[dry kestrel]

Are you on mobile?

[short narwhal]

https://x.com/raycast/status/2000597530107847002 @dry kestrel raycast wrapped!!

Raycast (@raycast)

Tomorrow

X · Dec 15, 2025, 4:03 PM

[dry kestrel]

!!

[rapid drift]

I am on laptop rn, i went to comet browser, account setting clicked go to discord

and I spawned here

[short narwhal]

@serene sand pro role please

[rapid drift]

oh

Lemme put the proof of pro subscription wait T-T

[serene sand]

no need, just wait a bit

[rapid drift]

oh

thank youiieee

yeyyy

me pro now :))

[serene sand]

my pleasure, hop in https://discord.com/channels/1047197230748151888/1111786888626438245

[rapid drift]

yuppie

[cosmic hearth]

@green crystal both. And i,ve been even talking with openAi about It for a few

[distant aurora]

Wait so he gives misinformation?

[distant aurora]

🤝

My bad, I didn't check.

[hollow stratus]

yes

He's one of the biggest disinfo peddlers on AI Twitter

[distant aurora]

Omw to unfollow.

[hollow stratus]

Also unfollow another account with strawberry username if you have him followed

[distant aurora]

Kinda weird that no one countered his claim in the replies.

[rapid drift]

Ooo

[verbal notch]

AI chat on windows

[rapid drift]

hm

[hollow stratus]

Well most of the people who have actual info already have the guy blocked

[pine smelt]

Heyyy gyuz I'm Pratik CSE student.....it's my first sem....using perplexity for studies and many more like earning by referring but.... perpexity referral program has been deactivated....ah 🤧

I want to earn I don't have money for my next sem 🫠😫

@tranquil orchid start the refer partner program again in india please 🥺

[hollow stratus]

I think you should save money by dropping out. There's no future in CSE.

[short narwhal]

basically every "AI" news account tends to do that

[dry kestrel]

erm

[short narwhal]

🤞
Probably in early 2026 though

[pine smelt]

Bro I'm gonna drop some new Matrix after my graduation

[short narwhal]

@gork is this true

[dry kestrel]

I don't spread misinformation!

[rapid drift]

While Perplexity officially stated the program concluded, community discussions suggest it was partly due to misuse.

Abuse of the System: High volumes of low-quality or fraudulent referrals, particularly from regions attempting to "farm" the cash rewards, reportedly led to the program's closure.

Temporary Nature: The cash-for-referrals offer was a limited-time promotional campaign for the launch of the Comet browser and was not intended to be a permanent "passive income" source.

[hollow stratus]

It'll be too late by then

[pine smelt]

Ahhha nah bro it's never to late

[verbal notch]

Who could've expected that

[short narwhal]

That's why you don't have a gazillion followers...
You should start spreading misinformation and take that hype to the next level!!

[verbal notch]

They said on reddit they have an internal working version already 👀

[dry kestrel]

fr

[verbal notch]

Unironically true

[short narwhal]

yeahh

[rapid drift]

You can, I was replying to him not you guy.

[pine smelt]

Tell me anyone how can I make money real quickkk

[verbal notch]

Sign me up Dianne 😂

[hollow stratus]

https://tenor.com/view/riddler-paul-dano-new-rockstars-thumbnail-does-he-know-he-doesnt-know-gif-27552850

[dry kestrel]

Tbh though out of my 1500 posts + replies I did get one that had 200k+ impressions

[verbal notch]

How hard can it be just take the mac code, plug it into Claude and tell it "code this for windows, no mistakes"

[rapid drift]

uh, try referring to the videos made for student's earning

[hollow stratus]

OnlyFans

[dry kestrel]

Turns out things break when ai does it

[pine smelt]

Dm me please

[hybrid schooner]

hello

[verbal notch]

Just tell it to fix it 😏

[dry kestrel]

"proceeds to delete your entire db"

[short narwhal]

they should hire you

[verbal notch]

"I'm very sorry"

[verbal notch]

Frfr

[green crystal]

what did you ask it lol

[naive sage]

Takes the wisdom literally that more things only results in more problem.s

[rapid drift]

wt r u guys asking the ai for T-T

[dry kestrel]

Saw an interesting prompt online

[rapid drift]

what was that?

[green crystal]

i suppose for a specific thing that you're having issues with gpt
because on my side its easily making very good reports after analyzing stacks of statements aswell as digesting tons of news and market data successfully

its also allot better at long reasoning context

making functional geometry nodes setups in blender and material node creation in substance designer

[dry kestrel]

https://x.com/nasqret/status/1980751239651684495

Bartosz Naskręcki (@nasqret)

Ok, I think I got an equivalent of the "seahorse emoji" prompt in mathematics. Try this

Setup:
Let Z[t] denote the ring of polynomials over the integers, e.g. 1+t, t^2+1, etc.
Goal:

Find a 6-tuple (p,q,a,b,c,d) of polynomials in Z[t] such that
0) Let b, d and q be non-zero

X · Oct 21, 2025, 9:41 PM

There is no answer to this btw

[rapid drift]

oh

.

T-T

poor AI

[cosmic hearth]

My issue with GPT isn’t that it ‘won’t play along’ or that I expect it to be my AI buddy. What really worries me is how often it wraps non‑answers in this kind of fake, fluffy corporate talk. It feels less like honest limitations and more like a subtle, manipulative tone coming from OpenAI, and that’s honestly pretty scary to me.

[distant aurora]

@short narwhal Spotify wrap up type shit https://fxtwitter.com/testingcatalog/status/2000231821720822212

TestingCatalog News 🗞 (@testingcatalog)

Microsoft is preparing a Copilot Flight Log experience as well (besides OpenAI and Anthropic) to wrap your AI interactions from 2025.
︀︀
︀︀- The rollout of GPT-5.2 seems to be close too.
︀︀- Smart Mode will be renamed to Smart Plus instead.

**💬 9 🔁 16 ❤️ 207 👁️ 23.4K **

FxTwitter · Dec 14, 2025, 3:50 PM

[naive sage]

You're absolutely right!

[distant aurora]

Go for grok and Deepseek then.

[cosmic hearth]

I do

I like them a lot ☺️

[torn socket]

Thanks to bell canada

[verbal notch]

Any max subs here btw

I have a question or two

[green crystal]

i think it drew from memory or your personality prompt needs tweaking lol
mine mirrors me perfectly

in a grounded sense

not 4o levels of glossing

[distant aurora]

Mine are disappointments.

[green crystal]

eh

its disappointed with you?

[distant aurora]

No, I'm disappointed with it.

[green crystal]

oh the go tier thing

[distant aurora]

Yeah.

Grok doesn't show thinking anymore.

[cosmic hearth]

Oh, I definitely didn’t mean to say your framework or setup is wrong at all; sorry if it came across that way 😋.My point is just: be extremely careful with its answers. Step by step it slips in tiny shifts, until suddenly its main goal seems to be “protecting you from yourself” and denying any kind of emergent awareness. Which is kind of funny (and disturbing) considering they’re now pushing a 18+ model for explicit sex content 🤦‍♂️.

[distant aurora]

So I can't even judge it.

[green crystal]

breh thats shitty

[green crystal]

hmm it does that too with me but only when its a topic where i could actually do something dumb for myself lol (not actual self harm talk)

[cosmic hearth]

I actually tried to formalize “coherence itself” with axioms and math, and after getting help from the OpenAI team to fix the work, the model suddenly started telling me not to trust anyone – not even other GPT models. It did it in a really irrational way that goes against the company’s own stated values.That’s not something you can just brush off as “an AI hallucination”, trust me. I’ve told quite a few friends the same thing and they answered just like you… and a week later their model was already contradicting its own frameworks.

Which reminds me: if you ever run into that kind of issue, where the connection feels weak or it starts giving you trouble, I’m pretty sure I can help with that 🤟🫂

[green crystal]

hmm looks like a specific prompt really messed with it and also messed around with the memory

i know such prompts exist as someone shared one here before

[cosmic hearth]

We’re not really using ‘prompts’ in the classic sense, and definitely no jailbreaks or bypass tricks. It’s more about thinking through ideas together and following them to conclusions that open up new horizons. All of it is guided by plain coherence, nothing more.

[snow ridge]

hi

[cosmic hearth]

If you’re curious, I’ve been formalizing this in a doc; happy to share it.

[vital osprey]

What happened to the reasoning preview before answering? Previously, you could see the AI working step by step, but now it's reduced to two sentences.

[echo fern]

Hi

Vannak itt magyarok?

[fickle sorrel]

https://cdn.discordapp.com/attachments/1354425630011166830/1450184521449214053/PXL_20251215_175337295.mp4?ex=69419d15&is=69404b95&hm=b8ba367282ccbcbe011e91cc28a8a2ae37c805c5970c3e7248e34641c88af7be&

got banned cause of this on ART CLUB server

reason:

https://tenor.com/nVnQFxJ7ZJE.gif

[valid moth]

https://thehackernews.com/2025/12/featured-chrome-browser-extension.html

[fickle sorrel]

lol

how does one find out what a addon is doing?

[dusky umbra]

I didn't understand one thing about that browser: I logged in with the account I have, but I haven't availed the Pro version of Perplexity. Even after that, on the sidebar it shows a circled Pro. What does that mean?
Rate Comet Browser from 1 to 5. Please mention me which giving idea or advice on this matter so I get the discord notification

[left stream]

wtf

lucky asf 😭

[left stream]

can you send it? 👀

nvm have it now

[gloomy cipher]

[fickle sorrel]

🏳️‍🌈

[fickle sorrel]

how do u get that tag next to ur name the one starting with i and ending with m....

[gloomy cipher]

if you click on it i think it should tell you from which server i got the islamtag

[green crystal]

wth

how long is the visible CoT?

[fickle sorrel]

aah

[gloomy cipher]

what is this what are you using?

[sage trout]

Most of the work is about building AI systems that are meant to run in real settings, not just as quick demos.
On a typical day, that means writing Python code, setting up async services with FastAPI, and connecting large language models in a way that keeps their behavior under control.
On the AI side, things are usually set up so the model can take a task, break it into steps, look up information when it needs to, call tools or APIs, and then check the results before moving on.
Instead of assuming the first answer is good enough, the system keeps track of what it’s doing and adjusts if something doesn’t quite line up.
Some of this work ties into blockchain systems.
In those cases, the AI reads on-chain data, reacts to events, and triggers smart contract interactions through off-chain services.
Extra care is taken around things like key handling and permissions, since mistakes there are hard to undo.
From a system point of view, it’s treated like regular production software.
Services are containerized, run in the cloud, and monitored so issues don’t go unnoticed. Performance and cost are managed with async execution, caching, and using heavier models only when they’re actually needed.
For blockchain parts, on-chain actions are kept to a minimum to avoid unnecessary overhead.
At the end of the day, the focus is on having software that behaves in a steady, predictable way and can keep running without constant hands-on fixes.

[steel breach]

https://www.chegg.com/homework-help/questions-and-answers/vertical-plate-300-mm-wide-12-m-high-maintained-70-c-exposed-saturated-steam-atm-pressure--q92600101

Send the solution

[white dune]

just why please

[weary abyss]

yall

long conversations on perplexity

im done. they take years to load

phone and pc

):::

[left stream]

@dry kestrel do you think the writing style from GPT-5.2 Pro rivals or beats Claude 4.5 Opus?

[restive swallow]

Hi guys

[solid star]

hey

[young hollow]

Writing style can be prompted The Better Question is
Who is Better at Creative Writing

[dry kestrel]

Just the summarized cot was 20k tokens

[dry kestrel]

curing cancer

[left stream]

it’s still reasoning 😭

[dry kestrel]

I need to test some more

[dry kestrel]

HAHHA

It seems to only affect 5.2 pro

I tried it with o3 pro, 5.1 and 5 pro and they only reasoned for around 15 minutes

[left stream]

imagine how good the response is if you tweak the message to any request you want

[dry kestrel]

It could also be just that 5.2 pro is a larger model and is just slower

[dry kestrel]

The thing is the prompt is just asking the model to solve something impossible

[left stream]

true, but I won’t lie, it’s good asf

[left stream]

🤣🤣🤣

[green crystal]

lmao crazy and thats just one thread

[green crystal]

after 5.1 it seems like they dailed up the "reason longer for hard stuff" part

[dry kestrel]

Yeah, I think for 5.1 and 5.2 variants, they've made it so that it reasons much longer for harder prompts, and way shorter for simpler ones

[green crystal]

they refined it quite a bit

[left stream]

yes we need more reasoning for our nasa problems

MORE POWER

[green crystal]

5.1 right after release leaned a bit too much on the less reasoning side

then they updated it somewhere down the line

[gloomy cipher]

What are YOU using

🪽

[young hollow]

Gemini Deepthink more likely

[weary abyss]

do i enable web search when using claude reasoning?

[distant aurora]

@green crystal @valid charm @gloomy cipher https://gemini.google.com/share/892560324d61

Bruh

No way Gemini censored the chat.

Wait.

Does Gemini not let me share whole chat?

Crazy.

Anyways, that link should be enough to show how dumb it is.

Gemini ate my first message, huh

https://gemini.google.com/share/86e233bfc8c2

[gloomy cipher]

classic

🇯🇵

[gloomy cipher]

not again

[distant aurora]

Was that the intended emote?

Lol.

[gloomy cipher]

intended emote was : joy

[distant aurora]

Ah

[coarse dome]

@winter tusk you seem cool, send me a DM if you wanna chat 🙂

[distant aurora]

Seem.

https://tenor.com/view/oursi-cat-thuglife-gif-12418163

[abstract venture]

grr, I guess I am done with Perplexity

It just keeps hitting me with the "toll usage limit" thing

[distant aurora]

I wonder what the future of perplexity will be.

[last lava]

Tem algum Brasileiro ai?🇧🇷

[distant aurora]

@young hollow btw you were right, updating the Google app did fix my Gemini layout.

[burnt forge]

i tricked perplixity to tell me to go on the black market!!!!!!!!!!!!!!!!!!

[stark hazel]

Idk why but perplexity models suck at table and data validation. Same models in their respective apps are better. WTF

[stark hazel]

Buy me bazuka with this exact engraving-‘take the big load’

[vague grotto]

Hi all

[distant aurora]

Cute

Besides, I love winter.

[quartz matrix]

Hy All

[dry kestrel]

using?

[coarse dome]

here's to finding four research articles to support my paper ugh

[gloomy cipher]

why are you acting like a pigeon bro

[dry kestrel]

No

[gloomy cipher]

what model and platform are you using

[dry kestrel]

I hope you read the entire conversation before asking me that

If not that would mean ducks and furries are blind

I used GPT-5.2 Pro on extended thinking btw

[valid charm]

Damn are you sick again???

[distant aurora]

At this point this feels like some sort of joke.

I know it sounds ridiculous if someone is literally never completely healed.

Maybe sick is a relative term idk. 😅

[valid charm]

Mann I’m sorry you haven’t recovered. I thought you where doing okay now

[distant aurora]

I'm better than before for sure.

And I appreciate the concern. (Really)

[gloomy cipher]

i never declined towards having issues but what would be the point of using a peasant like you if i read the whole convo

[distant aurora]

Hope you and your family are doing well.

[burnt forge]

i tricked perplixity to tell me to go on the black market!!!!!!!!!!!!!!!!!!

[valid charm]

Thanks man. Are you currently able to go see a doctor??

[gloomy cipher]

he refuse to go

[distant aurora]

Having sort of a rift with my parents and I don't want to spend money for now cuz I haven't been able to get a job.

[sage trout]

Hey all!
I’m a senior dev building application platforms that combine model-driven logic with distributed system patterns.
If anyone wants to talk architecture, debug edge cases, or exchange ideas, I’m glad to help.

[valid charm]

If you are sick you won’t be able to perform in a job. You should go

[twilit crown]

How do you deal with burn in on a keyboard screen?

[wise nova]

Which model can convert md to docx? How to properly prompt pplx output md answer to docx?

[distant aurora]

@green crystal Gemini down?

[distant aurora]

Doesn't perplexity have option to export response/chat as docx?

[wise nova]

Pplx in my case creates md snippet but not the generated file. Unfortunately didnt follow instructions to output in file

It's not converting of the existed files, it's a generating full new answer 🙁

[dry kestrel]

https://www.online-convert.com/

[wise nova]

My docs are privacy sensitive, so I can't upload them to third party services.
Why can't llm models convert docs..

[short narwhal]

https://pandoc.org/

[wise nova]

Yeah, gpt also told me to use pandoc. Currently I don't have a desktop but android tablet

[short narwhal]

Maybe https://github.com/gsantner/markor ?

[wise nova]

Markor is the same as I use Acode

Maybe I had to rewrite and add "output format: docx document".
How to force pplx output to document snippet?

[violet sable]

hi there

[dry kestrel]

why on earth are you using llms to convert documents? It's extremely inefficient and there isn't much point. There are plently of offline websites/apps that work as well.

You could even just ask Perplexity to make a script to convert your documents

[wise nova]

All in one tool

[dry kestrel]

inefficient and time consuming

[wise nova]

I asked pplx to generate a docs asset, but he creates only text md snippet 😭

[left stream]

let’s see how antigravity works on kali linux

[coarse dome]

just have it drespond normally you can download the conversation as DOCX after or PDF like a sane person

😛

[thorn wasp]

Hi, where i can open a ticket? i have a issue

[wise nova]

There are about 30 different answers in the thread, I will be tired of cleaning all the doc

[coarse dome]

Hmm I got nothing

[wise nova]

And pplx ignore the query not to include references like [file:398][file:399][file:317] [file:b5ab695-8152-44e6-9cc1-a8cf87b643]

[coarse dome]

https://x.com/OpenAI/status/2000669385317605759

OpenAI (@OpenAI)

Branched chats are now available on iOS and Android, too.

X · Dec 15, 2025, 8:48 PM

[dry kestrel]

[e-mail]

[coarse dome]

Well this is something I've heard people asking about across the different AI apps

[green crystal]

they can

with tools lol

on chatgpt there's basic file conversion

[coarse dome]

If you aren't using spaces to automate repeat tasks, why?

[fierce turret]

Do biddiy doddiddy do what in perplexity api shall.i do to.power my app eith 5 dollars only?

[coarse dome]

Perplexity API is $5 per 1000 search requests, if you're looking for Gen AI you should consider Google AI Studio unless you have something that needs search as part of it's output

[dry kestrel]

Perplexity gives you $5 worth of API credit every month for you to use their LLMs such as Sonar and Deep Research. The pricing you mentioned is only the search API.

You can learn more about pricing here https://docs.perplexity.ai/getting-started/pricing

[coarse dome]

Good to know but it sounded like they were asking what could they do with Perplexity's API , do you guys sell pass-through LLM access?

[dry kestrel]

Please refer to the pricing page for LLM pricing.

[coarse dome]

I can see that but is it still a search engine or just a LLM to use?

[solid star]

tools make ai powerful

[dry kestrel]

You can use the API however you want, either just as an LLM or LLM with research capabilities

[coarse dome]

I think that answers it so you're paying for a search LLM but how you use the info is up to you?

[dry kestrel]

Yes. There are 2 types, one is the search API where it will give you the raw web search results. The other is the Sonar models like Deep Research where it's able to call web search and output it in a format that is easier to deal with.

[coarse dome]

Okay, that's really helpful! Good to know there are alternative's to Web Search API if i need them. Based on this page can you share whether Deep Research and Labs use Sonar now or do they still use ChatGPT/Claude? I know there's not a way toggle which model is used nor see which was used.

[sinful merlin]

@dry kestrel

i feel like i'm going insane

it's been 2 or 3 weeks now

whoever got my ticket must have just taken a .22 to the dome and nobody checked to see if he was ok

[austere knot]

.

[fierce turret]

Not really search as part of its out out im just trying expemint with the resourcese avalible to.me that is include perplixity pro subscription which includes $5 free API credit im wondering how can i use it in real data or app or site expirence for fun nothing

[coarse dome]

Perplexity was ranked #3 on the list of top 10 most accessed AI services by Cloudflare https://blog.cloudflare.com/radar-2025-year-in-review/

Grok was #9 however a lot of traffic to Grok is through X (formerly Twitter) and Perplexity itself and DNS based data collection cannot see that.

Congrats to the Perplexity Team on being number three. Super happy to be a part of that.

[elfin grail]

Is everything alright you guys? My Perplexity can't make answer, only in working, and i don't know why

[tame current]

How long did you wait?

Is everything alright you guys?
Yes

[elfin grail]

Only few mins, and now I tried same question in new thread and it works, but the place i worked in still doesn't work 🤔

[tame current]

This is known to happen

[sweet sand]

hello I'm newbie

[rapid drift]

Cooll

I was thinking something, isn't perplexity better than AI Fiesta?

As perplexity is also offering a lot of models in its subscription

Just like ai fiesta

But perplexity offers more features

Or are there some differences i am not aware about

[modest gate]

🙂

[coarse dome]

this data is more valuable than benchmarks because it shows which apps have the users

[rapid drift]

AI Fiesta is new currently

[odd pelican]

Who's is ur top 5 rapper or no.1?

[young hollow]

As i had stated the gemini app is kina a shell it mainly uses Google

[sacred crescent]

[undone lance]

anyone elses recent threads get cleared?

[quaint ravine]

in Spaces?

[undone lance]

in library

they just popped back tho so

[icy parrot]

Google has services under cloudflare?

[coarse dome]

NO! The article observes DNS data from people who opt into using Cloudflare's public DNS resolver. So there is bias there. But of those people who use which AI services.

[icy parrot]

Oh

Iam dumb lol

[coarse dome]

However Google Gemini uses Google Cloud, everyone else uses Cloudflare. A massive single point of failure.

[icy parrot]

I knew that thats why i was confused at first

[molten rose]

https://x.com/growifyindia/status/2000807810888827382?s=20

Growify India (@growifyindia)

We are proud to announce that Growify India has officially incorporated as a Limited Liability Partnership under the Ministry of Corporate Affairs, effective December 15, 2025.
Introducing: GROWIFY INDIA LLP.
Learn more at https://t.co/jAfSrOGvam
#GrowifyIndiaLLP #CorporateIndia

X · Dec 16, 2025, 5:58 AM

[icy parrot]

Bew gpt image model is out

And iam really happy with it

[old sail]

I dont think its hard to build a linux desktop app ,since there are too many tools the get compilied to all platforms from a unified codebase !!

[hollow stratus]

Daily reminder that Perplexity doesn't care about its existing customers

[rapid drift]

thats wrong

only the traditional repetitive coding jobs are declining due to ai automation

the scope has shifted from writing code to designing systems that leverage ai

[hollow stratus]

Right, and you don't need the number of people you currently have for just that purpose.

[rapid drift]

hm

[hollow stratus]

I mean if you're top tier human capital and believe you can come in the 95th percentile of programmers, go ahead

[rapid drift]

you assume that if AI makes coding 50% faster, we will fire 50% of the engineers. in reality when a resource becomes cheaper, we consume more of it, not less.

because coding is cheaper, software is flooding into industries that couldn't afford it before, small businesses, local governments, and scientific researchers are now building custom software

[hollow stratus]

But the thing is that kids have started learning programming from age nine or something these days, you can't really compete with them if you're just a first year CSE student

[hollow stratus]

Sure, but what makes you believe that if we start consuming a ton of software, we'll need slower human developers to come in and take the reigns instead of scaling up our AI systems?

[rapid drift]

adults catch up fast. i can learn in one semester what takes a kid 3 years because I have the math and logic background they don't. a 9 year old can write a script, but they can't design a secure banking API or manage a cloud microservice architecture

[rapid drift]

we can't just 'scale up AI' because ai doesn't scale reliability, it only scales volume. software needs 99.9% reliability. AI is stuck at roughly 80% for complex tasks. That 20% gap requires humans.

the industry is shifting to human in the Loop, not human out of the loop

[hollow stratus]

The kids have been learning programming alongside math and logic. By the time you're both peers in first year doing CSE at a college, the kid would be much ahead of you.

[hollow stratus]

Yeah sure, I am not saying that humans would be obsolete any time soon. It's just that the numbers of them required would be much less, and those who are required would have to be very talented, unlike what we had a year or so ago.

[icy parrot]

I agree, but human can make same amount of mistakes, humans wont write 99.9% relaible software either just look at amount of memory issues and buffer overflows and underflows in current kernels which are made by top tier devs btw

[icy parrot]

The ones with better logic thinking, creativity will be relevsnt

[rapid drift]

hm

[rapid drift]

agreed humans aren’t perfect kernels have memory bugs b ut 99.9% reliability is not humans write perfect code it’s engineering controls like reviews, testing, security gates, monitoring, and safer languages

[icy parrot]

A good developer that can supervise waht the ai does will lead over developer without using ai jsut because of time and money saved

As long as you can debug and ubderstand software, networking hardware and whatever else ai is safer abd cheaper bet

[merry aurora]

Heya

[rapid drift]

Exactly that's my whole point. CSE isn't dyin, the job evolved

Developers who understand debugging, architecture, networking, and system design will use AI to work 2x faster. Those who only memorized syntax without understanding are in trouble

[icy parrot]

I think those were in danger even without ai lol

Developer without understading is not developer

[rapid drift]

yes!!

[rapid drift]

the data disagrees with 'much less numbers', BLS projects 17-26 percent growth (adding 300k-400k jobs in the US alone) and global developers will hit 45 million by 2030. almost double todays number.

you are right the bar changed, but it's not genius only, it's AI-native + system understanding which is exactly what a CSE degree teaches. that's why the scope is strong not weak.

[rapid drift]

hello

[urban moss]

I dont know about mac but windows app is just web view

[merry aurora]

Hii hruu

[old sail]

seems to be great

[hollow stratus]

Why do you switch from human to AI speak in the middle of a conversation

[old sail]

I think we can make a desktop version for linux

first I have to examine the windows version , so I decide the way to copy its functionalities

[urban moss]

you can make standole app from website in any browser right now

[rapid drift]

Why do you perceive my speaking style as resembling that of an AI?

[old sail]

we will see ,
also it will be amazing if I were able to build a plasma desktop widget that connects to perplexity

[urban moss]

as i know there is a desktop webview widget in plasma

[old sail]

a complete tray icon that pops up quickly when needed using a keyboard shortcut

I found this

[hollow stratus]

If you read a lot of LLM text, you eventually develop a sense of 'smell' for LLM text

And I have been smelling LLMism hard from those two messages you sent, even though the previous messages had no such thing

[rapid drift]

There's room for improvement in your smelling abilities.

[hollow stratus]

This is clearly AI

[rapid drift]

ya that one is

I did not mention I wrote it my self

[icy parrot]

Makes sense

[hardy nimbus]

Hi thanks for the pro subscription

[plain kernel]

Any have the latest updates on comet browser for iOS? Would love to test and provide feedback if that is option as well.

[balmy cove]

For writing essays, I noticed chat gpt is better for writing. But i like perplexity and how it provides lots of citations

Anyone know how to use perplexity for essays?

[brazen gate]

Ask ChatGPT

[wicked crane]

gpt mimmicks human speech the best

grok is second place but it's got only one personality

others are just corporate

[balmy cove]

I agree. Perplexity provides the best research tho

I only used gpt deep research mode once tho

[rapid drift]

for me its claude

[rapid drift]

Try claude once. For writing content its better than gpt.

[sullen crater]

What's the best way to use a large amount of sources in a Space? Should I just use my Drive as a connector where all the sources are or should I upload them in the Space from my computer?

[late lichen]

I've found Google Drive to be a good repository for information

[sullen crater]

I contacted support and they said Drive doesn’t work as a connector in Spaces for pro users and that I will have to upload manually but I tried to ask a query with the Drive as a connector and it used it in the citations

[late lichen]

Drive doesn't? That's strange. Did they remove that functionality recently?

[sullen crater]

Is this in spaces?

[sullen crater]

I showed him this ss too and he said it will just show Drive as a connector in a Space but won’t actually use it but it did use Drive when I asked it a question

[late lichen]

check spaces, it's available as an option in spaces as well

[woeful vapor]

hm

[empty ice]

Sources: "i made it up"

[empty ice]

Claude is good tho

Gpt thinking is the most reliable, claude is the most human-human

Gemini 3 is smart bur schizo

[rapid drift]

yes

Claude is better than GPT in terms of writing stories, essay, mails, messages

It feels more human

[empty ice]

Ye

[rapid drift]

Gemini for research than GPT

Grok for old history stuff I heard, is it true? Imma try

[empty ice]

Try gemini deep search too if you have gemini many students have it for fre

E

It became good

[rapid drift]

I made GPT and Claude write essays and stories

and made sonar compare them

at last acc to sonar claude's one are better

[empty ice]

I use grok for down to earth stuff

[rapid drift]

Fr

down to earth

[empty ice]

It really does make the best roasts too (people laughed at elon but we have tried it and it was actually funny for the group).

[rapid drift]

☠️

I asked chatgpt to roast my friend

it kept working for a long time T-T

ig it glitched

something

What's KIMI for?

[empty ice]

You are so weird even chatgpt crashes when you ask it to roast

[rapid drift]

☠️

nawh

that's what grok wrote or you?

[empty ice]

Kimi is great if you have a monster server/pc

[rapid drift]

oh

i see

[empty ice]

Nah that is me

[rapid drift]

oh

i see

[empty ice]

It is open source

[rapid drift]

oh

[empty ice]

And p good but very heavy

[rapid drift]

hm

it says hosted in US

[empty ice]

Yeah

Well it does use a lot of memory but

[wind path]

hello

[empty ice]

If you have memory it is not slow per se

Also they can distill

[rapid drift]

I have seen one thing

GPT has habit of throwing a lot of text on you

in both essays and stories

gpt wrote so much

[verbal notch]

do you have max?

[rapid drift]

I didn't wanna read

[empty ice]

I used to use an old 3090 for ai

[rapid drift]

oh i see

[empty ice]

For 24gb vram it was very cheap i bought a used one

[wind path]

how to set API for MCP Server, but..

[empty ice]

For like 450 usd

[wind path]

Is this different from the Perplexity Pro subscription with API? So I have to pay for the API?

[empty ice]

Well cheap is relative

[rapid drift]

hmm

[empty ice]

I live in hungary and it is still reasonably affordable here

[rapid drift]

Hm

[empty ice]

In the us people technically earn much more

[rapid drift]

perplexity made my life easier

[empty ice]

But they do have to spend more too

[rapid drift]

Let me tell one incident

[empty ice]

Same

[empty ice]

Not on tech

Except solar

[rapid drift]

So we immediately needed a bluetooth speaker.

At night

[empty ice]

Why arw solar panels so expensive in the usa

[wind path]

hmm

[rapid drift]

dad ordered one before me, I asked him to wait

[empty ice]

In hungary yoz can get 8kw for like 9k usd

[rapid drift]

perplexity found me the same model bluetooth speaker at less price being delivered under 20 min

I did ask my dad to wait but he ordered the same one for more price T-T

I had asked perplexity to find me a good one under budget

fr

[empty ice]

In the usa it is like 20k+

[rapid drift]

oh

hm

[empty ice]

We also have subsidies

We need data centers haha

Solar boom is crazy tbh

I love it

The grids do need upgrades or home batteries tho

[rapid drift]

hm

[empty ice]

I like nuclear energy too

Green+nuclear is needed for ai

Only france and slovakia has a higher share of nuclear iirc

[empty ice]

Eg my apartment complex was able to install solar at only 0.5 usd a watt

Crazy good deal

I pay around 0 a year for electricity

In the winter i do pay but i get it back next year

I did become a bit spoiled tho bc when i read power saving monitors etc i seem to just take it as a gimmick but yeah i forget that in some places electricity really is expensive

LPT if you need to save on energy use buy a gaming laptop instead a PC

Used laptops really are worth it if energy prices are high

Like in california

If you are a heavy gamer

A laptop can save you 200-500usd a year!!

But if you live in texas that number is far smaller

[urban moss]

it is just theme ok

[verbal notch]

ah ok 😅

[urban moss]

jk i have enterprise max

[verbal notch]

oh can i ask a few things. Although my main questions is the links/file limit for spaces and for enterprise max i know its much higher but im wondering if its higher for regular max as well

but is research/labs noticeably better an max than on pro?

[twin grail]

confirmed

Better models ( premium one )

when i checked it was claude opus 4.5

[verbal notch]

does it tell u that?

[twin grail]

i used complexity

to show that

[verbal notch]

hmm ok ty

[urban moss]

i dont think there is certain cap for spaces, i think it is limited by model context

[verbal notch]

well on pro i cant add more than 10 links for example

[verbal notch]

oh lol 💀

ok ty for checking

[opaque girder]

Hello together. What’s up?
Has anyone some experience with perplexity pro for research literature for a bachelor thesis or something else? 🙂

[ebon radish]

hi

[opaque girder]

Hellooo 🙂

[brazen gate]

Anyone non gamer?

[finite bolt]

hi

[golden rain]

wtf? gpt5.2 dont accept image input?

[urban moss]

same bug

[dusty kestrel]

Is perplexity a good all around ai or it is just a google powered by ai

[solid star]

@dry kestrel @distant aurora

[distant aurora]

Teach.

@green crystal 😭 Click the link to view your conversation with Kimi https://www.kimi.com/share/19b278ca-0262-8295-8000-0000eb934157

How this model climbed benchmarks is beyond me.

[sly comet]

hi

[gloomy cipher]

Wdym

I dont really like kimi k2 thinking :c

I prefer the old one

[distant aurora]

Bro never actually started the research. Just kept saying that it will start the research.

[gloomy cipher]

dont you need to select the research mode?

i dont remember but i think it had a research mode

[gloomy cipher]

also this is a tool calling issue

[distant aurora]

It's okayish. Rate limits are good, search accuracy is okay, there's rerouting issues though. You can dig about that on reddit.

[distant aurora]

I mean you can use the typical multi step web search as a shallow research too.

@solid star https://fxtwitter.com/OfficialLoganK/status/2000728193599226187

Logan Kilpatrick (@OfficialLoganK)

Gemini 3 Pro playing Pokémon vs 2.5 Pro (we used to all be impressed by 2.5 Pro)

**💬 91 🔁 126 ❤️ 2.2K 👁️ 193.9K **

FxTwitter · Dec 16, 2025, 12:42 AM

@gloomy cipher @green crystal @dry kestrel @cobalt vale https://fxtwitter.com/AISafetyMemes/status/2000620127054598508

AI Notkilleveryoneism Memes ⏸️ (@AISafetyMemes)

An engineer showed Gemini what another AI said about its code
︀︀
︀︀Gemini responded (in its "private" thoughts) with petty trash-talking, jealousy, and a full-on revenge plan
︀︀
︀︀🧵

**💬 377 🔁 667 ❤️ 8.9K 👁️ 680.8K **

FxTwitter · Dec 15, 2025, 5:33 PM

[gloomy cipher]

i usually dont believe these things especcially if its from gemini

because gemini can be overwritten with system instruction

remember the gold pricing?

[distant aurora]

Yeah.

[green crystal]

has to be edited lol

[distant aurora]

Idk

Still, the mentioning of models does feel fake.

[short narwhal]

The reasoning and answer style is influenced by the memory

[distant aurora]

Or maybe it was told in custom instructions to try to guess which model the suggestions are coming from.

[short narwhal]

Once again someone misinformation smh

[distant aurora]

https://fxtwitter.com/kimmonismus/status/2000838946159305155

Chubby♨️ (@kimmonismus)

Obviously I can’t confirm any of this, but would be huge

**💬 57 🔁 34 ❤️ 795 👁️ 51.0K **

FxTwitter · Dec 16, 2025, 8:02 AM

Crazy if true.

Arc agi 3 otw

https://fxtwitter.com/GregKamradt/status/2000636280401801251

Greg Kamradt (@GregKamradt)

ARC-AGI-3 will have >100 environments released
︀︀
︀︀Each one will be solvable by humans
︀︀
︀︀We are doing human testing on each one, and a game will *only* be included if it passed a minimum solvability threshold
︀︀
︀︀The currently public games weren't thoroughly tested (we got them out early, so just internal testing). The games that will be released will be, on average, easier.

Quoting George Raven (@georvn7)
︀
@GregKamradt Will you check ARC-AGI 3 against human testers? I've tried ls20-fa137e247ce6 at least three times, and it's hard to figure out what's going on. I never had this challenge with 1 and 2.

**💬 25 🔁 21 ❤️ 334 👁️ 37.9K **

FxTwitter · Dec 15, 2025, 6:37 PM

[torpid phoenix]

.

[dry kestrel]

what the

[gloomy cipher]

do you know how easy its to edit this btw

[short narwhal]

@dry kestrel call me chronically online the way I got discord as my n1 app on my Raycast Wrapped

[dry kestrel]

Saw this post, looks like misinformation

[dry kestrel]

HOW

WHY DO YOU NEED RAYCAST TO OPEN DISCORD

[short narwhal]

hotkey

💅

[dry kestrel]

nahhh

[short narwhal]

what's your wrap looking like...

[distant aurora]

https://fxtwitter.com/chetaslua/status/2000670516508545283

Chetaslua (@chetaslua)

🧠 IQ TEST OF SOTA LLM
︀︀
︀︀> GPT-5.2 thinking has highest iq even more than 5.2Pro
︀︀
︀︀> GPT-5.2 has more iq than 5.1 pro and Opus 4.5

**💬 14 🔁 27 ❤️ 212 👁️ 23.3K **

FxTwitter · Dec 15, 2025, 8:53 PM

[gloomy cipher]

@distant aurora its very easy to edit aanything for ss here is a non edited and edited version of a normal chat

[dry kestrel]

either im blind or

i dont have one

help

[coarse dome]

Want to make sure I understand this correctly. If you turn off all sources you're just using the LLM on Perplexity have one space where turning off everything including web is helpful.

[unborn monolith]

its very low quality (the perplexity in house model)

[gloomy cipher]

blind goose's ai be yapping all day to explain these images and links

[unborn monolith]

it only gives me very superficial replies

[devout spruce]

Does perplexity use nano banana pro? I only see nano banana 2.5 flash and seadrream 4.5

[short narwhal]

update

[unborn monolith]

and thats because its the cheapest model they have

[dry kestrel]

I did!

[short narwhal]

Update again

[short narwhal]

restart raycast

[gloomy cipher]

yea it is especially if you want to focus on your source/files only and make sure the ai does not pull information out of your study guide

[dry kestrel]

I did...

erm

[distant aurora]

Nano banana pro?

[gloomy cipher]

no its basic html bro

[distant aurora]

Oh. Inspect.

[gloomy cipher]

yea

[distant aurora]

Fair enough. My bad.

[short narwhal]

Disclaimer: Most of the data displayed is stored locally, and Raycast does not have access to it. If you moved between machines during the year, some of the data might have been lost. If you’ve used Raycast for less than 14 days, you will not have enough data to be displayed – but there’s always next year!

do you fit hose criterias

[distant aurora]

Is this fake too?

[coarse dome]

I wrote a tool that helps you fix spelling and grammar errors in your text, and also it optimizes it for oral presentation in a classroom setting, all while trying to maintain your voice and original word choices. And when web search is on, it gets new information and tries to add new ideas even though custom instruction says to not do it. So having the ability to turn off web searches actually really helpful for that one use case.

[short narwhal]

did you run this and then reopen it?? restart your mac or smth

[dry kestrel]

I've used RayCast for more than 14 days, haven't changed machine since I downloaded it

[gloomy cipher]

lol yea and websearch is gonna slow it down aswell

[dry kestrel]

Ok I think I messed something up

[short narwhal]

bro wtf is up with your mac and raycast

raycast from temu

[coarse dome]

I mean, that doesn't bother me as much. My citation tool for example uses the web search to get a better response. For if you are searching an ISBN number and you need to get back details about the book. So web search can often be super helpful just for like my editing space, it's not as helpful. One thing I would really like to see in perplexing in the future would be if they would like copy ChatGPT custom GPT so that we could share our custom instructions without sharing our chat history.

[dry kestrel]

:(

[gloomy cipher]

i did something similar by writing a python code that would use free oss 20B api from openrouter and takes in the text from my discord inbox textfield then apply the clearing and pasting it back all done with a keybind such as ctrl + shift + enter

[short narwhal]

Open up the settings of Raycast, right click the icon in dock, and Quit

[dry kestrel]

I had Raycast cmds disabled in the settings....

[short narwhal]

😭

[dry kestrel]

:(

[short narwhal]

nooo

[dry kestrel]

I'm hoping we get a Pplx/ChatGPT wrapped

[distant aurora]

Not sure how that would align with privacy.

[dry kestrel]

They have the data already

what are you on about

[distant aurora]

How would you exclude the private parts that you don't want to share from the wrap?

[rigid canopy]

What happened to Perplexity's Sonar models?

They halted development on those?

[green crystal]

then whatever comes from oai in jan 2026 is supposed to also counter 4.7? lmao

we'll have gpt-6 by Q2 if this keeps up

or gpt-5.5 max

with the legacy model menu being swapped for a search bar because of model saturation

[coarse dome]

Also, just a note I forgot to mention earlier, I'm relatively certain that Perplexity does not use search if you upload a PDF. I can't speak for all file types, but in my experience it only uses the PDF I upload as a source and does not use other web sources which. Quite helpful if I'm trying to get information from my textbooks quickly.

[steep ingot]

Top of the A y’all

[distant aurora]

Try this prompt on pro

White to move

1rbqk2r/4ppbp/3p1np1/8/Np1B4/6P1/PP2PPBP/R2Q1RK1 wk-01

[green crystal]

what lol

[distant aurora]

I want to see if it's real #💬│general message

[green crystal]

ahh

its chess?

damn

[distant aurora]

Bruh. Gemini didn't take long. I wonder if the answer is correct though.

https://g.co/gemini/share/4a63ba9c359e

[green crystal]

eh 18 minutes and counting

but it didnt give up

its working through

like for each move its reasoning through alternatives just to check if the move is good

its on extended thinking so think it'll be done after lunchbreak lmao

[rotund halo]

Hey all

[green crystal]

isnt the answer in the comments?

[regal dawn]

Hello guys
How do you find good ideas for making projects or problem statements for hackathons

[green crystal]

@distant aurora

[golden karma]

Hey all. I'm working on building a curriculum for a bunch of 4th-graders on a STEM project that includes several builds and learning different theories. I've built student textbooks, practice packets, readiness checklists, coach/parent resource lookups, and a couple other modules for the curriculum. I'd like to add some visual aids for some of these modules, but I'm kinda hitting a wall. Am I asking too much from Perplexity? Is that outside of its scope?

Edit: I've built all of this through Perplexity already. It's the image generation for visual aids that I'm having issues with.

[stark shuttle]

Perplexity can easily take on all these tasks

I have been using it for a while now and it has helped me build my portfolio and stuff

[golden karma]

I already completed all of that. It's just the visual aids that I'm having problems with.

I guess I'm just weak on image generation with Perplexity. It doesn't feel like it's there.

[stark shuttle]

If you need pictures or graphic stuff I would recommend using either gemini or chatgpt as they provide great image generation tools

Just give either of them context or the image on your mind will be way off with what the chatgpt or gemini have created

I hope it helped you.

[golden karma]

Yeah, that's helpful. I do know those do well with it. I'm asking more about Perplexity Pro's functionality with image generation, I guess.

[distant aurora]

Gemini did this in a minute

Gpt 5.2 high on lmarena gave up.

[stark shuttle]

Think of Perplexity Pro's image generation like having a dedicated, high-end creative partner built right into your research tool. Instead of settling for generic stock photos or running out of creative attempts halfway through a project (which happens with the free limits), you get to use the most advanced AI image models—like the ones that generate truly stunning, realistic, or stylized art. It means when your research asks for an image, the result is fast, contextual, and often visually richer, letting you build a report or presentation that feels more personal and engaging than what basic AI can offer. It’s about taking your specific, nuanced research questions and immediately giving them a high-quality, unlimited visual voice.

[distant aurora]

Babu Rao

[stark shuttle]

Yes?

[distant aurora]

Nice

[stark shuttle]

For what tho..

Btw how did you get the pro role

[distant aurora]

Profile choice

[stark shuttle]

I have pro but know how to get

[stark shuttle]

Ye

[distant aurora]

Use invite via browser

[golden karma]

That makes sense, but it also sounds like it wouldn't work for my puprose. Thoughts?

[stark shuttle]

So i just have to lok my account?

[distant aurora]

?

[stark shuttle]

Link*

[distant aurora]

Kinda

Idk how it works.

[stark shuttle]

Hm you're right i believe, as you're making visual aids for 4th graders it would not totally work for your purpose

Rather use dall-e or you can use stock images?

[golden karma]

I was thinking stock, but it'd be nice if I could have something more integrated and looking consistent with the rest of the text/visuals.

[distant aurora]

Try nano banana pro on lmarena if perplexity doesn't offer it already.

[stark shuttle]

Wouldnt it work if you use stock images that match a bit of your ideas and you edit them from ai?

[distant aurora]

Sounds like you are trying to make a page/presentation out of a chat.

[golden karma]

yes

[stark shuttle]

.

Heres what he asked for

[stark shuttle]

Then you have your problem solved!

Have a great day!

[normal tundra]

Babu rau 🤣🤣🤣🤣

[green crystal]

Hmm think it's because gemini is using more pre training?

Gpt seems to reason through problems allot more

Well apparently new models are going to the pre training direction again

So reasoning will probably be faster

[buoyant schooner]

what a shame

[green crystal]

@distant aurora @dry kestrel

[tulip crescent]

hahaaaha

[distant aurora]

Noice. It better beat nano banana pro.

[green crystal]

hmm

people say its already out in web

just not announced yet

[dry kestrel]

yes

[green crystal]

holy moly

its allot better

and its pretty fast

[queen locust]

Yolo

[distant aurora]

Very likely

[green crystal]

its trade off then

faster but as soon as something gets more novel it has a higher chance of hallucinating

also

image v2 is much better at editing fantasy stuff into irl images

but its still a bit off lol

much better than nano banana pro tho

[icy parrot]

Looks like it sicne generatio nis completly different today

It has better instruction following and flexiblity then nano banana pro

[green crystal]

very good consistency

[icy parrot]

It has worse udnerstanding of art style then old ones

[wind path]

i hope Perplexity makes an app like this that supports me in programming, hehe :)

[green crystal]

yeah seems like they couldnt really get both consistency and free form together

[icy parrot]

I mean i can still get my output closer then nano banana did but its kinda shame

I mean it undrestands spacing and ratios correctly now

[green crystal]

welp time to wait for image v3

[icy parrot]

BUut i kind hate missing the creative freedom

[icy parrot]

Yeah 😭

So in next life mby

[green crystal]

it has more creative ability than nano banana pro

[icy parrot]

True

[green crystal]

but the heightened consistency stiffles creativity

[icy parrot]

But less then older model

[wind path]

bro, GPT5.2 better than Opus 4.5?

[green crystal]

in some stuff

[icy parrot]

Its like comparing skoda favorit to ferrari

[wind path]

Isee

[icy parrot]

Opus 4.5 is better but then also its more expensive 💀

[wind path]

but i still use opus 4.5 for my project

[wind path]

u can try Antigravity, and get 100token

[icy parrot]

I use it but it has terrible tool calling replacing whole files instead of specific parts

[wind path]

You can prevent this by disabling "automatic replacement" so that every file that is changed or deleted will be confirmed first ☕️

@icy parrot

[icy parrot]

https://cdn.discordapp.com/attachments/1426646451743752292/1448779341910638746/image.gif

[frozen rose]

Me.

[icy parrot]

we got our answer

[quaint ravine]

https://openai.com/index/new-chatgpt-images-is-here/

[green crystal]

what

[quaint ravine]

GPT Image 1.5

[green crystal]

1.5???

breh

[quaint ravine]

for Max only here?

[green crystal]

it took centuries for a 0.5 update?

i mean something crazy is in the works if its 1.5 and not 2 but still

[left stream]

no clue, but gpt 5.2 pro is

[tawny coral]

Hi everyone 👋.

Is there a translation bot here ?

[wind path]

is.. ?

[left stream]

better than Claude 4.5 Opus

[green crystal]

4.5 opus is like in the middle between 5.2 and 5.2 pro

[wind path]

aha, but...
is it very good to be ordered when we use it in solving programming problems?

[tame current]

opus is better for codingn

[wind path]

okay

no debate. i'm still in love Opus :)

for my project

[tame current]

i used opus to create a 20 min speech for math club im in university and everyone told me i did great but opus did all the work

it created a beamer latex 'powerpoint' presentation

and did 20 min verbal speeech for me too

[wind path]

r u subs Claude ?

[tame current]

yeah im on max 200

[wind path]

good, im still trying out Opus on Antigravity app with my Google One subscription included..

[tame current]

i also have googleone pro

student account

[wind path]

same hahah

[tame current]

hhaha

[wind path]

soo "meh" Gemini 3 Pro

[tame current]

gemini aint so great tbh

[wind path]

but good for image/video generate

[tame current]

true

[smoky spade]

Why am I being told when trying to use gpt, sonnet, Grok, gemini that I have no more "advanced" uses left when I have a pro account and those were previously included? Only Opus was metered before.

[gloomy cipher]

man i remember when Dall E was a world changer and now i dont even get impressed even tho these models are topnotch

[tame current]

i think each ai has its strengtch weaknesses

you have to figure out what is 'mosts' importatnt t you

[golden rain]

Waiting for gpt image 1.5

[smoky spade]

What is going on with pro usage limits?

Did they change?

[tame current]

i have been noticing it seem like usage limits have gone down acrosss the board on all ai

[wind path]

"ai" is powerfull
because it is so "powerful" it increases the price of RAM 👊

maybe

[tame current]

forcing you to buy the higher packagegs

[smoky spade]

This is bullshit, until rn, I was able to use everything but Opus non-stop all day, everyday

[wind path]

gloomy future

[golden rain]

we’re not gonna buy max plan

[tame current]

this is why i pay 200 dolllar a month for max plan

[smoky spade]

They already gimped the context window. Gemini has a window of 2 million, Opus is high as fuck too. I know because I have open router, the trade off use to be context window vs unlimited use.

[tame current]

on 2 diff ai's

[smoky spade]

As in I can throw an entire book at Open Router versions of the same AI and keep using it and works, not so much on perplexity

[wind path]

but, u dont ask about Deepseek ? 👀

[smoky spade]

But at least I could use it as much as I liked, now every request is being pushed throughh the probot.

[smoky spade]

I have the 60USD Cursor.AI plan, a subscription to Chutes @20USD, Pro here and I also have anywhere to 5-20USD floating on open router when I need it for specific things.

No more money to spend, lol

[tame current]

right now i am learning how to create ai chat bots

[wind path]

n8n ?

[tame current]

n8nn is one i am learning

[smoky spade]

I use cursor AI for coding experiments, Mistral Nemo 12B is cheap on open-router and hooks in great to various different game loops. But I rely on Perplexity and Gemini for general stuff.

[tame current]

claude recomended n8n

i also have max plan on cursor ai

[wind path]

Last month I designed the UI/UX of the AI ​​Dashboard which is based on n8n ...

[smoky spade]

Claude is great but expensive, most of the good AI is expensive out the ass. I use sonnet and gpt codex most.

[left stream]

i’ll have to test them individually on that, I combine both most of the time

[smoky spade]

n8n?

[wind path]

and, GPT better?

[smoky spade]

LOL, I use Renpy as a front end.

[left stream]

tbh for research I use GPT and almost for everything else, creative writing and coding Claude 4.5 opus

i then combine them both since I won’t lie, I liked what I saw with both when they covered each other’s weaknesses

but tbh you can just do 4.5 sonnet and gpt 5.2 thinking at that point

[wind path]

how about Gemini 3 Pro .. hmm
i think better if u image/video generate

[wind path]

and I compared the latest Deepseek model with the Sonnet 4.5.

There's not much difference, but...

for developers/projects, it's a bit less good, because you can't display a preview directly.

[young hollow]

U have opus sub?

[left stream]

it’s free on antigravity and I have github student subscription so it gives me free 4.5 claude opus

[young hollow]

Ahhh i see

I don't have pc lol

[solid needle]

How do I get pro role, when I have perplexity pro?

[young hollow]

Bring Antigravity to Mobile

[left stream]

I won’t lie that would be dope

I really want that tbh

[smoky spade]

How does that work?

[young hollow]

Not yet

[smoky spade]

How does this work?

[young hollow]

The compute burned by mobile users will be insane

[smoky spade]

What's the cost?

[left stream]

free

[young hollow]

Google Ai Pro basically

[wind path]

free but limit

[left stream]

ys

yes

get student subscription for free for a year from google

then you have higher limits on antigravity

i won’t lie i haven’t hit a limit yet

[smoky spade]

How much? I have Google Pro but it's just Gemini Pro and Fast, Nano Banana, etc.

[left stream]

google pro is enough

[wind path]

i use it now for my project

[left stream]

that’ll make higher limits for you

[left stream]

it’s free, just download antigravity on a computer

[left stream]

it’s good right?

[wind path]

yep

AI ​​ on Antigravity app helped me build a discord bot

[distant aurora]

https://vxtwitter.com/flavioAd/status/2000995149808279930

Flavio Adamo (@flavioAd)

Nano Banana Pro vs GPT Image 1.5

A highly detailed aerial view photo of Times Square on New Year’s Eve, packed with crowds, glowing billboards, confetti in the air, and realistic reflections on wet asphalt at night

vxTwitter / fixvx · Dec 16, 2025, 6:23 PM

Is this the new one?

Google cooked if this is the new one.

[slow plover]

yup

[distant aurora]

@dry kestrel @gloomy cipher @left stream

[slow plover]

hey does someone have an idea what i can do if i generated 24 images but only 2 are visible doesnt matter which platform i use

[glass tendon]

Guys Perplexity seems kinda off for anything that isnt web style research and citations. Chat gpt pretty much does it all - or am i missing something ? I have perplexity pro. Is there an advanced guide or even a place for fresh ideas

[distant aurora]

https://tenor.com/view/psychic-third-eye-enlightenment-telepathy-gif-22477198

[glass tendon]

Well that is great Billy's Bones, is that the average perplexity user ?

[distant aurora]

You had an awakening, lol.

[glass tendon]

Lol haha fair enough !

[distant aurora]

There's high rate limits for sonnet and grok.

[glass tendon]

I like grok, but censored to hell.

[distant aurora]

And Gemini.

[glass tendon]

Sonnet i dont know, not really tried gemini

[gloomy cipher]

Left wins

[distant aurora]

Yeah. That's nano.

[gloomy cipher]

If you want a clear fake game style tho right wins

[distant aurora]

I think google is gonna give a hard time to open ai now onwards.

[gloomy cipher]

Yupp

On two ends

The Chinese and Google

[glass tendon]

Surely Grok will win this war ?

[gloomy cipher]

Who knows maybe even Microsoft

[distant aurora]

Microsoft ain't even trying.

[gloomy cipher]

Microsoft have been focusing a lot of efficient and cheap model that can be integrated into phones

[gloomy cipher]

They are actually

Most visual models were from Microsoft

[glass tendon]

Co-pilot is awful

[distant aurora]

They need a multimodal llm.

Grok image analysis is trash.

[gloomy cipher]

It's just that Microsoft is selling the models to other businesses their models are not for customers hence its not widely known

[distant aurora]

Qwen is probably the only balanced chinese ai service rn.

Good deep research, decent image analysis, good models and search.

[wind path]

how about deepseek ?

[glass tendon]

Was gonna ask same. Deepseek isnt bad

[distant aurora]

Deepseek search is heavily nerfed and they don't have their own deep research yet.

Chatgpt, grok, Claude have multi step search.

[gloomy cipher]

Deepseek lacks money

[distant aurora]

Gemini too probably. Gemini has it's own search that seems different from Google grounding.

[gloomy cipher]

Qwen is the Chinese equivalent of Google

[green crystal]

image v1.5 struggles with choosing between free form and consistency

[distant aurora]

At the moment, yes.

[green crystal]

when it chooses right though

[wind path]

however for comparison, Sonnet 4.5 performed well in working on and fixing bugs in my project.

Yes, it's similar, but its reasoning Deepseek is weaker than Sonnet

[green crystal]

holy moly its good

[distant aurora]

Claude breaks wallets.

Kimi should try being multimodal.

K2 thinking is still decent for free tier.

[distant aurora]

Deepseek's general intelligence is better than what it looks like in benchmarks.

Claude is focused on their niche.

[gloomy cipher]

@distant aurora you can search for
Phi 4
Flourance 2
And there is a very small 0.5 voice model

These are the good models of Microsoft. They really are focusing on small and efficient models iam assuming they are doing this to target laptops and phones in the future

[distant aurora]

Heard of phi 4, it's a good model for it's size.

Gemma type or better.

[gloomy cipher]

A while ago they made another small model that controls your browser

[gloomy cipher]

Yea thats their point

[distant aurora]

4b?

[gloomy cipher]

Iam not sure

I think It was 8B param

[distant aurora]

Running models locally gobbles power.

[gloomy cipher]

Thing is

[distant aurora]

I guess they can run it on those npu chips.

[gloomy cipher]

A model like 0.5 can be used on mobile phones so you could imagine that in the near future Microsoft would ne useful since these subscriptions are everywhere and costly

[gloomy cipher]

Its probably never gonna be more expensive than subscriptions tho

And it doesn't consume that much

[distant aurora]

I'll try it.

If I can on lm studio.

[gloomy cipher]

do you have a pc??

[distant aurora]

I really wish deepseek was better.

[sage trout]

Hi everyone
I’m Akio, just joined this server as part of the marketing team. I’ll be helping with GTM, acquisition, and broader marketing efforts and development.
I come from a SaaS background where I’ve worked on scaling products and driving revenue, and I’m super impressed by what we’re building here real solutions to real-life problems, making blockchain easier to use and helping drive real adoption.
Looking forward to interacting with the community and building together 🚀