💬│general

[obsidian umbra]

for less data, and only canada wide

they priced o3 super well imo

[indigo kiln]

That's because Google released Gemini 2.5 Pro back then :D

before, o3 was priced at 10$/40$

[obsidian umbra]

ik

also o3 price change was quite a bit after 2.5p no?

[young hollow]

@indigo kiln your Verdict?

[obsidian umbra]

you're prob close

[indigo kiln]

I can't tell for sure, but I think you're pretty close

[young hollow]

Gotcha

Tysm Again!!!

[vagrant merlin]

Is there any free platform that provide free api key that provide 200k token rather than openrouter

[indigo kiln]

Google literally mocked OpenAI back in June :D

[indigo kiln]

What model exactly are you looking for?

[vagrant merlin]

For vibe coding and thinking

[obsidian umbra]

google is still mocking them

😭

2.5p is still quite a bit cheaper

and its free limits are INSANE

[obsidian umbra]

https://ai.google.dev/gemini-api/docs/rate-limits#:~:text=250%2C000-,100,-Gemini 2.5 Flash

[vagrant merlin]

Yeh but it's not working properly sometimes

[obsidian umbra]

100 free uses of 2.5p

a day

bluethumbsup

[indigo kiln]

You can try using Google Gemini via their AI Studio at https://ai.dev

[indigo kiln]

Oops

[obsidian umbra]

this is the perfect mdoel for a discord bot

😩

also cloudflare workers "beta" llms that are completely free are very nice

[distant aurora]

Pirates know what this is about.

https://cunnyx.com/weirddalle/status/1957993748098359602

no context memes (@weirddalle)

💬 313 🔁 22.4K ❤️ 276.3K 👁️ 4.33M

CunnyX · Aug 20, 2025, 2:31 AM

[indigo kiln]

My friend successfully uses it for his unhinged Telegram chat bot :D

[vagrant merlin]

I have tried it but it creates problems and sometimes even tells me it is invalid

[obsidian umbra]

i too use it for my unhinged bot

gemini is so easy to bypass

[indigo kiln]

There's no enforced moderation at all to begin with

[obsidian umbra]

well, there is

it still has safety

its quite little, but you cant ask it about uh

smoke bombs for example

[young hollow]

I know very Well

[young hollow]

How about flash bangs

[indigo kiln]

It's got safety, not the oh no you said poop kind of vibe

[obsidian umbra]

it'll tell u how to make a pimpbomb if u ask

[young hollow]

What's a pimpbomb

[obsidian umbra]

uh

just leave it

it goes kaboom and does alot of damage

and no one should ever make one

[young hollow]

Daym

For sure

We don't need kaboom

We need technologia

TECHNOLOGIA

We need AGI/ASI

[obsidian umbra]

yea

tbf, no api does

atleast the ones ive tried

[young hollow]

What if i ask ai to make a very advanced laser

🤔

[obsidian umbra]

why in the fuck does restoring from icloud need to take an hour

give me my damn phone

[young hollow]

📱

[obsidian umbra]

by advanced, do u just mean

powerful?

[young hollow]

Just powerful laser

[obsidian umbra]

like a 25w laser?

[young hollow]

No damage

[obsidian umbra]

i mean

if u want one, u get the other

[young hollow]

I see

Idk lasers have always been cool and fascinating toys to me

[obsidian umbra]

if u just want that'll go one forever

just go on amz

they sell those

[indigo kiln]

Damn, that's disappointing

[obsidian umbra]

just dont be stupid w them

[obsidian umbra]

LGHTMOde

AHHHH

[young hollow]

It thought about harmful weapon lasers

And i meant just advanced lasers

[indigo kiln]

I accidentally did one of those above

[obsidian umbra]

its ok

gngy is on it

🙏

[young hollow]

Gotcha

[obsidian umbra]

i really need to put this back on my vps

[indigo kiln]

I just don't like using a dark theme in a bright environment

[obsidian umbra]

oh god

[indigo kiln]

Thanks for the reminder

[young hollow]

I guess if we give it proper instructions for making a advanced unharmful laser it would do it

[indigo kiln]

I need to get VPS in the first place

[karmic isle]

lol

[obsidian umbra]

aws ec2

for bots

bluethumbsup

[indigo kiln]

Yeah, sure, it would comply if you force it to view your context as "learning about completely safe and highly advanced lasers that will never hurt anyone"

[young hollow]

Jailbreaking is also there

[indigo kiln]

What about using Termux on a $5 phone instead

[young hollow]

Or uncensored models

[obsidian umbra]

ec2 is free

[young hollow]

Sounds like me

[obsidian umbra]

gngy yapped

[young hollow]

Daym

[obsidian umbra]

[obsidian umbra]

if u really wanna pay

i use racknerd

11$ a year

u get what u pay for, but they are pretty good

[indigo kiln]

I wonder what's under the hood for such a low price

[obsidian umbra]

1x vCPU Core
10GB SSD Storage
768MB RAM
1TB Monthly Bandwidth
1Gbps Public Network Port
Full Root Admin Access
1 Dedicated IPv4 Address
KVM/SolusVM
Linux O/S
Available in Multiple Locations

good enough for a dc bot

and maybe som other useless things

they have tiers on their yearlys tho

[young hollow]

Welp gn guys is 2 am

And try to sleep early

[obsidian umbra]

"early"

[fierce turret]

good night guys we will talk to more prep perp!!

[indigo kiln]

We run private proxies on such low-tier nodes, and they work just fine :D

[obsidian umbra]

that too works

🙏

[serene sand]

Automatic comes from Greek ‘automatos’ meaning ‘self-moving’ and basically anything that runs, reacts, or clicks on its own without you even thinking about it

🌀

Aravind Srinivas (@AravSrinivas)

Comet can autonomously set up a @Shopify Store for you. How is this @tobi?

X · Aug 21, 2025, 8:18 PM

[dry kestrel]

What happened to the thumbnail 😭

[serene sand]

https://youtu.be/3KZ6mIsbDc4?si=3vUxgTVlfNcb7PUf

[dry kestrel]

One of my favourite actors

[gloomy cipher]

Arvin was moving the monitor

[cunning aspen]

I am using Perplexity Pro and can't access the Tasks menu. i searched the entire interface and can't locate it. Anyone else have this experience?

[indigo kiln]

Try using a direct link: https://www.perplexity.ai/account/tasks

[dry kestrel]

Click on your pfp

[timber niche]

No hat 🙁

[dry kestrel]

I'm not changing my PPLX avatar

[timber niche]

😢

[indigo kiln]

DeepSeek V3.1 is a bit too verbose by default... (1.25 tokens per word)

[dry kestrel]

@fast meteor Came across this today - 26M parameter model doesn't seem too bad, and finetuning this should be much easier and faster. https://github.com/jingyaogong/minimind/blob/master/README_en.md

[distant aurora]

God watching his children be like https://www.vxinstagram.com/reel/DNjVeXSM8um

[ebon slate]

Guys. PPLX pro doesn't even come with Claude opus 4?

Is it just sonnet 4.0?

[distant aurora]

Nope. That's stuck behind 200 dollar sub.

[vapid plover]

It's just Sonnnet 4

[ebon slate]

Damn. Sonnet 4 is available for free in Claude ai itself

What's the use then

[vapid plover]

The thing I don't like the most is gpt 5 thinking is for max only but it's and price in apis as gpt-5

It's like same price as normal but costs 200$ perplexity for some reason

[serene sand]

more logic @ pplx

[ebon slate]

Like? Does the explanation differs? It's more clear and stuff?

[distant aurora]

That's the non thinking variant.

[dry kestrel]

Opus 4 is incredibly expensive.

[ebon slate]

PPLX has thinking varient?

[dry kestrel]

Yes

You get more sonnet 4 usage on PPLX Pro then a Claude Pro subscription

[distant aurora]

Guys please check api pricing before demanding expensive models like opus and O3 pro under 20 dollars.

[ebon slate]

Hmm I mean I'm only intrested in Claude ai as I'm a dev. But currently can't afford Claude pro. But was surprised when I've seen there is no opus and 4.1 sonnet so

[dry kestrel]

You get 300+ uses of sonnet 4 every day on Perplexity, the same can't be said for Claude

[ebon slate]

Yes but somewhere I've thought PPLX is just offering as much as what free tiers are doing in the available ai's

[distant aurora]

Where are you getting sonnet thinking and O3 for free?

[wise nova]

Do you seen GPT5 Thinking in Max?

[ebon slate]

Oh so Claude kinda restricts user when it comes to messages too

[distant aurora]

Yeah just today. Unacceptable.

[dry kestrel]

Yeah, a lot

[woven brook]

How long does the comet trial last. I asked perplexity it didn't know 😉 7-days?

[cunning aspen]

Ok. found it. I was in the app, not accessing through a web browser. Why is this feature not accessible via the app. I am using macos - latest version.

[wise nova]

Betrayal

[ebon slate]

So sonnet 4.0 thinking in PPLX is better than the Claude sonnet 4.0 free varient in the Claude ai website? Over-all

[serene sand]

PPLX - logos

Claude/chatGPT - mythos

@ebon slate

[distant aurora]

Yes

[dry kestrel]

I'm not too sure, you can make a feature request here #💻︱feedback-desktop

[timber niche]

https://www.facebook.com/share/p/1AnDZPKjmh/

Has this been addressed

[distant aurora]

This feels out of context.

[ebon slate]

Okay

[ebon slate]

So Claude in PPLX is kinda better with thinking compared to the actual free varient right?

If that's the case

I'm happy

[dry kestrel]

Yes

[vapid plover]

Perplexity enterprise gives unlimited Claude 4 sonnet thinking and if you think no opus is insane the Claude.ai 20$/month package doesn't get you much prompts I've seen it myself the 100$/month and 200$/month do give enough opus prompts for actually using

[timber niche]

Great

[distant aurora]

Unlimited?

[timber niche]

Huh

[dry kestrel]

It's partially patched from what I understand

[timber niche]

Hn

[indigo kiln]

Some people feel a bit frustrated with Perplexity, because they expect overkill models in a highly optimized RAG + web search environment with focus on factual accuracy and low latency

[ebon slate]

Okay. So the value for money we get here for PPLX pro is actually worth it. And more text and tokens compared to the Claude free tier. Then it's good thing

One bad thing about PPLX is it's android app. Idk why but it sucks

Website is good

[timber niche]

Hopefully they fix it completely

Because it is very concerning

[distant aurora]

Overkill?

[serene sand]

#👑│pro-lounge message

[ebon slate]

Why am I not given the option to select the model in android application? Like in windows or website

This is a bit frustrating because i only wanna deal with Claude

[dry kestrel]

https://cdn.discordapp.com/attachments/1398497101125783572/1407486387317510245/i9LIt2K.jpeg?ex=68a84198&is=68a6f018&hm=13d0e1d4792547fef442ba7292b5cb3b89f581c70e9375430c07f24adf3e8b3b&

https://cdn.discordapp.com/attachments/1398497101125783572/1407486387560775760/2kpFxTj.jpeg?ex=68a84198&is=68a6f018&hm=ccb2179ea3a767d305f4508473641ff673cc29d2c2c2c6f9890e823ae7f7ff22&

https://cdn.discordapp.com/attachments/1398497101125783572/1407486387833536553/BljsOQd.jpeg?ex=68a84198&is=68a6f018&hm=1686c73136602b825ad5c3573960172dc17aa2db825349ac62e19cdc6ded16fb&

[wise nova]

Consider perplexity system prompts which may distort your queries

[indigo kiln]

Haven't noticed any search limits honestly

[distant aurora]

300 per day

[dry kestrel]

The system prompt on PPLX is pretty standard, just tells the model about tool calls and such

[distant aurora]

Or 400ish

[feral marlin]

On my computer I have GPT-5, on my phone I only have GPT-4.1

[dry kestrel]

#💬│general message

[dry kestrel]

Please update the app

[ebon slate]

Thank you bro. Should have been straight option like in web to select a.model

[distant aurora]

You aren't countering anything with that reply.

[feral marlin]

I don't see an update, let me try reinstalling

[dry kestrel]

?

[serene sand]

just read exact link

[wise nova]

How do you know it?

[dry kestrel]

I'm not trying to counter anything? Just offering a more detailed explanation

[ebon slate]

So any model we use we get 300 ish prompts with PPLX?

[distant aurora]

To me?

[distant aurora]

Yes.

[dry kestrel]

300+ queries a day across all models

[distant aurora]

300+

So I would say 400ish

[dry kestrel]

Just try extracting it

[distant aurora]

There's no official explanation about it.

And last time ed tested, it stopped working after around 400.

[dry kestrel]

Plan:
• 200 Pro queries/week
• 20 Deep Research queries/month
• Unlimited Best queries
• 50 Uploads/week
• 3 videos/month (8s, Veo 3.1 Fast)
• 40 Comet Agent queries/month
• 0 Computer Credits/month
• Monthly limits apply to: GPT-Image 1, Seedream 4.5, Gemini 3.1 Flash Image (Nano Banana 2)
Note: Computer credits has to be bought separately

Enterprise Plan:
• 400 Pro queries/week
• 50 Deep Research queries/month
• Unlimited Best queries
• 100 Uploads/week
• 5 videos/month (8s, Veo 3.1 Fast)
• 80 Comet Agent queries/month
• 500 Computer Credits/month
• Monthly limits apply to: GPT-Image 1, Seedream 4.5, Gemini 3.1 Flash Image (Nano Banana 2)
Note: Comet Agent limits ≠ Comet Assistant limits

Plan:
• Highest level access to Pro Search, Deep Research, Create files/apps (Labs), and Comet Assistant
• 15 videos/month (Veo 3.1 + option to use Seedance)
• Extended image generation (includes Nano Banana Pro)
• Access to Model council
• Access to Computer (10,000 Credits + 20,000 Limited Time Bonus)
• Upgraded Comet Assistant (Opus 4.6 Thinking)
• Advanced Deep Research powered by Opus 4.6 Thinking

All query limits operate on a rolling window basis and regenerate gradually over time. Limits do not refresh on a set date.

** Comet Agent limits ≠ Comet Assistant limits ** Click here to learn more about the difference

Support Response (Not AI): Using the default model ("Best") doesn't count toward your limit.

You can visit https://www.perplexity.ai/rest/rate-limit/all and https://www.perplexity.ai/rest/user/settings to check your usage.

Limits subject to change per public documentation.

Regarding Perplexity Labs/Create Files and Apps #💬│general message

[ebon slate]

Id say that's good enough. More than enough for a basic user or programmer.

[distant aurora]

Correct me if I'm wrong.

[distant aurora]

If you can work with the 32k context, it's amazing.

[dry kestrel]

your quota slowly regenerates. If I consumed my queries in one burst in the morning. I'll have it back in the afternoon.

[feral marlin]

I cant belive it was this easy

[ebon slate]

The tokens is really bad tho. 32k is way too less for 2025. Like gpt 5 comes with 400k ish tokens. Should have atleast been 100k

[distant aurora]

Gpt 5 thinking is 196k on chatgpt.

[dry kestrel]

Yeah - for context all the models on the ChatGPT Plus plan used to be 32k. Now GPT-5 Thinking got bumped to 196k

[distant aurora]

Yeah

[indigo kiln]

within a rolling window
Thanks, it explained everything instantly :D

[wise nova]

So, it's better to pay $20 for ChatGPT Plus for Thinking instead $200 pplx max

[dry kestrel]

Disagree - you get o3 pro and opus 4.1 on PPLX Max

[distant aurora]

If you only want gpt 5 thinking, yes.

Perplexity max is mainly meant for opus and O3 pro.
@green crystal do you see any glazing?

[dry kestrel]

And labs + better comet assistant

[ebon slate]

How exactly does PPLX work? I've heard from itself that it doesn't exactly respond like gpt itself. But it uses gpt integrated with PPLX search engine to generate responses. So the responses we generate using PPLX gpt 5 would be actually very different from the gpt 5 from open Ai itself right?

[wise nova]

Why I need Opis 4.1 if already have GPT5 Thinking

[dry kestrel]

Opus 4.1 is better at coding

[ebon slate]

Yeah it is. It just got much better. Even 4.0 is good

I like it over gpt

[frank spoke]

Has anyone tried coding along with perplexity, if so then what the best model and how you guys manage your projects with it? Thanks!!!

[dry kestrel]

I mean, it uses the same underlying model (GPT-5), they could tweak things like temperature, reasoning effort. PPLX uses their own search API and custom system prompt.

The model capabilities should be roughly the same.

[wise nova]

Pplx can be suitable for coding? Or just for searching?

[dry kestrel]

Try using sonnet or labs for coding

[dry kestrel]

I prefer Opus 4.1 when searching - you can code small simple things with PPLX

[distant aurora]

20 dollar sub on perplexity uses the dumb non thinking variant of gpt 5.

[indigo kiln]

I tried creating a space with a system prompt identical to my ChatGPT, and with web searches disabled. Guess what happened? It answered almost exactly like GPT-5 in ChatGPT

[distant aurora]

You got perplexity max?

[dry kestrel]

I have a claude subscription.

[left stream]

Yes, or jailbreak agent and have it do it for you

[distant aurora]

Oh.

[left stream]

Or jailbroken comet x jailbroken agent

[distant aurora]

No way you waste your opus on searching.

[dry kestrel]

I bought the $100 Max just to use search with Opus

[distant aurora]

😭

[ebon slate]

How did you disable web searches in PPLX?

[distant aurora]

Hard to believe.

[dry kestrel]

Press the globe icon.

[indigo kiln]

Uh...

[wise nova]

$200?

[dry kestrel]

$100

[wise nova]

How??

[dry kestrel]

#💬│general message

[ebon slate]

So it's better we use Claude sonnet 4.0 also by disabling web search I guess. Because i like sonnet 4.0. it's better I don't get any modifications from PPLX. I'd prefer it raw

[dry kestrel]

You can make a space with web disabled by default to streamline the process.

[indigo kiln]

That defeats the purpose of Perplexity...

[wise nova]

So I can buy $100 Claude sub, $20 ChatGPT, $20 Pplx, it is $140, not $200

[ebon slate]

But I'm still getting more prompts and responses.

With pro

[ebon slate]

That's nice

[dry kestrel]

I get around 80 uses of Opus 4.1 every 5 hours. You get much higher rate limits with a PPLX Max plan.

[indigo kiln]

Well, that goes without saying :D

[distant aurora]

Unlimited according to perplexity page.

[wise nova]

I use pplx rarely than 80/per 5h

[dry kestrel]

Then buy a claude sub?

[distant aurora]

#💬│general message

[ebon slate]

I didn't exactly compare and try how is web search helping while being interested with sonnet 4.0 to RAW sonnet 4.0.
But not sure if it can actually make responses better than the raw Claude ai

[dry kestrel]

Just get the subscription that fits you the most.

[frank spoke]

Thats' interesting.. I never thought a so called web search replacemnt can be used as NOT a web search tool 🫡

[wise nova]

Not yet, still using pplx and google AI

[indigo kiln]

This can be used to rely on your own sources in Space instead of searching the web, for example

[dry kestrel]

Then why are you so insistent on getting Opus 4.1 onto the Pro plan? Just get a Claude Max subscription if thats what you want to use.

[ebon slate]

Ai tools just made learning and coding much easier to grasp. There were times we had to deal with textbooks and carry the doubts we get during studies for the next day to a lecturer.

I bought 1 year subscription as I'm weak in coding. I hope I'll make full use of it and get better at coding 😂

[indigo kiln]

AI has made coding more difficult for me personally. Instead of doing the classic Ctrl+C -> ChatGPT assist me with refactoring please -> Ctrl+C -> Ctrl+V chain, the world is now forcing me to use integrated copilot agents :D

[wise nova]

I think it is possible to budjet opus 4.1 in pro costs

[dry kestrel]

It's an extremely expensive model, at the end of the day PPLX has to make money and introducing Opus 4.1 would make them go bankrupt.

[distant aurora]

Nah

Nope

[indigo kiln]

Unsustainable

[distant aurora]

15 dollar input 75 dollar output per million tokens.

[ebon slate]

Even VSC comes with copilot agents. While I write my code it spoils the entire experience by giving all the code I'm just about to write. I feel like my ideas are being overridden by Ai. Not just the text.

So I kinda disabled it

[distant aurora]

Compared to like 1.5 and 10 of gpt 5 thinking.

[wise nova]

Give us 100k tokens, not 1M

[distant aurora]

That's not how it works.

What about the other customers and other model options?

[indigo kiln]

For me, the only thing worth leaving for was the inline autocompletion feature in PyCharm :D

[distant aurora]

The ability of web search costs extra.

[dry kestrel]

Just look at the cost man

[frank spoke]

I tried using labs but I wish if I could to that in spaces, set a fixed model per space and then do my coding stuff there. But the model selection is not persistent.

[dry kestrel]

Yeah it's unfortunate that you can't change models for labs :/

[ebon slate]

Labs are just 50 a month. That's really sad. 50 a month? We can exhaust all in a single day 😂

[dry kestrel]

Better than ChatGPT's DR at 25/month

[distant aurora]

https://tenor.com/view/donut-glaze-satisfying-gif-13180530

[dry kestrel]

(also they refill slowly rather than at the start of each month)

[wise nova]

You understand that by collecting $20 from each person on average, Pplx pays the total amount, but someone uses 1%, someone uses 90%, someone uses 200%. Due to the average amount, the economy is going

[ebon slate]

I think it's just better to use search rather than labs. Once labs are exhausted we get demotivated getting used to them

[dry kestrel]

I'm pretty certain that they are consuming VC money right now

[distant aurora]

Yeah but opus is too much of a risk under 20 dollars.

[dry kestrel]

50 labs is more than you think (they also refill slowly so in theory you could get more than 50 uses a month)

[distant aurora]

If they did offer it for that cheap then everyone would dig in and perplexity would go bankrupt.

[serene sand]

deep dive ➤ #💬│general message

🤿

special attention to Tutorial II

[ebon slate]

One question is like one lab right? 1 lab = 1 prompt + 1 response by pplx
If I'm not wrong

[dry kestrel]

yes

[indigo kiln]

With great power comes great responsibility. If everyone knew that Opus 4.1 was the best AND available for $20/month, everyone would use it for all their tasks, even if it's redundant

[distant aurora]

https://rebrand.ly/pplxt

[wise nova]

You will collect from everyone, for example, $2000/100 subscribers, but they spend only $500 for opus

[dry kestrel]

With great power comes great electricity

[ebon slate]

The 20$ subscription that comes with Claude is pretty useless as per what I've been told. The responses are too low for a paid subscription

Not sure how many though

One of my friends tried

[dry kestrel]

Less than 10 uses of Opus 4.1 every 5 hours

[wise nova]

Ok, forget opus. What about GPT5 Thinking? Why it not incl in Pro if cheaper than opus

[dry kestrel]

and thats if your prompts are like 1k tokens

[ebon slate]

Damn that hits hard

[indigo kiln]

It's still better than what Mistral offers, isn't it?

[dry kestrel]

You could send like 50 "hi"s though

[distant aurora]

When you use a model through api, there's parameters to control. Perplexity alters them to balance cost and quality while trying to maintain accuracy in responses.

[dry kestrel]

mistral models aren't too great in performance :/

[distant aurora]

Every third party provider does this to some extent.

And even chatgpt natively has it's own set of parameters.

It's not as unhinged as raw api.

Raw api without tuning would have unpredictable costs.

[ebon slate]

Yeah it sure is. So it's 200 responses per day. But like berd said it's just 10 4.1 for 5 hours. That's the only thing that's bad i guess

[ebon slate]

Yeah that's for sure.

[wise nova]

I will be enough with 10 queries opus 4.1 in Pro

[indigo kiln]

Custom agents are saving the day for me :)

[ebon slate]

That's great then.

For me it'd be less

As I'm a beginner to intermediate level guy and I do like 2-3 questions every 1 hour

[dry kestrel]

Assuming each output is 40k tokens, your 10 prompts would cost them $30 USD. And that's not even counting input tokens

[ebon slate]

If I'm from intermediate to advanced. Id do 1 program for every 2-3 hours as it gets tough

[dry kestrel]

I mean, it's great for just having conversations not so much for STEM related tasks.

The speed on the thinking models is amazing though

[indigo kiln]

Before buying Pro, I was also afraid to use Perplexity :D

[wise nova]

Pplx plans not about unit-economics. It's about shared total

[indigo kiln]

Magistral models tend to spend 5-6x more thinking tokens than popular models though

[ebon slate]

Are you feeling comfortable with PPLX now? How's your experience been

[dry kestrel]

It's simply not sustainable for PPLX to add Opus at the current pricing.

[dry kestrel]

They think a ton and then the output is really mid

[wise nova]

One used just $1, other maybe $10, and third $30, so total $41, but collected $60

[dry kestrel]

130M reasoning tokens is just insane

[indigo kiln]

I was enjoying the free tier already, but something clicked inside me and I decided to upgrade to GPT-5 answers :D

[dry kestrel]

If you require Opus 4.1, just get a Claude subscription instead.

[distant aurora]

Is it worth it though? If the model is dumb it would lead to just unpredictable overthinking.

[indigo kiln]

That's if the model doesn't get stuck in the middle of the reasoning process

[distant aurora]

Open router best.

[ebon slate]

Okay so you upgraded to PPLX pro just for gpt 5 ?

[dry kestrel]

100% not the case for using Opus

[distant aurora]

That gpt 5 is non reasoning variant 😭

[dry kestrel]

Yeah lol

[distant aurora]

I'm trolling.

[indigo kiln]

Not only that, but there's also o3 and FINALLY more than three researches a day

[ebon slate]

Yes

[dry kestrel]

It's not - the model is low in performance

[indigo kiln]

I'm okay with non-reasoning models. I'm already enjoying using GPT-5 (and GPT-5 mini, yep) in ChatGPT on a daily basis

[ebon slate]

I think gpt 5 better than o3

[distant aurora]

You can just add "think hard" to your query to trigger gpt 5 auto thinking in chatgpt.

Lol.

For now atleast.

Hope they don't patch it.

[wise nova]

Does the pplx max have this code capabilities?

[wise nova]

Even in pplx?

[distant aurora]

Nope

Only in chatgpt.

[indigo kiln]

o3 has much better intermediate planning when you really need it, but mostly I don't. And GPT-5 is basically GPT-4.1, but without the silly emojis and sycophancy. It's just what I always wanted :D

[ebon slate]

I really don't know much about these researches and labs thing. What are researches exactly? All i use is search that's all

[distant aurora]

Thorough multi step strategic searching.

[indigo kiln]

It uses GPT-5 Thinking with low effort, which is mostly enough. I wish it also worked for web searches. That's exactly why I bought Perplexity Pro

[wise nova]

So why to paid $200 for max

[indigo kiln]

ChatGPT still uses some sort of GPT-4o mini for web searches...

[dry kestrel]

It gets you GPT-5 Thinking on the $200 subscription on PPLX

[distant aurora]

For opus and O3 pro. And buffed comet assistant and labs.

[wise nova]

I get GPT5 Thinking for $20 from Plus sub

[distant aurora]

Gpt 5 thinking is not the strong selling point of perplexity max 😭🙏

[dry kestrel]

No, it gets you the equivalent of 64 juice. In the API if you select low its 16 juice

[indigo kiln]

Research == model plans everything and triggers web search much more often

[dry kestrel]

Then get the $20 ChatGPT sub

[indigo kiln]

It doesn't feel like Medium honestly :D

[dry kestrel]

If you just want to use GPT-5 Thinking, get a ChatGPT subscription

[wise nova]

As we found out Pplx haven't code capabilities of Claude Max

[distant aurora]

Can I get the source for this?

[dry kestrel]

Prompt yourself

[distant aurora]

I saw this on your Twitter.

[dry kestrel]

Yeah - I tried to consolidate all the possible combinations

[distant aurora]

Sweet.

Eddi be testing everything.

[dry kestrel]

Hehe

[indigo kiln]

So it really is GPT-5

[dry kestrel]

Testing Catalog has more info than me though

[wise nova]

How to activate thinking in Pplx Gemini 2.5 Pro?

[dry kestrel]

Prompting think harder in PPLX does not result in GPT-5 Thinking

[dry kestrel]

It's already activated.

[ebon slate]

The one option I'm missing in PPLX pro is to create a project and then create several new chats in a single project.
I had the feature in gpt plus while I had the sub i really enjoyed it

[dry kestrel]

try out the spaces feature

[indigo kiln]

Yeah, it doesn't. It's probably just a remnant of GPT-5's internal router

[dry kestrel]

It's close to the projects thing on ChatGPt

[wise nova]

Then why Gemini in non-reasoning section?

[dry kestrel]

Because you can't see the reasoning summaries

[ebon slate]

How is that similar to it? Can we create multiple chats in one space by naming it python or something

[dry kestrel]

and to avoid users from asking "Why is gemini put in reasoning when I can't see the reasoning"

[dry kestrel]

All the chats started from a space is organized into that space.

[indigo kiln]

I'm still curious why every AI company makes projects and folders so exclusive that they [people] need a subscription to access them

[wise nova]

Original Gemini still havent project feature

[dry kestrel]

Gives people a reason to buy the subscription

[indigo kiln]

Perplexity didn't pressure me into buying a subscription, though

[ebon slate]

So I can just prompt like add all the chats with start with the word DAY. and create new chats with the names DAY 1, DAY 2. and then all are organised?

[wise nova]

I see "researching.." when ask Gemini in pplx.
It's very good if Pplx Gemini uses thinking/reasoning model

[dry kestrel]

All the chats that you start from here gets consolidated into the space underneath.

[wise nova]

What magic word will force Gemini to choose thinking mode? "Think hard and deeply"?

[dry kestrel]

There is no non reasoning variant for 2.5 pro

[distant aurora]

@dry kestrel perplexity should consider a credit points based system in the future for people who demand expensive models under 20 dollars imo. More flexibility for individual user needs. What do you think?

[indigo kiln]

There is no such magic word. 2.5 Pro is already forced into reasoning by Google, but it's reasoning process is cloaked

[dry kestrel]

Depends on how their credits point system works

[distant aurora]

Gemini 2.5 pro always thinks.

[ebon slate]

Thank you. I'll try it

[wise nova]

How thinking differs from reasoning?

[distant aurora]

Kinda like Merlin or sider but more transparent about limits.

[distant aurora]

Same thing.

[dry kestrel]

It's the same thing

[distant aurora]

If your model isn't thinking on Gemini, you are probably using 2.5 flash.

[indigo kiln]

Oh

[distant aurora]

To make 2.5 flash think, you'll probably need to make a gem.

[dry kestrel]

You need to be in the server for a while to get embed perms

[wise nova]

I mean Gemini in pplx

[distant aurora]

It thinks. It just doesn't show it. We have tested it with tough math problems and stuff.

[indigo kiln]

I guess I just overlooked that part, thanks

[dry kestrel]

If I was on the PPLX team, I would never add anything like this.

Makes it confusing for new users

[wise nova]

One cool feature in pplx which I like is Spaces, where I can choose different models in on thread or project. No one gives this possibility

[ebon slate]

@tranquil orchid yo you can respond here?

[distant aurora]

How? It's basically like in game money.

[dry kestrel]

Nope

[distant aurora]

You want more expensive stuff, you pay more.

[ebon slate]

Just guessed as it's PPLX discord

[distant aurora]

There was an ask perplexity bot but it's deprecated now.

[dry kestrel]

This is extremely confusing and hard to remember.

[indigo kiln]

It's basically Windsurf

[distant aurora]

No one's gonna remember the whole price sheet. Different niches of users will remember costs of their preferred models.

[dry kestrel]

There's a reason why PPLX has a usage based system rather than a credit one.

[distant aurora]

What you are saying is like keep very limited items in a mall or else people will get lost.

And a mall clearly has way more items than this pricelist.

[ebon slate]

I thought it's PPLX model compatibility list for a sec. Then seen 4.1 opus ticked for pro. 😂

[distant aurora]

A restaurant menu is another example.

[dry kestrel]

You asked for my opinion, I'm not here to debate this

[distant aurora]

Alright.

@green crystal what are your thoughts on this?

[green crystal]

Dafook is goin on

[distant aurora]

Asking your opinion on some stuff. Check mentions/pings.

Sorry if it got spammy.

[indigo kiln]

It could be a small tooltip, like what Complexity uses

[green crystal]

The credits system?

[distant aurora]

That. And another ping.

[distant aurora]

Yep.

[wise nova]

Wow, what the service?

[distant aurora]

Wait.

[indigo kiln]

Our beloved Perplexity plus cplx.app

[dry kestrel]

Could induce FOMO with something like this - It would be interesting to see for sure if they do decide to implement something like this

[distant aurora]

Gemini 3.0 pro?

Wtf?

[wise nova]

Why we stiil on 2.5 Pro

[distant aurora]

Wait.

[indigo kiln]

||F12 / Inspect element trolling||

[indigo kiln]

We're so AGI now

[dry kestrel]

Fr

I got that special access from Sam Altman

[gloomy cipher]

Does anyone know how to open the same chat on open router mobile and pc? Ima using the same account but the chats aren't in sync

[distant aurora]

Gpt 6 became autonomous and learned time travel.

[dry kestrel]

Fr

[dry kestrel]

Afaik not possible

[indigo kiln]

It would be great if all services supported Mermaid charts

[gloomy cipher]

Such a weird con of open router

[wise nova]

It is windsurf?

[dry kestrel]

Hard to implement, I know that the devs at T3 Chat had a hellish time doing this

[rigid ether]

I used sonar models through API, even sonar deep research through API. It's expensive for long usages but one thing I understood that perplexity is very low* quality compared to direct API. That's why they are giving nearly unlimited msgs In pro plan. Reason is quality is 70% reduced.

[indigo kiln]

No, I just gave an example of how it could look like if it was done right :)

[distant aurora]

https://tenor.com/view/zhongli-hmm-zhongli-geo-geo-daddy-daddy-gif-21570059

[wise nova]

But Gemini 3.0 is already exist

[indigo kiln]

I'm sorry, but I think you're imagining things

[indigo kiln]

on an isolated Google TPU for internal use only :)

[distant aurora]

So is quality 30/100 or 70/100 ?

[rigid ether]

Your wish, i already used today 5usd by directly using sonar deep researcher, sonar reasoning APIs

[gloomy cipher]

Every llm has it tho :/

[indigo kiln]

I'm not a fan of how things are set by default...

[rigid ether]

Sonar deep research and sonar reasoning are legendary models when I used on API, but in perplexity they are behaving like poor reasoning

[gloomy cipher]

Just create a shareable link and let my info go viral 🧘‍♂️

[indigo kiln]

Way better now :)

[dry kestrel]

Are you talking about real time streaming of the message across multiple devices?

[distant aurora]

Is it 30 or 70 on perplexity?

[dry kestrel]

I know for sure that ChatGPT and claude does not support this

[indigo kiln]

If anyone can pull this off, they should be brought on board at FAANG right away

[rigid ether]

Original API direct use is 100% and unbelievable quality

Same models on perplexity are showing up only 30% of direct API usage quality

[green crystal]

Where is o5 pro

[dry kestrel]

Hire Theo

[distant aurora]

Bold claims.

[wise nova]

YES

[distant aurora]

Learning how to travel between dimensions.

Istg

[gloomy cipher]

its gonna be for the new pplx ultra max subscriptionist

[distant aurora]

These are all flash memes.

[distant aurora]

500 dollar sub.

[indigo kiln]

Pro Max*

[distant aurora]

I hope this perplexity from another dimension has bigger context. 😭

[wise nova]

Max Pro

[indigo kiln]

1k more, 33k total

[rigid ether]

It started with one prompt, direct API query, then I shocked if it is really perplexity product or not. Direct sonar deep research through API gave literally equivalent to gemini deep research and chatgpt deep research quality. But in perplexity it is only 30% of original API quality. I already spent just now 5usd in this testing and realised direct API is expensive even though quality is awesome

[warm scroll]

why is there even a downgrade button the hell 😭

[wise nova]

And Opus 4.1 in pro

[distant aurora]

Their opus will be as smart as haiku then.

[short narwhal]

Haiku 4 wen

[gloomy cipher]

what is more funny is when you run out of limits, it suggest that 😂

[distant aurora]

Cancel subscription option basically.

[indigo kiln]

So you can purchase another annual Pro subscription, of course

[dry kestrel]

Never

[distant aurora]

Fr.

[warm scroll]

okkk.... and why would I want to cancel my free 12-month subscription?..

[dry kestrel]

To stop it renewing on your next billing period?

[indigo kiln]

Llama 3.3 70B we... oh wait

[warm scroll]

how would it renew if my card isnt even on there tho?

[dry kestrel]

It won't

[distant aurora]

Because not everyone gets it for free like you.

[warm scroll]

that's a good point

[dry kestrel]

It's for the people who actually paid an annual sub

[distant aurora]

How did you manage to get pro without a card?

I had to put my card there just to claim gift card.

[gloomy cipher]

how did the 11/10 enhancement go ? 🧘‍♂️

[dry kestrel]

you could remove your billing information (I think)

[warm scroll]

some guy on instagram said that there is a promotion between perplexity x t-mobile

so i said fck it and gone done it

[distant aurora]

Teach me this. I need to.

Okay. Mares confirmed it.

Cannot be removed.

[dry kestrel]

Emoji man back at it

[distant aurora]

Soundwave vibes.

LMFAO

[dry kestrel]

Seems like you could just press the "x"?

[distant aurora]

Wild.

[short narwhal]

Haven’t used it too much yet, but when chatting it seems to be more hyped. Other queries seem the same besides maybe a few more emojis? The same basically though

[dry kestrel]

Oh lmao

[distant aurora]

I'll have to wait for it to expire before I remove it I guess.

[serene sand]

try if you can, it’s How Stripe operates…

[distant aurora]

Damn.

Gemini level.

How much did 1 research cost?

[green crystal]

Really?

[indigo kiln]

Is it still as bad as before?

[rigid ether]

Yes, it also generating pages of quality research, but more better structured and more meaningful info than gemini DR.

[distant aurora]

Gemini Deep research is best imo.

[worthy urchin]

Sleep

😭

[distant aurora]

Perplexity's research?

[dry kestrel]

This happens https://cdn.discordapp.com/attachments/1047649527299055688/1408220627449675908/Screenshot_2025-08-22-04-15-45-15_21da60175e70af211acc4f26191b7a77.jpg?ex=68a8f329&is=68a7a1a9&hm=14ceb555ffa79de005f9281a0f911fff1d0141683496ecd7ae51ede6ff5dd644&

Smh

[indigo kiln]

Gemini app experience compared to that of AI Studio or API

[rigid ether]

Not sure. It is same price as O3 model, input is 2usd per million, output is 8usd per million, i used back to back and completed all my 5usd credit.

[distant aurora]

App is shit. Use website.

[indigo kiln]

Yep, that's exactly what I was wondering about

[serene sand]

i know… emojied you, but you keep exploring/wasting time

[distant aurora]

How many were you able to for 5 dollars? And how much time did it take?

[spare plank]

Hey folks! 👋 Does anyone here happen to have a spare invite to Comet? Would really appreciate it 🙏

[dry kestrel]

you never know unless you try

and I am suspicous of the emoji man

[distant aurora]

https://tenor.com/view/tfp-soundwave-gif-27115566

[green crystal]

Lmao

[dry kestrel]

He stopped giving me millet and I'm getting hungry

[indigo kiln]

I'm actually a fan of Perplexity's Research :D
It's fast enough to keep me interested, but long enough to collect data from a significant number of sources

[distant aurora]

It's not perfect but there were significant improvements after the last rant.

[rigid ether]

I don't remember count bro, i even used sonar reasoning pro and sonar reasoning also, I kept on testing quality but never counted how many msgs. May be i made a bill of 5usd in 2 hours approx

[distant aurora]

Particularly reduction (not elimination) of hallucination.

Response length was buffed too.

It would say it's good if you are not dealing with too many variables.

[indigo kiln]

As for Gemini's Deep Research, it works like a charm even on 2.5 Flash
If the whole Gemini app experience wasn't such a bummer, I'd even pay for Google One with 2.5 Pro :(

[distant aurora]

Lmao perplexity logic.

[green crystal]

Because you don't need to pay $200 a month to use gpt-5 thinking on chatgpt

You get gpt-5 pro and much more for $200 per month

[distant aurora]

20 dollars, 3000 messages per week. 196k context.

[indigo kiln]

I know what offer you mean, but it doesn't apply to me :)

[green crystal]

Hope you aren't joking thinking being max only

That would be a scam

[distant aurora]

It is.

Lol.

[green crystal]

Oh

[rigid ether]

I think* they don't put in max also

[green crystal]

Bruh

Disgusting

[green crystal]

It's basically the cheapest reasoning model there is for it's class

Wth

[distant aurora]

Opus is too pricey but demanding gpt 5 thinking for 20 dollar sub is valid.

[green crystal]

Anyway time to go make some food

[indigo kiln]

If I'm remembering right, Perplexity uses their own DeepSeek R1-1776 model for researches

[distant aurora]

Yeah.

[gloomy cipher]

Honestly This is also a bad upgrade for max users because they should have gotten gpt 5 pro

[indigo kiln]

Exactly

[short narwhal]

Not in the API yet (unless it is nowadays)

[distant aurora]

https://tenor.com/view/donut-glaze-satisfying-gif-13180530

[short narwhal]

Not sure how that would help in perplexity in any way

[gloomy cipher]

It's not but my statement still stands

[dry kestrel]

It's not :(

[indigo kiln]

Still not in the API

[gloomy cipher]

Or gpt 5 high for max,

[short narwhal]

Perplexity has its own pipeline system for search, even if they added it to the model picker it wouldn’t actually be the one doing the tool calling (unless they added it to comet as the assistant model, but that would likely require post training)

[distant aurora]

https://tenor.com/view/vignesh123-vigneshfisrtgif-yoda123-yodamars-gif-13055175463724769097

[dry kestrel]

The models do the tool calling, they just use the PPLX API for searching though

[short narwhal]

🤨 proof?

[dry kestrel]

wait

I have the full prompt somewhere

[short narwhal]

Is this the agent search or just any model with basic search?

[dry kestrel]

and I can't find it

[dry kestrel]

Just normal search outside of Comet

[short narwhal]

Ehh, never seen any evidence before that it could actually be capable of that. Would be a nice suprise though

[indigo kiln]

Looks reliable to me

[dry kestrel]

Aligns with what I found previously, as well as when the model broke formatting or something and the tool calls were exposed

[eager socket]

Hi, I've got a problem between connection with Perplexity app for MacOS Destop and my browser using the same Pro credentials. When I am looging in with my credentials in my browser, I do not have my Pro subscription. What did I do wrong?

[warm scroll]

What is the best model on perplexity for coding?

[distant aurora]

Did you try refreshing? Might be that cloudflare verification issue.

[tulip nebula]

What is the size of context window of gemini in PPLX?

[distant aurora]

Gemini and Claude.

[distant aurora]

32k

[tulip nebula]

Why do they limit it? Isn't it originally 1M?

[dry kestrel]

Due to cost

[dry kestrel]

Try reinstalling the browser and clearing browser cache

[indigo kiln]

Because less is more.

Many large language models show a “lost in the middle” effect—performance is highest when relevant information is at the beginning or end of the context and degrades when it’s in the middle, especially as the context grows longer.
(c) Perplexity

[distant aurora]

@languid star @young hollow I love twitter.

For people who don't know context, use perplexity research to find out about e20 petrol situation in India.

[indigo kiln]

🥺👉👈

[distant aurora]

That's intentional sabotage by government to force customers to buy ev.

Potentially from some kind of pressure from automobile industry.

[indigo kiln]

Could have been handled better probably

[eager socket]

Thx solved. My mistake. Sorry 😅

[languid star]

pressure from automobile to govt, thats insane 😭

https://www.perplexity.ai/search/what-could-be-the-real-reason-8WKxIqyUQOCKpJcpfZjJCw#0

tbh ngl these reasons have some weight

[amber badge]

All my API keys and credits are gone?

[languid star]

https://www.perplexity.ai/search/so-is-this-right-way-to-shift-I6Gk5FTeRHyRjvh6N1A94A

instead of gradual govt is doing at fast rate, currently oil war is going on b/w countries who knows what could be the reason

[tranquil orchid]

@amber badge

FairLunch said:

@vague pagoda -- my API keys are gone and credits too. I'm using the same account. I had $20 in the API acc, then everything vanished, and I reloaded $5 to test it. My project that uses my API key is working/running

Please avoid @mentioning members of the Perplexity Team

[old leaf]

Ios 15 web while screen how to fix this?

[amber badge]

It's an emergency ;/ stupid bot.

[gloomy cipher]

.

@distant aurora

[dry kestrel]

Please email [e-mail]

[dry kestrel]

It seems that they updated how imagen works, I remember before it would automatically route you to another model. Now it straight up tells you "Image generation rate limit exceeded. Do not try again. Please let the user know they have exceeded their limits for the month and that they can try again in a few days or upgrade to Max for more credits."

[naive sage]

"for the month"? Oh wow, I thought imagen had just a daily quota

[dry kestrel]

GPT-Image 1 has a monthly limit of 150 images

Assuming they are using the medium quality and you are creating landscape'/portrait images, it would cost them $9.45

[naive sage]

I remember Sam Altman saying that the costs for generative AI generally goes down about 5-10x year over year, so as things get more efficient, etc, that $9.45 could be only $0.94

The problem is they'd still have to recoup a lot of their initial investment to build out all that expensive infrastructure... so I doubt prices will fall. But this is one reason that I think smaller models that individuals can run will be good enough for most people at some point. Niche, specialized applications (especially around academia and learning in general) will be more long-lasting premium offerings with a strong subscriber base, though. And maybe the really high-end models will still have some researchers and design studios or something.

[formal geyser]

So I hear there is Image generation on Perplexity now

[naive sage]

Yes, they've had it for awhile, just have to say generate an image from the prompt

[dry kestrel]

I think he was referring to LLM's rather than image generation. And it would make sense for OAI to decrease prices if they did manage to find optimizations in this area to undercut the competition.

[formal geyser]

Also video generation as well?

[naive sage]

I thought no, but supposedly that is a very new feature that they support (but very tight limits too, only what - like 5 a month or something?)

[dry kestrel]

Yep, 5 videos a month

[formal geyser]

Only 5? Even with Subscriptions?

[dry kestrel]

Would cost them 7+ USD if you used up your quota

[dry kestrel]

Video gen is incredibly expensive :/

[naive sage]

Yeah, video generation credits go brrrrr

Video generation is also possible locally, but you need like 16GB VRAM+

[formal geyser]

Damn that actually sucks but makes sense money don’t grow on trees

[dry kestrel]

and it's going to be worse than veo3

[naive sage]

Yeah, veo3 is very good

[formal geyser]

However I can say the quality is very good

Surprisingly

[serene sand]

#📢│announcements message

[minor shard]

Im curious why o3 is not Max but Gpt 5 thinking is

[cloud tartan]

Older model

pretty much

[minor shard]

Yeah but I would assume costs are cheaper for GPT 5 thinking

Im just suprised unless this is their GPT 5-Pro model. Gpt 5 Thinking isn't even locked behind OpenAI's Pro sub

[naive sage]

I think actually o3 is slightly cheaper [now, at least, not when it launched]

[minor shard]

yeah but not 10x cheaper

oohhh welll

Im still waiting for Deepseek V3.1 to show up in LM Studio/Ollama

[wise nova]

Why not to replace o3 by GPT5 Thinking?

[dry kestrel]

I guess Max subscribers get new models first?

[minor shard]

yeah idk

[dry kestrel]

Also GPT 5 costs more

[clever flicker]

Meow

[gloomy cipher]

HUH

[distant aurora]

Pov:
You buy perplexity max for gpt 5 thinking.
Gpt 5 thinking says you were stupid to buy it.

[minor shard]

lmao

[minor shard]

Understandably considering they dropped the price on o3. But still, its just going to push people to buy OpenAI sub instead of pplx. The non-reasoning model you can just get for free with limitations of course. I really want people to sub to Perplexity cause having multiple models is better than just one.

[distant aurora]

https://tenor.com/view/lilia-vanrouge-silver-vanrouge-silver-twst-twisted-wonderland-gif-11576704522248992570

[distant aurora]

And perplexity search is good.

[green crystal]

is it because of the $10 vs $8 token difference?

[dry kestrel]

also because GPT-5 thinks more

[green crystal]

[distant aurora]

Have fun reading it. https://www.perplexity.ai/search/analyse-the-angle-of-indian-go-o70Z0CLIQcWpmnW_uI6gWQ#0

[distant aurora]

Use medium. Not high.

🧠

[dry kestrel]

Thats not fair considering o3 was on high

[green crystal]

chatgpt uses medium right?

[dry kestrel]

Depending on your sub

[minor shard]

Isn't the GPT5 Thinking model Medium and not high?

GPT 5-Pro is high

[distant aurora]

On chadgpt or pplx max?

[minor shard]

In general

[dry kestrel]

This was from my testing https://x.com/EdDiberd/status/1955707761943093686

[distant aurora]

Bro leave the embed.

This was from my testing https://cunnyx.com/EdDiberd/status/1955707761943093686

EdDiboi (@EdDiberd)

Updated @btibor91's Juice summary as of today.
︀︀
︀︀API, ChatGPT Plus, and ChatGPT Pro Juice scores

💬 1 ❤️ 3 👁️ 282

CunnyX · Aug 13, 2025, 7:07 PM

[green crystal]

so plus gets medium then

[dry kestrel]

Takes up space

[minor shard]

I would assume PPLX would have labeled it GPT5-Pro if it was the high model

idk

[dry kestrel]

Ehhhh nah

I doubt they would do that since it's kinda misinformation

[minor shard]

I guess

[distant aurora]

It's still decent imo.

O3 ish level.

[minor shard]

So If GPT5-Pro isn't "high" then how do you get Gpt5 (high) lol

[distant aurora]

Valid question.

Api I guess?

Since O3 high wasn't an option on chatgpt either.

[green crystal]

in general above o3 imo

base model itself just seems to recognize things more easily during reasoning

[minor shard]

Well in API you only have low, medium or High. I would love to know if they are using High or Medium.

[distant aurora]

Here's a better option.

Give the pro users gpt 5 mini thinking.

[minor shard]

lol

[distant aurora]

Nah. nvm.

O3 already does it.

[green crystal]

medium is pretty close

and with the nice weekly limits i wont complain lol

[languid star]

horizontal tabs are such a mess, daym i miss vertical tabs

[minor shard]

medium is pretty good. I don't think high is needed majority of the time.

[distant aurora]

What we need is chatgpt like multi step search in pro.

The thinking mode drags it long.

[minor shard]

I can't live without verticle tabs anymore. Its just so much easier to navigate

[gloomy cipher]

I wana marry gpt 5 high

[distant aurora]

She can fix us.

🍻

[green crystal]

a mini deep research basically

[dry kestrel]

What do you use it for?

[gloomy cipher]

This part is really really weird

[dry kestrel]

I find that my screen isn't large enough for split windows

[dry kestrel]

The benchmark is already saturated

just like AIME 2024

[minor shard]

Its nice that I can make the verticle bar as slim as possible to get maximum real estate and then it extends. Its just a smooth experience

[dry kestrel]

Interesting, I've actually never heard of vertical tabs until seeing the posts in #1393681130842226698

[distant aurora]

Grok 4 has disappointed me.

[minor shard]

It took me a sec to get used to at first but I wont ever go back

[dry kestrel]

It's not available on chrome :(

[distant aurora]

They might have made it kinda complacent.

[minor shard]

oh yeah I stopped using Chrome a while ago. I liked Brave's privacy and ad blocking etc.

[dry kestrel]

Interesting, I couldn't get used to Brave's UI

esp the bookmark button

[minor shard]

I always thought they were pretty close although I personally don't use bookmarks too much. I have like my top 10 and they are in my bookmark toolbar haha

and then I unfortuantely use Edge for work 😢

[dry kestrel]

I keep on pressing the brave button when I try to bookmark something

[minor shard]

I thought you could move those icons around

I can't wait till I don't need to make bookmarks anymore and AI just does it for me lol

Comet plz bookmark

[dry kestrel]

Lmao

[dry kestrel]

I couldn't find a way :/

[minor shard]

If you right click the home button you can customize the toolbar but it won't let you move it to the right where it belongs 🙁

oh wait it does

[dry kestrel]

Oh wai what

[minor shard]

https://tenor.com/view/book-monster-logo-tale-story-gif-16469550

[dry kestrel]

Where is the home button 😭

[minor shard]

right click the bookmark button haha

oh wow

brave://settings/appearance\

[dry kestrel]

Found it

I can't move this button though

[minor shard]

No I would just turn it off and then right click the home button and customize the toolbar to put the one on the right

silly adjustments

[dry kestrel]

AHH

I can't move it to the right where Chrome usually has it

[minor shard]

Yeah its silly. You gotta turn that one off and then customize the toolbar to enable bookmarks on the right

[dry kestrel]

Ah

My brain is overheating

[minor shard]

yeah Im slightly annoyed I had to do that lol

Its great at blocking ads and stuff though.

I definitely love the direction Comet is going in though.

[green crystal]

@distant aurora

pplx needs this

[minor shard]

It used to show its thinking. Been gone a while now

[minor shard]

What model did you use 😮

[dry kestrel]

Sonnet

[minor shard]

I did too but it didn't show anything lol

Oh maybe I did the wrong question

Thank you ❤️

[dry kestrel]

Nice

o3 works as well

[languid star]

https://x.com/SebastienBubeck/status/1958198661139009862

Sebastien Bubeck (@SebastienBubeck)

Claim: gpt-5-pro can prove new interesting mathematics.

Proof: I took a convex optimization paper with a clean open problem in it and asked gpt-5-pro to work on it. It proved a better bound than what is in the paper, and I checked the proof it's correct.

Details below.

X · Aug 20, 2025, 4:05 PM

[minor shard]

Been using Copilot more with work. much more usable now lol

[languid star]

emoji_10~3

[minor shard]

thats a great step up for an opensource model

token cost is like 1/4 less lol which is great considering most models are using more tokens lol

[dry kestrel]

It's still kinda mid https://artificialanalysis.ai/models

And the HLE they showed was with tools

[minor shard]

is this with or without tools

oh ok

I mean.. for a model that I can run on my PC, I don't mind taking a hit in performance

Looks like qwen3 is better tho but thats 235b so I would just stick with deepseek 3.1, although it sounds like its not much different than R1

[young hollow]

Is About The Costing Bro

Being World's 7th Greatest Model

Which is Being Handled with Low Tier Chips

Cheaper then Closed Source Models

Free for General Users

If You Ask Me That's Really Something Big and Admirable China has Achieved

https://www.perplexity.ai/page/india-bans-online-money-games-G7hIFXf.Qkyg7rWLsoQauQ

[left stream]

I have finally been blessed

I’m gonna get trained by a boxing coach who trained world champions and golden gloves winners holy shi

[languid star]

https://tenor.com/view/peaky-blinders-no-fighting-no-fucking-fighting-fight-fuck-gif-16259749058826278291

[dry kestrel]

Are you sure you can run V3.1 on your PC

[young hollow]

Btw guys have any of you Tried and Tested Z GLM?

[minor shard]

Prolly a quant model haha. Idk about the full thing

[twin grail]

cholacdit

[pine abyss]

what's new in the new comet update

[formal rover]

💀

[tranquil orchid]

@sweet zealot

MrADHD said:

@vague pagoda Someone gave you the PPLX mode that has a higher rank here. Why can't you just connect to them? What's the benefit of being in this mode? Just to flex?

Please avoid @mentioning members of the Perplexity Team

[twin grail]

first of all, mod not mode

second thing

Everyone have a different role for each duty

Pplx staffs in here have different roles in the companies which u might know or not

So their duty wasn't being adding what request but further foward it to the team

Just buy chatgpt at this point

why wasting time complaining ?

20$ isn't the end of the world if you are a heavy worker

[languid star]

90% daym, idk where u got this percentage lmao, pplx gave free via other telecom too in other countries and samsung too

[twin grail]

Even a student can afford it, u use a research based AI to code

then complain about it ?

[twin grail]

That number is fake

[languid star]

obv

[twin grail]

Not when i understand why they don't add it and the actual perfomance of it

[obsidian umbra]

they never added o3 until the price was cut down to half, current gpt5 costs more than o3 iirc

esp with reasoning

im sure feedback does matter to them, but at the end of the day it is a cost thing

[languid star]

then they should add vertical tabs 😭

[obsidian umbra]

for comet?

make an ext and do it ur self

bluethumbsup

[languid star]

https://tenor.com/view/magnus-carlsen-magnus-carlsen-rage-slam-table-gif-3278955469026070231

[wheat kindle]

hey , how can i get an introduction to the ai ? i dont know a single thing about it

[languid star]

search it on youtube

okchuphoja

[wheat kindle]

😭

[languid star]

u watched terminator ? thats AI

[wheat kindle]

bruh

[quaint ravine]

how's your experience with it?

I thought it was really good until I modified a prompt I had (analysis of a PDF attached) - GPT-5 Thinking only gets it right like 20-30% of the time

while PPLX + Gemini 2.5 Pro gets it right most of the time and takes less than half the time too lol

[twin grail]

For math

i'd say it fits my need

I mostly use it for math and other coding stuffs

[twin grail]

I'm not really know about the medical part

[quaint ravine]

it's generally solid, but not clearly better (and often worse) than PPLX + Gemini 2.5 Pro or Sonnet 4 Thinking

[twin grail]

tbf, just buy chatgpt and refund like u said

Then why are u still complaining about it ?

About a product u don't use anymore

right...

[vague pagoda]

• I work at perplexity, my role isn't that of a moderator
• I've dmmed you

[karmic isle]

lol

[haughty birch]

then why you want AI bro, chill know

[silver drum]

Does research pro has limit like how many queries per day?

[dry kestrel]

#💬│general message

[dry kestrel]

Learn more about rate limits here #💬│general message

[dry kestrel]

https://www.perplexity.ai/search/give-me-an-introduction-to-llm-m8zV85klTSC.YHKNIBF3Mw

[silver drum]

Why it was showing me only 10

[languid star]

its app or website

[silver drum]

Website

In app nothing like this showing

[dry kestrel]

Looks like it's unintended

Maybe an issue with India based users?

[languid star]

mine show like this

[gentle whale]

Brother is a little behind

[languid star]

its a bug ig, happened before many times

[ebon slate]

guys where do i toggle web off in here?

I mean I'm not getting direct option like this

[silver drum]

Pro

[languid star]

this is a thread ig

within thread u cant disable

[ebon slate]

oh, so its not possible in a thread

[languid star]

no, start a new thread with web off

[languid star]

which device btw

[dry kestrel]

Please contact [e-mail]

[silver drum]

Tablet

[languid star]

mine

@silver drum

lol

[twin grail]

u did a lot of research lmao

[languid star]

https://tenor.com/view/shocked-dog-gif-14542099238045824336

[silver drum]

How to check?

[languid star]

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

[languid star]

show urs

[twin grail]

slayyy

[ebon slate]

How are you guys knowing how many searches left per day?

[languid star]

https://tenor.com/view/cat-meow-silly-cat-cat-explode-yellow-cat-gif-12654556298327310612

[languid star]

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

[twin grail]

Enterprise peak fr fr

[magic silo]

In the pro version I have limits for creating images. It is assumed that in pro everything is unlimited. What can I do to solve this problem?

[languid star]

ah, i forgot u r in another league

[twin grail]

150 for gpt1 image gen

the rest are unlimited

[silver drum]

This reset every day or every month

[twin grail]

for pro search and research

daily

labs is monthly

[languid star]

day

[ebon slate]

mine says the same, most probably not true maybe, because i just used up my 4 search pro enabled in past 30 mins

[twin grail]

pro search are unlimited

[dry kestrel]

#💬│general message

[ebon slate]

alright

[ebon slate]

no ig, its 300 per day maybe

like bred says they might impose limits once we exhaust 300

[twin grail]

They increased to 600

But if u divide it

Into 24 hours

[worthy heart]

how do you get the pro badge here on discord

[twin grail]

Ig it’s basically unlimited

[dry kestrel]

Check pins

[twin grail]

So no ?

[dry kestrel]

Hey hey Kari, read this #💬│general message

[twin grail]

I’m not quite sure abt this

[twin grail]

Ye i did

But what about the api

Showing 600

[dry kestrel]

Bug

[twin grail]

I see

[worthy heart]

thanks

[dry kestrel]

Check out #👑│pro-lounge

[ebon slate]

i only hope there are no limits on how many pro searches we can use per hour

[twin grail]

There aren’t

[dry kestrel]

You would need to spam queries to reach a rate limit

[twin grail]

https://tenor.com/view/spam-gif-10811817447080815426

[dry kestrel]

delicous

[ebon slate]

I won't. If i study for a 3 hours period. The max I can exhaust is 50 queries. So if that's not limited I'm good to go

[dry kestrel]

Yeah you are chilling with that usage lol

[left stream]

Bro I’m gonna workout for the ippo physique holy sh

Once I cosplay as ippo its over

https://tenor.com/view/ippo-hajime-no-ippo-boxing-punch-fight-gif-16417015

[ebon slate]

The good thing is that cloud sonnet 4.0 PPLX has thinking for pro users. I hope that really helps

Never used thinking model in Claude as free users don't have access

[twin grail]

it's good

[haughty birch]

ask Comet and ask it to do for you.

[ebon slate]

Thanks for the idea. I've just asked. Let's see how far comet can do

[haughty birch]

yes, login to your perplexity account , open assistant right side tab and say it "Control my browser and add pro badge to my discord"

[haughty birch]

if it cant do this simple task, then comet is a big piece of* big sheetttt