💬│general

[sly osprey]

And sold the data to Adobe

All your base r belong to them

[boreal dew]

again???

[orchid lily]

someone has aravind's no.?

[young pivot]

GOOD buye guys ...need to study or surely wiil get a F

[torpid sun]

Wait guys

Let’s dissect this

Adobe

[pale salmon]

AI ate my homework is a valid excuse

[torpid sun]

A dobe

A doge

Elon

[young pivot]

2025 excuse

[torpid sun]

“Studying is doubting your potential” homeless jimmy outside of the department of education once said

[brittle schooner]

I also got this
0 enhanced queries remaining today

[torpid garden]

is it down again chat

[strong tiger]

I also got this
0 enhanced queries remaining today 😦

[torpid garden]

ohno

@zinc quartz is the website down again :/

cant see library

[boreal dew]

#🔔│status message

[spare lodge]

Has been down for a couple hours already 😩😤

[torpid garden]

oh

[crisp roost]

days

[boreal dew]

(i quoted the wrong message at first lol)

[torpid sun]

Guys is it that serious?

They’ll fix it soon

[torpid garden]

my bad

thanks

[boreal dew]

and my bad, no worries😂

[void sorrel]

F

[torpid sun]

E

[green crystal]

Huh now it's down entirely

Did someone pull off a gitlab and sudo rm -r the entire infrastructure?

[torpid garden]

it was me

[torpid sun]

The new intern

[pale inlet]

Still down

No library, no spaces, all threads lost

[green crystal]

I just let my sub renew lol
Hopefully it actually gets fixed in the weekend

[crisp roost]

I’m giving up hope

Just wait for the next day

Or next decade

Idk

[green crystal]

They said they're working in the weekend

I think someone really deleted everything...

[pale inlet]

Worst part is if I just had previously generated responses at least it could work

[finite nymph]

Guys, a tip while Perplexity is unstable, the computer version seem to be working better with some limitations

[south aspen]

@wild palm hey

@wild palm Will the new GPT image generation that was just be released be added to perplexity pro?

[keen obsidian]

@wild palm do you know about Perplexity?

[south aspen]

probably down as well while they fix stuff. 1 more try

/askperplexity are you working

[ruby epoch]

Hey guys I just installed discord back again and joined this discord to ask if perplexity is down and if others are also facing this frequent interruptions

[crisp roost]

#🔔│status message

[spring cobalt]

Bros

Perplexity works for me now. Check it up

[ornate umbra]

Hey! Can you change the model to search?

[elfin gate]

I can on android app

[sullen vortex]

0 enhanced queries remaining today, btw 0 today XD

[spring cobalt]

:ccc nope. I got all excited since they've fixed the threads and stuff

[blazing sleet]

threads are back in web

[ornate umbra]

They are actually working on android app and on web too

[gilded gull]

Hey i just create an account with Revolut but I had an issue with it, do you know who I can contact?

[sturdy skiff]

Are there any Claude users here?
I’m curious — what happens when we exceed Claude’s context window in a single ongoing chat?
Will it continue the conversation using the most recent information that still fits within its context limit?
Or will it break the chat and force us to start a new conversation, making it impossible to continue in the current one?

[elfin gate]

It will wave his white flag proudly

[sturdy skiff]

it means we cant continue chat in that window anymore?

[elfin gate]

I have no idea,just a bad joke about the french..

[sturdy skiff]

ok

Are there any Claude users here?
I’m curious — what happens when we exceed Claude’s context window in a single ongoing chat?
Will it continue the conversation using the most recent information that still fits within its context limit?
Or will it break the chat and force us to start a new conversation, making it impossible to continue in the current one?

[elfin gate]

Go to feedback pro,bunch of knowledgeable gentlemen there

[elfin gate]

.

*pro lounge

[sturdy skiff]

but that is for app feedback, not to disucss this claude in perplexity

[elfin gate]

Pro lounge

[sturdy skiff]

cant find in my channels

[elfin gate]

Go to all channels and click everything.
Or buy pro sub if you don't have

[regal prairie]

Is it worth switching to chat gpt plus?

perplexity pro is so limited now and always down. Has completely ruined its functionality, especially when I need it most

[left jasper]

If auto-mode is the future, I think that knowing what mode/model was used in the answer is crucial. Added a comment about it: https://www.reddit.com/r/perplexity_ai/comments/1jm2ekd/comment/mk97f47/

If that is not clearly available then auto (the best mode for your question) can't be reviewed and people will not trust it fully.

[elfin gate]

Amen

You are our representative from this moment.

[regal prairie]

I'm talking about actual functionality. I actually really enjoy the idea of perplexity, of its use of different models, accessing real time information from the news, and the spaces. When it comes to dependability though, it has been terrible. Always being down and randomly not being able to access files which leads to a lot of hallucination , but hopefully they fix infra issues?

[static crater]

I didnt even know pro lounge was a thing. How do I gain access? I've been a pro subscriber for months

[elfin gate]

We voted u in

It's like a sect here

[left jasper]

You need to leave the discord server and re-join using the link on your settings page (when you are pro user)

[elfin gate]

Yes ,you're def the one who will ask all important questions to the CEO on reddit

[cobalt vale]

I thought Auto just uses Sonar, sometimes with pro mode

[distant aurora]

What a wonderful way to save tokens and waste pro searches simultaneously.

Lol.

[rough dove]

it was such a short time for both though. I get that 4.5 cost a lot, and they seemed to have an issue at start with it not actually counting number of uses. but i assume deep research high uses their own r1 model, and they'd have factoreed cost and compute requirements in.

there's several models of sonar on labs. does anyone know which are being used for perplexity search etc. they have thinking.reasoning models on labs, but that doesn't seem to be an option for search (just the r1, o3 mini)

[regal prairie]

how do you guys take advantage of complexity if you use it

[green crystal]

So let me get this straight there won't be a reasoning mode anymore and instead it'll just be a single pro mode where if you select a reasoning model that only then it performs reasoning am I correct?

[unborn viper]

wait 4.5 was removed? i thought it was a bug or something

[elfin gate]

No idea

[bold imp]

well they are not very communicative I tell you.

at least with bad news

[unborn viper]

yeah lol

[green crystal]

Well the amount of stuff breaking down seems to be well explained

They're switching from python to golang

[zinc quartz]

The user gets all the same functionality. This is just a simplification of:
• Auto - AI chooses the model for you
• Pro - you can choose whatever model you’d like (sonar, 4o, R1 for reasoning, o3 for reasoning)
• Deep Research - long, comprehensive reports vs quick answers

[green crystal]

Thanks for the clarification. I just got back from letting Claude explain it to me hah

[zinc quartz]

Damn I lost to an AI again

Also the added benefit is that pro is sticky aka doesn’t default to Auto for follow-ups

So for those of you who want to keep reasoning on for your work, you won’t continually have to change it back to reasoning again

[green crystal]

I hope the team can manage to release the new deep research not too far from now
I have quite a few tasks for it

[zinc quartz]

This was our approach to both simplify + address your frustrations re: default-auto for follow ups

[zinc quartz]

totally fair feedback imo. This is why aravind pulled me in on Sunday for 4 hours to discuss how we can improve this going fwd

I gave up speedrunning fall out 4 to manage yall lot

[green crystal]

Hope you guys also work on announcing all events, ongoing issues and progress more often.
And in more places like in the discord or official blog

[shut dagger]

heyy @zinc quartz , there's one issue. In spaces, whenever I send messages like "Hi" or "Hey", it responds with default Perplexity intro message, it breaks my character which I set in Space's instructions.

[green crystal]

Doing it for the current situation would've prevented quite a bit

[bold imp]

If you let me know, say, three days in advance — like, 'Hey, we talked it through and realized it's not economical and the benefit doesn't justify the cost' — I'm totally fine with that. I just want to be informed, rather than sitting there thinking, 'Wait, that’s gone too?

[zinc quartz]

Agreed. Some change log or weekly updates on changes and why is needed here

[green crystal]

Gemini update logs sound a bit corny sometimes but they're very transparent

[bold imp]

Yeah, it feels like I have to hunt down every single addition or removal, feature-wise.

[zinc quartz]

It may sound trivial to some but give us a bit of time. There’s so many teams moving so fast that even making sure we know exactly what hits prod and when and why is a Herculean effort

[zinc quartz]

Def not what we want yall to do

Not the type of Easter eggs we want yall to find

[zinc quartz]

Mind sending me a few screenshots or the URL to it?

You can DM

[shut dagger]

please disable this for spaces, instead set it for regular threads

[zinc quartz]

Mind inviting me into the Spaces? If it’s sensitive, can you jsut take a screenshot of the whole screen so I understand what’s being asked. I’m very confused on what I’m seeing here

[shut dagger]

matter of fact, in regular threads, it has no such default message

[shut dagger]

okay wait, i'll send you whole screenshot

[zinc quartz]

Just DM me

[dapper pine]

How do I access other AI tools within Perplexity, such as Claude AI, Chatgpt, or Copilot?

[shut dagger]

these are Pro features... $20/Month

[dapper pine]

does it give me direct access to them? I dont understand

[shut dagger]

You get AI models, like GPT 4o, O3 mini (both from ChatGPT), Claude Sonnet 3.7 +/Thinking (Best model from Claude AI), DeepSeek R1, and Sonar (Perplexity's auto model), and Gemini 2.0 Flash (Best Prod. model from Google)

[dapper pine]

oh

thank you

[shut dagger]

if you have pro subscription, you can select one of these pro models and have convo.

[dapper pine]

ok

[whole cape]

Guess the discord bot isn’t working 🤷‍♀️

[shut dagger]

it went down with perplexity earlier

never coming back, RIP

[left jasper]

Although it is working. In #1354883649882816563 . Might be that they limited it to one channel.
That said, ❓ doesn't seem to work at all

[shut dagger]

it kinda shies away from certain questions

[left jasper]

It is a public channel so I envision the system prompt sets the mood 😛

[zinc quartz]

We turned it off on #💬│general so people couldn’t spam it esp given people are tuning in for outage news

Not to hear yall try to get it to say unhinged shit

If you want to chat with it, #1354883649882816563 is always open

[left jasper]

It's only unhinged if you don't agree?

[sleek vortex]

it doesn't work in DM either

/plex

[zinc quartz]

You have to tag it in its own DMs to get it to reply

That isn’t a feature. We will enable DMs at some point

• allowing you to invite it into your own private servers to use for whatever products yall want

[vapid elm]

Hey anyone on the team, is Deep Research High off currently? Is it coming back?

[hollow pelican]

I guess that's a yes. 🤷‍♂️

[latent iris]

prolly

gm btw

[plush valve]

i don't see no o3 mino

[drowsy fox]

when they gonna add gemini 2.5 pro 🤔

[left jasper]

It was all a rouse. They asked for understanding to deploy a "new infrastructure" with all the sporadic outages through the day. Just to give you... the same offering you had yesterday 😄
Unless the new "infra" is live within the next 24h, the statement seems more like damage control than an actual planned deployment.

[minor shard]

I believe they said they had to revert some changes because of the issues and that they were going to work over the weekend to resolve them and push it all back out.

[vapid elm]

new infrastructure is not something you’re supposed to see

first infra then the features that take advantage of infra

[left jasper]

Difficult to know where to keep up with the updates. #🔔│status does not have any news indicating a rollback/delay

[left jasper]

Exactly. A company that has the option to do live "maintenance"... That's just an excuse for poor planning.

[vapid elm]

I wouldn’t bite them too hard tbh just from my experience. Judging by the status channel this doesn’t actually happen often so it might be a one off thing

still definitely seems like it was an engineering or devops mess up than a failed feature launch or something like that

[zinc quartz]

I'm chatting with infra team to give you guys an update shortly.

[zinc quartz]

A new deep research product is indeed in the works and it requires pulling from so many existing and new systems that some parts of our infra had to rebuilt so the 20-30 minute query doesn't fail + all work is lost. Proper backups and caching needed to be introduced.

[fierce grove]

does anyone else also get this error message when uploading attachments, it also says daily file upload limit 0 even though i havent used perplexity from 2 days

[zinc quartz]

What are you trying to upload?

[fierce grove]

png and pdf

used to upload just fine a week ago

it only began happening since yesterday

[zinc quartz]

what device? and is this free or pro?

[fierce grove]

pro version and on a macbook, both the desktop app and on the web

can anyone help?

[vale scarab]

I want to create a YouTube automation channel. Perplexity creates content that allows me to create a successful YouTube channel.

[zinc quartz]

Let me escalate to team

[keen obsidian]

@fierce grove can you share what you are uploading? This typically happens for explicit content or in a failure case, something that is classified as explicit content errorneously

[fierce grove]

a pdf containing a play i have written to be analysed and png of a poster to get suggestions on improvements that can be made to it ( graphic design )

it failed with them adn i even tried various other things too like txt files and stuff

[shut dagger]

It's under reasoning option.

[left stream]

That color looks dope asf

Try ai mode on google

I forgot what ai model it is

It’s definitely a gemini 2.0 though

[proper osprey]

wow man did you seriously read everything that happened since this morning to end up seeing my post over 500 messages above? 🙂 --- Thanks for the compliment! I really enjoy perplexity and the options it provides with a nice and easy interface 🙂

[pearl scroll]

Can. We generate ghibili photos using Perplixity

@tranquil orchid

[left stream]

No problem and agreed, I enjoy perplexity too. Looking forward to renewing my subscription with them soon 🤘

[proper osprey]

and the color scheme (along with other features), in case you didnt catch the person that answered first asking about it, comes with "complexity", a free extension that adds a few dozens new options, features, ui changes, themes, etc... it really makes it that much more amazing 😄

[short gale]

@zinc quartz Thank you for providing updates in #🔔│status and using the appropriate @ role to do so.

[amber dragon]

Sorry can you tell me how this is done ? Thank you

[rocky cape]

Damn we can’t swear in this server wtf

[tall galleon]

Your AI is always bugging, I don't know why it gives me half the answers.

[dark oasis]

chatgpt

[tardy junco]

I’ve been mining my 5000+ conversations using BERTopic clustering + temporal pattern extraction. Implemented regex based information source extraction to build a searchable knowledge database of all mentioned resources. Found fascinating prompt response entropy patterns across domains

Current focus: detecting multi turn research sequences and tracking concept drift through linguistic markers. Visualizing topic networks and research flow diagrams with D3.js to map how my exploration paths evolve over disconnected sessions

Has anyone developed metrics for conversation effectiveness or methodologies for quantifying depth vs. breadth in extended knowledge exploration?

Particularly interested in transformer based approaches for identifying optimal prompt engineering patterns

Would love to hear about ETL pipeline architectures and feature extraction methodologies you’ve found effective for large scale conversation corpus analysis

[open wyvern]

whut the promt?

[tame current]

I love purpurlexity

[ivory nova]

Unpopular opinion: I've been using Perplexity consistently and haven't visited GPT in nearly 6 to 7 months. Due to recent downtimes of a few days, I thought of using GPT, and it's damn buttery smooth.
Now I'm thinking of moving to GPT, but I do love having access to Claude for coding-related tasks.

[visual latch]

Hey there, I am using perplexity however it has experienced frequent downtimes in the past one week. For some reason its mode changes to auto once i ask it any question.

Perplexity is bette r than chatpgpt. i love it. but due to recent issues I am dissatisfied

Whenwoud this be solved

[distant aurora]

I think there's no fixed answer to this. I tried O3 mini high on Merlin and got a different result https://www.getmerlin.in/chat/share/42a5b738-8700-4e5b-88c0-ea2d4c61fc73

Deepseek R1 https://lambda.chat/r/y_12TJP?leafId=19c8ffd0-80f2-41a2-b56b-e1366d200783

O3 mini (via chatgpt) https://chatgpt.com/share/67e7a31e-690c-8013-9227-9c6bdf8a3ead

[golden geyser]

Guys can someone with GPT 4o restyle my photo to ghibli style?

[zinc quartz]

Totally get it. There's been quite a few outages. Aravind posted a reponse on reddit addressing a few of these concerns

[distant aurora]

I thought free users already have 4o access.

[golden geyser]

Yeah. But only + and premium user can restyle images. OpenAI block all request for free users

OpenAI GPU are melted

[distant aurora]

Send image

Lol.

Or maybe just use some 3rd party Ghibli filter.

[distant aurora]

I read that. Didn't have any definite answer about the outages as far as I remember. Only something about reworking deep search.

[river bronze]

Is odd the reason given is a new 30m deep research but we don't have that yet so how is it causing outages if no one is using it

Perplexity is a marketing company primarily, even when apologising for the site being offline it is in the form of an advert for a new feature 😄

[zinc quartz]

Sure but the team can either spend the 24 hours after major outages hardening the systems that failed or figuring out what really led to these failures and figuring out a path forward. Our priority is to do the former first and then get back to everyone with clarity on how we'll mitigate this after.

[zinc quartz]

outages aren't simply caused due to usage of a product. developing a totally different product requires major changes to existing infra to be compatible + framework change from Python to Golang. These changes were made quite hastily, thus affecting the functionality of the existing platform. Could we have done this more methodically? Absolutely.

That's our intent over the next few weeks until we're sure the infra issues won't persist.

[river bronze]

So the issue is testing in production instead of a UAT environment with a progressive migration basically

[distant aurora]

@river bronze are you a meat duck or a rubber duck?

Funny how duck can be replaced there. 💀

[zinc quartz]

Most likely we migrated too aggressively w/o sufficient testing in staging + proper checks and unit tests in place; this worked even 3-4 months ago but now we are way past that and existing processes are breaking.

[wheat pike]

pplx growing too fast for its own good 💀

[distant aurora]

This response makes way more sense, thanks mod.

[wheat pike]

released in 2023 to 8bil valuation in 2025

[tranquil orchid]

Hey @distant aurora!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[river bronze]

I saw someone posting your client if it fails to connect tried to reconnect 30 times a second, so when there is an issue you are creating a bonnet of clients sending you a DDOS, idk if that helps lol

[zinc quartz]

LOL yes, in some ways our webapp was DDOS'ing us + our DB

[distant aurora]

The rate limits are quite generous.

[wheat pike]

yea definitely

[zinc quartz]

We def flew a bit too close to the sun with our shipping velocity and desire to experiment to find the ideal user experience + design for this new world of AI-powered search.

[wheat pike]

prolly how it grew so fast

but what worked then doesnt work now

[zinc quartz]

The very thing that users loved about us has led to the cracking and breaking of existing systems, processes, and even the duct tape we put around some parts of our stack.

Huge wake up call for us

[wheat pike]

better late than never

and what a timing with your arrival too lmao

[zinc quartz]

Also, many people don't know that we're around ~200 people. Anthropic is at 1,000 and OAI at 3-4K people

[muted pine]

lot of error today

try to make a research avanced

have only error

[wheat pike]

oh wow

[mystic reef]

add gemini pro

[mystic reef]

hustler

[zinc quartz]

if PPLX is down, yall always have poogle.ai

https://poogle.ai/

[mystic reef]

im just grateful tbh i dont even know how my wifi service provider gifted me a whole year of perplexity for free

[wheat pike]

did you guys buy this domain recently

i swear i tried it yesterday and it didn't work

though that might have just been because everything was on fire

[zinc quartz]

oh nah this just leads to prod so if our prod was down, you'd have the same outages lol

please don't think this is a backup instance of prod; it was 100% a joke

[warm pumice]

Is anyone else running into really slow responses from the API?

I get the first word of the response, AI waits a couple minutes, and then continues

or it just ends after first word

[visual latch]

I am getting slow responses from Perplexity for the past one week. hopefully this will be resolved in 10 days and when is grok 3 being released.

[austere granite]

Please do not use ai for "legal expert analysis" 😭

[shell pewter]

how do u use image generation in perplexity?

[river bronze]

That has to be the most asked question here, they need to just add a button for it

[short narwhal]

Go to the image tab and press the big generate option

[short narwhal]

What are people even doing with all these generated images 😭

Absolute slop fest

[shell pewter]

ion see no image tab

[river bronze]

It's not clear because you have to ask a question first when you don't want to chat you want to do an image

[shell pewter]

im tryna have it edit some picture for slides and thats it

[short narwhal]

Are you on the website?

[shell pewter]

oh so i gotta ask sum and then when the answer shows theres an image tab

right?

[short narwhal]

Yes

[shell pewter]

they look mad weird bro im js tryna make some dudes kiss each other for the slides

dw

[short narwhal]

🤨 ey you do you dawg

[shell pewter]

how do u get the pro role btw

i heard u gotta join the discord link off of the web but ion see it either

[short narwhal]

You have to join through the link in your settings in perplexity

[serene sand]

https://discord.com/channels/1047197230748151888/1225557012532695060

[boreal dew]

i did that and it didnt add the pro role for me😔 ✊

[serene sand]

#💬│general message

[short narwhal]

We’re you in the server already before clicking it

[shell pewter]

i don't see the discord button...

[boreal dew]

no, time to leave and rejoin and leave and rejoin and try turning it off and on again😂

[short narwhal]

https://cdn.discordapp.com/attachments/1237372301243191327/1274608969989754911/catBOOMM.gif

[shell pewter]

@boreal dew can u show me where the link is at in the settings

[short narwhal]

This is where it’s on mobile

[shell pewter]

thank u ima try it

[shell pewter]

i got it love u bro

[short narwhal]

Yey

[boreal dew]

tbh i cant even find it anymore on the web application rn
it was there yesterday i promise🤞

ah i might have it

if youre on pc on the web application you gotta go to settings, preference, then pro support contact, scroll all the way down and then join our discord community

they hid it

[short narwhal]

Silly

[shell pewter]

they dont want me to get the role

😞

[boreal dew]

it was a single button before, now they trynna hide it😂

left the server, joined via the link, they dont want me to be a pro😔

[short narwhal]

😭

[serene sand]

you have it already

[serene sand]

https://discord.com/channels/1047197230748151888/1111786888626438245 hop in

[boreal dew]

yippie!! thank you!🥰

[shell pewter]

i said it as a joke bro thanks for the answers either way😍

[valid charm]

Good morning PPLX family

[valid charm]

Infra Challenges: We're working on a new more powerful deep research agent that thinks for 30 mins or more, and will be the best research agent out there. This includes building some of the tool use and interactive and code-execution capabilities that some recent prototypes like Manus have shown. We need a rewrite of our infrastructure to do this at scale. This meant transitioning the way we do our logging and lookups, and removing code written Python and rewriting it in GoLang. This is causing us some challenges we didn't foresee on the core product. You the user shouldn't ideally even need to worry about all this. Our fault. We are going to deprioritize shipping new features at the pace we normally do and just invest into a stable infrastructure that will maximize long-term velocity over short-term quick ships.

[tranquil orchid]

Hey @valid charm!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[tranquil orchid]

Hey @valid charm!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[valid charm]

@tranquil orchid Do you mean the original reddit post?

[valid charm]

This sounds like music to my ears

[verbal scroll]

Sorry if this was already said, but is read aloud gone temporarily?

[vapid crag]

what model is the "GPT-4o" in pro search, is it really "GPT-4o" or "GPT-4o-latest"

[haughty aspen]

is it only me or the assistant on android is very broken?

it often stops working and i can't use it anymore while outside the app, using android 13 with OxygenOS

[rapid bane]

perplexity is worst than a free ai ngl

[shell pewter]

what free ai

[celest nebula]

How to turn off web search?
in a Space: I use pro+ 3.7 sonnet and turn off all web and academy but it still search and get data from the internet

[rapid bane]

any lmao

[valid charm]

Hmmm you are doing it right. It could be a bug due to all the changes being done in the backend

[placid coyote]

yeah, saw this happening to me too. usually if you write it, the pipeline AI understands it. like "without any web search" or "do NOT search on web"

[short jewel]

Hello

[last osprey]

Is it down again?

[lament imp]

Not really sure the general chat of a Discord server is the best place to pitch a business proposition... May be best to do that via email 🤔

[golden geyser]

Can some explain why I can’t upload any file to perplexity?

[left stream]

If you remove anything related to search o3 mini high starts getting it

Initially when o3 mini medium reasoning first came out on chatgpt, in the very very first hours

It actually answered correctly all the time

But then a day passed and it started answering different and got the question wrong almost all the time

O3 mini high without search got it right though

[valid charm]

@zinc quartz spam?

[zinc quartz]

Ew. Deleted. Thanks for flagging

[valid charm]

They are currently rebuilding the infrastructure for file attachments and a whole lot of other stuff. This is why things randomly stop working.

https://www.reddit.com/r/perplexity_ai/comments/1jm2ekd/message_from_aravind_cofounder_and_ceo_of/

[dull patrol]

No, it’s not

[valid charm]

Here is more info on the changes going on now

[zinc quartz]

What’s the error you’re getting? Mind DMing me?

[zinc quartz]

You are totally welcome to introduce yourself and share more about what you’re working on. But if your intent is to source strategic partners, this ain’t the place for it

[golden geyser]

"File upload failed"

[sturdy skiff]

@zinc quartz , atleast please dont mind and share context window size of what ever models you have information for from perplexity for pro subscription

[sturdy skiff]

i need atleast one reasoning model to have atleast 128K or 200K at minimum, atleast if one reasoning model supports that, life is atleast sorted until your next updates about context window size increase.

Note: dont say Auto mode, that is not powerful for my use case, it never impressed me

[agile depot]

Hello

[sick arch]

@zinc quartz

[orchid quartz]

Is reasoning with r1 better or with 03 mini?

[sturdy skiff]

i am exactly in middle of testing that

[orchid quartz]

Yeah just now reading your message.

What is auto mode Phil?

[sick arch]

sonnet 3.7 extended thinking

[inner pine]

hey guys this might be a stupid question but .

if I need the most up-to-date data, which API model should I use? I need detailed and precise, up-to-date data.

[orchid quartz]

Is that the deep research mode?

[sick arch]

nope, in reasoning mode

[orchid quartz]

But that’s not available

[sturdy skiff]

actually all 3 reasoning models in perplixity are giving truth answers for my complex questions, so you can use any of them. i personally fan of openai, so will use o3 mini

[orchid quartz]

Only o3 mini and r1 are available

[orchid quartz]

What model does deep reasoning use?

[orchid quartz]

Oh the ios app doesn’t have access to to that yet sorry 😭

So sonnet is better?

[sick arch]

share ur ss

[sturdy skiff]

if you ask me, all three are good

[sick arch]

best amongst all available

[sick arch]

update ur app

[sturdy skiff]

claude is more good in listening to your custom instructions, o3 mini also equally good, R1 i dont use more @orchid quartz

[orchid quartz]

Updating now

Is the r1 the biggest available deepseek model running on perplexity servers?

[sick arch]

did u find sonnet

[orchid quartz]

Yeah 🙂

Thanks

[sturdy skiff]

perplexity might be using o3 mini low or medium, so i will also not suggest you to use it, so you can try R1 or Claude 3.7 thinking, i will use o3 mini until it hurts me, as of now it didnot hurt me

[sick arch]

what u use the most

[sturdy skiff]

for reasoning i use o3 mini, and deepresearch , i never use Auto and Pro because those both are giving wrong answers

[orchid quartz]

Where is o1 in the graph?

[sick arch]

do u find o3 mini response on plex better than sonnet?

[sturdy skiff]

o1 is underrated legend, i used to use it a lot when i used to have chatgpt pro subscription, yeah they didnot mention in graph, but o1 is the intelligent model i ever used

[orchid quartz]

O1 is the best model available right now right?

Not o3 mini

[real knot]

When presented with the same math problem, its performance is usually at the 'o3mini low' level, though it occasionally reaches 'medium' (I don't know how to trigger this).

[sturdy skiff]

o3 mini has more human way of thinking bro, so i use it for reasoning, claude is too much technial thinking, for example 1+1 = 2 then claude says it is 2, but o3 mini cross check if for this particular scenario the answer may be something else and gives the right answer, i saw this observation multiple times

[orchid quartz]

I’m an engineering student so I really need the right answer to my questions lol 😭

[sturdy skiff]

as per leaderboard o3 mini high is best, but o1 is unknown legend

[orchid quartz]

They shouldn’t give me the mini version

[sturdy skiff]

for 20USD no one can provide o1 brother

[orchid quartz]

Woah didn’t know that, when I was using chatgpt plus I always used o1 lol

[orchid quartz]

Chatgpt could

[sturdy skiff]

if they are using o3 mini medium, it is equivalent to o1 , but we dont have answer from perplexity whether they use medium or low

[sick arch]

whats ur use case, coding or general search query use?

[orchid quartz]

I have perplexity free true school

[sturdy skiff]

not at all coding, my use case is deep research of uploaded files and making decisions out of it

[sturdy skiff]

for coding people can straightly go to chatgpt, claude , gemini2.5pro, grok3

[orchid quartz]

Can we use deep reasoning in spaces already?

[sick arch]

i have used 3.7 sonnet for coding, its really good.

[sturdy skiff]

just imagine you wrote 3000 lines of code and perx forgot dur to context window issues and screwedup your code?

[sturdy skiff]

yes

[orchid quartz]

What model has the largest context module

[sick arch]

its limited to 32K in plex

[sturdy skiff]

this is what we are asking @zinc quartz ,

[sturdy skiff]

@zinc quartz , please mention us atleast one reasoning model which has 128K or 200K in the perplexity, thanks in advance

[orchid quartz]

Who is kwak lol?

[placid coyote]

they say 32k in docs, but in reality in search mode (not spaces) it can inject to context a beginning of a file 62k tokens worth of text. not guaranteed, but behaves like this for months. some more numbers: https://monnef.gitlab.io/by-ai/2025/pplx-tech-props

[sick arch]

https://tenor.com/view/avengers-infinity-war-captain-america-chris-evans-robert-downey-jr-gif-10724671

[orchid quartz]

Whahaha best answer ever

[sturdy skiff]

bro i got confirmation over email that, they are not talking about chat context window, that 32000 is upload files context window for pro subscription users. so chat context window officially no one knows

[placid coyote]

if only that promised 1 million context worked...

[sturdy skiff]

that also they confirmed in email as "1 million context works only with Auto mode" nothing others

[orchid quartz]

Omg web beta tester

So you use the agent webbrowser?

[placid coyote]

that is from my testing. model can see text correctly upto around 62k tokens. that is no PR, just facts

[orchid quartz]

@placid coyote

[sturdy skiff]

then fine, thanks for letting me know, and is it true with reasoning models as well, that is my requirement

[placid coyote]

I was trying mainly with pro, can test with reasoning.
though as I wrote - 32k is what they have in ads, docs, etc. so they might lower in the future. similarly with "300+" daily uses which not counting like two weeks was always 600uses per day

[sturdy skiff]

can you pleas test for reasoning and let me know which model in reasoning you used, in your testing if you found its bigger context window please let me know the value of it, i will not blame you if it didnot have that much, but atleast some peace while using it

[sturdy skiff]

@placid coyote please let me know in personal DM your fidnings , here i am hvaing discord bug, even though people tag me, sometimes those msgs are not visible

[rain ruin]

Hello kids

[dapper pine]

When emailing [e-mail] am I talking to a human or an ai bot?

[bronze mountain]

Yes

[zinc quartz]

R1 and deep research has 128K input context

Some models have 32K

[sturdy skiff]

Can you please let me knoe about o3mini reasoning and claude3.7 reasoning as well

[placid coyote]

hmm, not sure what is happening here - my query vanished and response doesn't look finished. maybe 128 steps is some limit?

[sick arch]

refresh ur page and go to library, u will find this query

[real surge]

it's probably been asked a million times this week, but can i do the ghibli thing using perplexity pro 😭

[sick arch]

i mostly do it with gemini, its works amazing there

[placid coyote]

hmm, it lost all steps and no response. but yeah, at least it didnt swallow the query. I learned to backup every query to clipboard the hard way (once lost a query I was writing for like 15 minutes; my clipboard manager keeps a long history). perplexity used to lose them much more often than now 😅

[harsh verge]

CPLX has an auto-saved prompts feature

[dapper pine]

which one I'm sorry?

[placid coyote]

I know, it is a nice feature. though even preplexity implemented some form of prompt persistence. but it doesnt work that well with multiple pplx tabs I think. last time I tried Complexity, I think I found out I would have to implement special handling (to not collide with cplx) in my tooling (perplexity helper and ailin)

[harsh verge]

Yes, indeed, multiple extensions can sometimes conflict with each other.

[foggy stone]

hi

[real surge]

gemini using perplexity?

[sick arch]

https://gemini.google.com/

[real surge]

what promt do you give it? bc it's giving 'inspired image', and not a converted one

[sick arch]

yes it doesn't convert, thats only downside😐

[real surge]

bruh

[harsh verge]

For image editing with Gemini, you need to use "Gemini 2.0 Flash (image generation) Experimental" in Google AI Studio.

[sick arch]

It's worse there😅

[harsh verge]

It works really well. I'm not particularly into Ghibli-style things, but for practical purposes, it's a great model for image editing.
(https://developers.googleblog.com/en/experiment-with-gemini-20-flash-native-image-generation/)

[twin herald]

@tranquil orchid

[left stream]

@distant aurora which response was better in your opinion, here is the original request: How does a nuclear reactor work, give the best explanation possible. The goal is to make whoever is the reader to be able to fully comprehend how it functions and understand it conceptually.

[distant aurora]

Response 1 I think.

Though I feel like the basics of fission should have been explained first before that ceramic structure statement.

And the use of control rods was better explained in response 1 I think.

Depends on the purpose I guess.

Response 2 keeps it simple to understand.

Response 1 is a more complete answer.

Shiit is hella expensive.

You get 1 million points per month for around 20usd

So roughly 19 prompts for 20 dollars.

[azure abyss]

when will be the grok 3 added?

[distant aurora]

Grok 3 is not worth it apart from the deepsearch.

The model didn't live up to the hype of the "best model".

Benchmarks were likely rigged.

Though grok in itself is a very good conversation companion. Doesn't have that bot-like tone.

[indigo garden]

In my opinion they could remove the sonar lol and put the Grok3

It is much faster and is also very accurate in searches.... Although I like the deep search, Grok3 would be better suited to the Pro search.

[distant aurora]

Sonar reasoning or normal Sonar?

[indigo garden]

The normal one...

Although Sonar here in Brazil at least, only has the normal option

[distant aurora]

Huh. I guess I'll agree with you there.

Llama 70b isn't very impressive.

[indigo garden]

It's a model that, between you and me, nobody uses, right?

You shouldn't use it, I don't use it, it's a bit outdated... And little known... In programming it doesn't do well either... In research either... And since you already have Grok2, just upgrade it by removing sonar.

[distant aurora]

They ain't doing that cuz it's signature model of perplexity. They won't just dump their in house project I believe.

And they have their claims.

It is also very likely a part of effective cost management.

Though, some upgrade/improvement in the in house model would be appreciated considering the current standards.

[short narwhal]

1 million output tokens is 600 dollars yeah

[short narwhal]

no api yet

[distant aurora]

They're most likely busy with the deep search revamp.

[distant aurora]

Wait. Really? I thought it would be available by now.

[short narwhal]

lmao no

[river bronze]

Sonar is the model that does all the searches I believe then passes the info off to your chosen model. So everyone uses sonar

[short narwhal]

at this rate it won't come out ever

and even when it does, multiple models are already better than it

no reason to use it really

[river bronze]

Yeah Grok missed its moment as being the top model by just not being available to use at all

[distant aurora]

Except deep search.

[river bronze]

Now everyone else is well ahead

[distant aurora]

Grok is more of a political research tool imo.

[short narwhal]

true, but now google has its deep research for free with 2.0 flash thinking, and GPT also is soon free (with heavy limits of course)

[distant aurora]

I really like how uncensored it is.

[distant aurora]

Google's limits are like 5 free searches per month.

[short narwhal]

is it?

[sleek vortex]

gpt is gonna be like 1/mo ngl

[short narwhal]

3 a month is what sam said

[sleek vortex]

oh right

[distant aurora]

Still grok is way cheaper even if free.

Perplexity had good timing with the deepsearch.

And the rate limits seem generous compared to competition.

[short narwhal]

Wonder if free users can get any high deep research uses

[distant aurora]

Bro, I feel like open ai probably bribed Deepseek to stay out of competition.

[river bronze]

How much is Grok? Gemini is part of the Google one for like £18 a month and has deep research

[short narwhal]

30

[river bronze]

With that you get 2TB cloud storage and tons of other perks too

[distant aurora]

I'm a free user. I don't see any high switch on perplexity.

[short narwhal]

yeah it's not yet a thing really

[river bronze]

Perplexity deep research high does not exist yet, the one in complexity just uses standard as high isn't available. The complexity dev told us today

[distant aurora]

Grok gives 5 deep search per 2hr for free.

Wild stuff.

[short narwhal]

makes sense

[river bronze]

So everyone who thinks they are using high are not 😄

[short narwhal]

😭

[distant aurora]

Grok rate limits are very strict though. Even for basic.

Have to downgrade to grok 2 after like 10 to 15 prompts.

[sleek vortex]

perplexity is ab testing between 2 deep research types

there is the one type where it shows a list of tasks like grok sorta does

and the other which is just one like linear list of things like it was before

[sleek vortex]

what does this even mean lol

[river bronze]

The testing has also destroyed their infra lol so I guess it's on pause for now

They researched too deep it burned all the GPUs

[distant aurora]

Deepseek is back already. They could have fixed their web search by now. I feel like open ai paid them to stay out and not fix the web search.

[short narwhal]

crazy we were supposed to get comet waitlist rollouts last week + high possibly but got neither in the end...

[sleek vortex]

it makes 0 sense for openai to bribe deepseek to "keep websearch off"

web search is like such a small part of their offering, just a feature on their site

the real threat in deepseek is well, the company, the talent they have, the model(s)

[short narwhal]

^^

[sleek vortex]

the fact that they OSS, so on

[river bronze]

Openai can't steal their talent it's okay, deepseek confiscated their passports

[sleek vortex]

web search being dead for a week is probably just a deepseek issue, maybe their search provider went down

[distant aurora]

It's about the subscription. The official website kills the price competition. Most sources that offer R1 charge a fee.

[sleek vortex]

You either use inference from china, or pay some 5x price for US hosted

They're slowly figuring out how to run it faster in the USA but

It still costs way more

[river bronze]

Or you use OpenRouter who have search access for every available model

[distant aurora]

Too slow

[river bronze]

And only pay what you use no subscription

[distant aurora]

I tried open router.

You are better off using chat.minimax.io

Or Blackbox ai.

Both offer R1 with web search for free for now.

[river bronze]

OpenRouter is great I almost exclusively use it for my projects. Great latency and througput

Ofc milage may vary

[short narwhal]

free or are you paying?

[distant aurora]

Likely paying per tokens. Free is sluggish.

[river bronze]

Bit of both, small personal stuff I use a free model but the rest I pay

[short narwhal]

Is there some app I could just plug all my api keys into

[river bronze]

OpenRouter is one API key for every service, they are like a middle man routing the requests

E.g. I use Gemini Flash on it a ton

Instead of paying Google

I do believe if you want though you can provide your own keys via them if you really wanted to

[short narwhal]

how does that differ from just using AI studio or gemini for that matter?

[river bronze]

I'm talking about using the APIs in code mostly

[short narwhal]

Also as a sidenote I don't like gemini in the sense that it looovess to yap

[short narwhal]

ah

[river bronze]

But for their web chat it's also kinda convenient just all in one place no subscriptions

[short narwhal]

fair

pop off deekseek

[distant aurora]

What's this?

[short narwhal]

openrouter

[distant aurora]

What's with the dates? Got stuck in some loop?

[river bronze]

The best deepseek is V3 with a bunch of numbers on the end

That's the latest one

[distant aurora]

Just call it 3.1

[river bronze]

Pro max

[distant aurora]

~ Claude 3.5 sonnet in coding

Impressive considering free.

I hate Claude cuz of the rate limits.

[fickle flicker]

does anyone have a favorite service for storing embeddings? I am storing email data and calendar information as context for an LLM

[torpid garden]

can we expect option for mcps in perplexity?

[regal coyote]

where is voice mode in the app, guys? what happened?

[tame current]

i think they workin on making it more 'agentic'? i would also really like to see that feature

there are workarounds tho

[elfin gate]

Hi

[tame current]

yo

[river bronze]

They have their own MCP they offer so I would hope so but they have never said

[radiant laurel]

how do i add the perplexity app to other servers properly? tried adding it to another server and i got errors when i would query it

[left stream]

Damn

[left stream]

Gotcha, I wonder what model was 4.5 and what the other model was.

Response 1 might’ve been some new model

[short narwhal]

doesn't seem like it yet at least, works in dms though :P

I made my own slop @grok gronk gork @wild palm version so if you really want it you can do that

[hollow mulch]

hi guys.

[elfin gate]

Hola

[hollow mulch]

yo.

[left stream]

@distant aurora try claude 3.7 extended on perplexity, it’s finally on ios I barely realized 😭

[frail dust]

@wild palm is there a way to prevent Grok from being one of the choices that Pro uses for my queries and prompts?

[left stream]

Definitely liking it a lot

[short narwhal]

no

[elfin gate]

Grok?

[worthy harness]

When are we getting Gemini 2.5 and DeepSeek V3.1

[dense crest]

How many threads can we have in library?

Are they unlimited?

[worthy patrol]

Hello:)

[frail dust]

The AI from Twitter. I don't trust their data security and don't want to bring an API that touches twitter into my use of Perplexity.

[elfin gate]

So Elon is spying on people

Interesting

[valid charm]

What social networks are search with the 'Social' toggle? I've never seen anithyng other than reddit

[edgy obsidian]

I feel yes in manage account u can use AI of your preference

Go. In settings
In manage account
AI SETTINGS
you can change default to what ai u prefer

I want to use perplexity for sports betting but they don’t have specific AI FOR IT

[left stream]

I never thought about using ai for sports betting, that’s genius. I’ll use it to bet on football

[edgy obsidian]

It’s not the best but it’s gives some idea

[austere granite]

Please do not make financial decisions based on ai advice

[sturdy skiff]

hi guys how is the new update of perplexity, i didnot check performance yet, but those options are good right, they are no more defaulting to Pro, they have heard us repaired for us

[left stream]

It’s awesome. I’m liking claude 3.7 reasoning a lot.

@distant aurora I’ve been thinking of making a report but using all the possible deep research models available online and have them all contribute in it. I wonder what it would turn out after all of them contribute.

[valid charm]

I did this and it came out great. The hardes part was combining all the info into a comprehensive document with all sources properly cited

[sturdy skiff]

Hey bro, othan perplex, gemini, openai, grok what other deep research options you are aware of.

[valid charm]

https://manus.im/

I don't like Manus but some swear by it

[sturdy skiff]

These are expensive things, not for me

[cedar latch]

https://media.discordapp.net/attachments/1089978843470766093/1128717835112296488/voicemessage.gif

[valid charm]

Preach Phil

[left stream]

You.com’s ARI there’s others I forgot since there’s so many deep research to remember 😂

[sturdy skiff]

Ok you.com is providing only 20 ARI searches in a month for 30USD

[left stream]

I have access to ari but mine is glitched out or something. It renews 20 ARI searches within a day

I got mine by having an edu email so it’s definitely a glitch

[sturdy skiff]

So glitch is giving you 20 a day instead of 20 a month ❤️

[gritty siren]

wats up with perplexity man

everyday they change it its getting annoying

now i can only select Pro or Deep research

[sturdy skiff]

See their announcement bro, you can choose model

[quick moat]

what is R1 1776? is it deepseek's model?

if so, then why are they calling it perplexity's model? where is deepseek & why did they replace deepseek's R1

[gritty siren]

My deepseek gone 😦

[unique umbra]

https://github.com/sentient-agi/OpenDeepSearch

I want my momey back fr fr

[dry kestrel]

Its intended

people who signed up via waitlist gets their thing refreshed daily

people who bought teams plan get theirs refreshed monethly

[shell pewter]

with this new update 4.5 is gone again 😭

[wise edge]

The new version of deep resarch doesn't seem to need deep seek anymore, because I found that the illusion of deep research has obviously decreased, which is a good thing~

Does anyone know what models are used in current deep research?

I prefer sonnet 3.7 > gpt4.5

sonnet 3.7 has context length :200K

gpt 4.5：
128,000 context window
16,384 max output tokens

[random ocean]

R1 1776 is a specialized version of the DeepSeek-R1 language model, developed through post-training techniques to eliminate Chinese Communist Party (CCP) censorship and biases.

[random ocean]

As per CEO reddit post, it’s not a single model, but rather different models working today.

[wise edge]

#1354883649882816563 message

[sudden tiger]

Possible to get Promo Code for Enterprise? Alternative Question, can I share my account with my parent?

[distant aurora]

This is literally sonar reasoning.

Wait. Are people literally using sonar reasoning labelled as R1 then?

[distant aurora]

Yeah fr. They should add twitter.

[distant aurora]

Nice https://www.perplexity.ai/search/find-me-assault-rifles-with-fa-oSZBk3OLQa6RTZj6JHoBfw

[distant aurora]

Google, open ai, grok, perplexity, open deep research (hugging chat), open deep search (vercel), storm ai, idk if there are others.

[paper violet]

Perplexity changes UX? any new model added?

[tame current]

math ai.

make your own math ai by yourself. and solve differentiation problems algebra etc...

or train your ai model on my math ai data. anything you want.

i don't know how ai model can do math.. but you can try to train and find some solution to it yourself.

train x
sin(x)+sin(x)+x+x train y
((2*sin(x))+(2*x)) ai model equation simplifier. i tried this but didn't work.

[harsh cape]

grok网页版挂了

[river bronze]

It is Deepseek R1 with some fine turning to remove the censorship that was built into it

[left jasper]

I really wish they would write which mode and model was used when running Pro + Best. I really want to use the full auto setup but not being able to get some insight into how and why a certain mode/model was selected prevents me from going down that route.

[tribal skiff]

Did anyone get double charged by perplexity this month?

[serene sand]

please contact [e-mail]

[tribal skiff]

Perfect thanks

[last cipher]

I think perplexity is down atm.

[patent karma]

currently it working fine

[last cipher]

Sorry, working again fine. Thx

[tranquil orchid]

Hey @last cipher!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[plain patio]

Why does perplexity change the search model availability every 2.5 minutes? Everytime I load the page a feature is gone. I'll be surprised if there is even a search box by the end of the week at the rate they are going. And is it me or does gpt 4.5 not show for anyone else as well?

[left jasper]

gpt4.5 is fully gone for me as well

[elfin gate]

I'm pretty sure 4.5 was removed long time ago

[plain patio]

https://tenor.com/view/smh-gif-smh-meme-smh-steve-harvey-i-can't-gif-13893533684179296052

[elfin gate]

But don't worry,now we have a new defsult model selection

[left jasper]

It was here on Thu and early Fri (although pplx was rather unusable at the time)

[elfin gate]

I've seen it pop out a few times by accident,I think it was bugging and not a proper 4.5

[gritty siren]

tell me abt it ..im annoyed

[desert cedar]

W change ngl

[gritty siren]

it cant seem to even read pdf files properly now

[desert cedar]

The auto model selection was...

Something

[gritty siren]

ya tat was annoying AF lol

[elfin gate]

Dont you like transition from absolute freedom of choice to what we currently have?

[gritty siren]

im surpised they cant get these things right..it has nothing to even do with LLMs and AI

they just have a bad UX team

[plain patio]

Ikr. Wasting all that time removing DeepSeek's CCP seasoning only to ingest it themselves

[left jasper]

It's intentional that they push users to Auto mode - they want the LLM to decide the level of compute that is needed for each users question.
If done right, I can see the usefulness here. But if they don't even show you which mode (reasoning/thinking or only search) and model that was used, it's impossible to know if you should ask it to rewrite using reasoning for example.

[elfin gate]

I am scepticalv

[gritty siren]

exacatly..i understand they want to make even the UX 'smarter' but there is a reason why its called 'Artificial' intelligence

[left jasper]

Me too. And as said, if I can't even know what mode and model was used I'm not going to use it tbh.

[gritty siren]

right now with this mode its getting things wrong such as the tax rate for my country being used...

[elfin gate]

Developers thank you for removing my rights I had before

I'm curious what did they say as a reason for all this

[gritty siren]

im going to request a refund

too tired with their constant changes and not listening

[elfin gate]

If they said we want to have more money then I would understand

[distant aurora]

Chatgpt is down.

https://downdetector.in/status/openai/

[slim jay]

the ai is having problems again, it can't see files I upload with code

[distant aurora]

[sturdy skiff]

What happened bro, i missed things, did they reduce anything for us

[left jasper]

I mean, combining search models and reasoning in a single drop-down is actually better. Now we don't have to pick between 3 modes at least.

[elfin gate]

Haha

Web UI is different

[slim jay]

must be having problems again or is it just me ?

[elfin gate]

You can't choose

[slim jay]

anyone ?

[gritty siren]

the thing is i want to pick u see..i use the reasoning one with deepseek and its the reason why i subbed

the other models are not to my likng and the other modes serve diff purpos

[elfin gate]

You can still always use it

[gritty siren]

deepseek removed on web ver

[sturdy skiff]

There is option to choose bro, left side select pro, right side there is option to choose models. Even they showed in screenshot how to choose models. Its in announcements

[left jasper]

You can choose

[gritty siren]

well they call its 'Perplexiti's R1'

[elfin gate]

R1 1776 is the same R1 reasoning that was before all the time.
They just renamed it and said it's new

[gritty siren]

yeaa thats the thing tho..what does Perplexities R1 mean..coz wheni use it its not the same as before

[elfin gate]

It's same but different.
How you don't understand? 😂

[gritty siren]

the results not the same i'd say

they tweaked it somehow

[distant aurora]

That's sonar reasoning I think

[elfin gate]

If I had to take a wild guess,they changed something with this r1 to appear different so they can say it's completely new model.
But I think there's no reason not to trust the team's explanation on this.
I'm just speculating

🙅

[gritty siren]

yea prob is ..it wasnt what some of us paid for

haiz

[elfin gate]

They properly cleaned the old one from Chinese dust and now when it's all clean and Americanised,they blessed us by introducing it

[distant aurora]

No. It's probably mostly patch work. I don't think remaking it from scratch would even be worth it.

[elfin gate]

They used air compressor to blow out Chinese dust

[distant aurora]

Maybe

[elfin gate]

And put American sticker when they finished

[distant aurora]

Now that chatgpt is down, I guess imma switch to perplexity.

[elfin gate]

Gemini is free

[distant aurora]

People like their spyware American.

[elfin gate]

2.5 pro

[distant aurora]

Lol.

[elfin gate]

I sold my soul to Google many years ago....

[distant aurora]

Is there no 2.5 flash?

[elfin gate]

Flash 2.5pro is free for all users

I don't understand why they did it but thanks

[distant aurora]

It says 2.0 for me instead of 2.5

[elfin gate]

I'm sure I've seen a few headlines about it being free

[elfin gate]

Last one

[distant aurora]

I mean, is there no 2.5 flash?

[elfin gate]

Last one is the one you mean

Click on it and enjoy

[slim jay]

do you guys have problem uploding txt files with code to debate with the AI ? or is it just me, anyone have similar problems, just asking ?

[distant aurora]

Okay

[elfin gate]

Last one click ,don't ask it too much questions about which model is running etc,and enjoy yourself

[elfin gate]

I've been using 2.5pro experimental for a few days ,I approve.

[velvet dune]

Now there is no way to get rid of Pro Search, it pops up in any model.
Who even needed this Pro Search 😅

[copper rune]

pro search is always enabled for every model under the pro subscription

[elfin gate]

I just realised,I am very sorry devs

[velvet dune]

It's a pity that it's impossible to use the model directly without this Google search lag

[copper rune]

you can stop it from searching

[elfin gate]

I'd I select best for my defsult model,then it's deciding which model it is gonna use(all pro all reasoning)
And if I select defsult model any specific model I still have everything like before .

So this is actually good

[copper rune]

if you just want it to write

just disable all three

[velvet dune]

Until today's update, everything worked like that, but now disabling the search doesn't help—it still keeps searching for something.

[copper rune]

strange, i cant replicate that

didnt search for anything

[velvet dune]

Can you see on your screenshot if there is one step of the search?

[copper rune]

nope if i disable sources it doesnt search for anything

[arctic idol]

I am starting to get confused about the auto-select, as an user I paid so I can choose model, there are normal and some super user, why not make a settings option. What is the strategy?

[copper rune]

you can choose your model though

[ruby heart]

this is not a real person

[arctic idol]

I want Sonnet 3.7 but don't have it on all my platforms, then moving to strategy using auto-select to detect based on context what to use is wrong, it should be user based preference not context. Everytime I start on web a folllow-up or new converensation, I need to change it.
This is like dark pattern (a deceptive design pattern)

[copper rune]

cant say im having that issue

it sticks to my choice

[valid charm]

Good morning PPLX town

[bleak hull]

its time for Perplexity Pro to die

all model on PP is worse than 2.5 Pro now

except claude 3.7 sometimes

[valid charm]

Sadly yes., even ChatGPT DR is better than PPLX DR

[elfin gate]

You can choose

[valid charm]

I'm mostly using 2.5 Pro now

[elfin gate]

Me too

[elfin gate]

Only 2 areas where it is lacking compared to perplexity is using 2 languages(at least one I speak) and occasionally real time information

[plucky bison]

Did perplexity deep research get the new reasoning update that they tweeted about?

[valid charm]

Not yet

[plucky bison]

Is there an eta or no?

[left stream]

Open deep research (hugging chat)? Which one is that one?

[left stream]

That’s dope, I thought it was a glitch lol I was making sure not to abuse it so I wouldn’t have it removed 😭

[empty snow]

anyone has pplx issue?

for me now it is extremely slow

[left stream]

Do you use it on android, ios, or website?

[empty snow]

web

[left stream]

Ah ok, I have no issues on ios so far. I’ll check web

[left stream]

Does it happen randomly?

[empty snow]

it seems, probably it depends by the prompt

[silent patrol]

Hello guys. What is the difference between Deep Research and Pro mode?

[empty snow]

deep research is meant to deeply search on the web the information you asked for. i think it's not really fit for complex or reasoning problem. Pro mode is used for all the avaible models and offer more resources

clear?

[short jewel]

Hello

[distant aurora]

https://huggingface.co/blog/open-deep-research

[sterile barn]

GM

[remote portal]

Hello guys 😄

[edgy obsidian]

When can perplexity integrate
Deep seek new model
V 3

[ocean wharf]

Some Arab Metal fans are arrogant, despicable and provocative, laughing at me just because I use AI and still they didn't manage to make one song like my songs. To think that using AI is trivial while there are some songs that I spend 6 days to make them. Strangely enough, I sent my songs to several Western metal groups and they didn't underestimate my songs. As AI professionals we need to spread awareness about AI to fight conspiracy and bullying against human development.

[valid charm]

Anybody here use the PPLX news section at all??

[left stream]

I didn’t even notice they did a new update on gpt 4o

[distant aurora]

It was yesterday I think

[valid charm]

What was the update?

[elfin gate]

Openai or here

[left stream]

It should apply to both

[elfin gate]

There shouldn't be wars around the world

[distant aurora]

Yeah. Maybe another pandemic is the answer. 😂

(I don't support mass destruction, I just hate people crowding.)

[valid charm]

March 27, 2025
We’ve made improvements to GPT-4o—it now feels more intuitive, creative, and collaborative, with enhanced instruction-following, smarter coding capabilities, and a clearer communication style.

Smarter problem-solving in STEM and coding:
GPT-4o has further improved its capability to tackle complex technical and coding problems. It now generates cleaner, simpler frontend code, more accurately thinks through existing code to identify necessary changes, and consistently produces coding outputs that successfully compile and run, streamlining your coding workflows.

Enhanced instruction-following and formatting accuracy:
GPT-4o is now more adept at following detailed instructions, especially for prompts containing multiple or complex requests. It improves on generating outputs according to the format requested and achieves higher accuracy in classification tasks.

“Fuzzy” improvements:
Early testers say that the model seems to better understand the implied intent behind their prompts, especially when it comes to creative and collaborative tasks. It’s also slightly more concise and clear, using fewer markdown hierarchies and emojis for responses that are easier to read, less cluttered, and more focused. We're curious to see if our users also find this to be the case.

This model is now available in ChatGPT and in the API as the newest snapshot of chatgpt-4o-latest. We plan to bring these improvements to a dated model in the API in the coming weeks.

[tame current]

Hello

My voice command is not working

What to do

[elfin gate]

Not in web

[valid charm]

What do you mean? Only on mobile?

[elfin gate]

This text is from openAi or perplexity

[zinc quartz]

Which rights?

[elfin gate]

This was a hallucination , ignore it

[valid charm]

Ahhh from OpenAI

[zinc quartz]

You mean a mistake? How are you hallucinating

[valid charm]

https://help.openai.com/en/articles/6825453-chatgpt-release-notes

[elfin gate]

Not enough tokens in my head

[elfin gate]

It's different company

[left stream]

@distant aurora

[elfin gate]

Dead sea scrolls are word for word exactly the same

[distant aurora]

Emkay ultra

Wait

Are you using a jaybee prompt?

[sturdy skiff]

hello guys how is perpelxity

afte those updates

i didnot get chance to use yet

[elfin gate]

Same

[sturdy skiff]

ok bro

[left stream]

No

[distant aurora]

Then why does it say "Alder"

[quasi cape]

Does anyone know if you sign in with a school account, can the school see your chat history? Nothing in the terms seems to allude to this but im not sure

[short narwhal]

like just logging in with your school email?

[quasi cape]

Yes

to get the one month of perplexity pro

[short narwhal]

No they can't see

[elfin gate]

That should not supposed to be legal

[short narwhal]

Not illegal either per se

[lusty bane]

Hey, I’ve noticed that Cloud 3.7 Sonnet randomly stops generating code mid-way. Sometimes it can crank out 3000 lines no problem, but other times it cuts off after like 300. Any idea why this happens? Is it a performance thing, server settings, or something else? Any tips on how to fix it?

[tame current]

no school do that

i think about comet atleast once a day since announcement

i can't wait

[jagged cargo]

are there any referral program in perplexity

[little zephyr]

Who celebrates Eid

Eid el Fitr

[elfin gate]

They removed it

[jagged cargo]

ok

[jagged cargo]

me

[little zephyr]

Eid Moubarak

[jagged cargo]

Eid Mubarak

[digital veldt]

one of my threads is super long and it wont let me scroll back past a certain point

is all that lost?

[rustic kestrel]

how do i turn off pro search

[grand wharf]

where is r1

[elfin gate]

You can't.
You can set best mode but it won't be non pro.

[rustic kestrel]

😦

[elfin gate]

Why would you turn off pro search

[rustic kestrel]

how do i set it to best mode

[rustic kestrel]

cuz it slower

[elfin gate]

Click on the thing left from mini globus

[rustic kestrel]

isnt this jus for models tho

[elfin gate]

Yes

It will all be pro

[rustic kestrel]

rip

[elfin gate]

But you asked me how to do it

[rustic kestrel]

i mean i thought u could turn it off

[elfin gate]

You're the only guy who preferred having free acc I guess

[rustic kestrel]

its jus thats it so slow

like see i do this basic search and

its still searching

[elfin gate]

Try less complicated prompt

[elfin gate]

Set default model sonar

[rustic kestrel]

alr lemme sdeee

[tribal skiff]

It's not that slow... If you wanna see slow... Try GPT4-o1-pro-mode. Now THAT'S slow. Sometimes it takes 8-10 minutes for an output.

[rustic kestrel]

😭

[left jasper]

How do we trigger pplx to generate images these days? Can't seem to get this generate button to show up no matter what I do.
But it running some python code to generate images with text (just plain gradients), that it can do...

[grand wharf]

perplexity is so ass for coding without the r1

god damn

[merry pendant]

When gemini 2.5?

[grand wharf]

gemini is overrated

[merry pendant]

Nope

[elfin gate]

Nope

[wide stratus]

? Where is there any statement regarding the outages, in there???

[elfin gate]

Itn not necessary

[alpine flicker]

I have a code for perplexity and want to use it, but do not want to give my credit card data. How can I do it?

[severe forge]

Bruh
I got Gemini 2.5 pro
In free tier in the Gemini app

This is supposed to be exclusive to Gemini advanced

[serene sand]

please contact [e-mail]

[merry pendant]

How?

[alpine flicker]

The code is valid until tomorrow (from t-moble moments) and support is way too slow to respond.

[elfin gate]

They said it will be free

[serene sand]

afaik, you can’t use it without adding the card

[alpine flicker]

It was possible in the past

I did so 1 year ago

see no card info

[serene sand]

great, only [e-mail] can help you with that

[severe forge]

What?

[elfin gate]

Yes

[severe forge]

Dawg.
What's the point of paying Gemini advanced

[elfin gate]

No clue

[severe forge]

Just for Workspace+ 2tb storage?

[elfin gate]

I'm not Google

[merry pendant]

I dont see gemini 2.5 without paying for advanced

[elfin gate]

Some guy earlier also didn't see

[merry pendant]

Do you see it

[elfin gate]

I have googleone

[merry pendant]

Wat

[severe forge]

See this

[elfin gate]

I see it

[severe forge]

It's in free tier

I don't know what's happening 😔

[elfin gate]

Why are u sad

[severe forge]

Sometimes Google looks like the good guys

[elfin gate]

First time in my life

Ok maybe I lied

Good but dumb guys

[severe forge]

Unless u know they get a massive chunk of revenue by selling ur data +ads

[elfin gate]

I don't mind

Actually iceo

I do

Wish I was selling my data to china

[tame current]

The point is to get 2.5 Pro stable version later

[hot stag]

hi! after last few updates like the deep search. i'm having hard time to generate images.. the Generate imagem button, doesnt show anymore, any tips on how to make it work?

[torn osprey]

can i use the new qwen model in perplexity?

[light veldt]

fr

[light veldt]

buy a $10 mastercard gift card or smth

privacy.com also

[merry pendant]

imagine losing a year of pplx because of not wanting to use a credit card or privacy.com

[light veldt]

yeah doesn’t perplexity just use stripe anyways

not even perplexity managing your card info

[sleek vortex]

yeah

or even better use through ios or android stores

because if you aren’t giving your cards to apple or google then might as well never spend anything online

[knotty ocean]

was Deepseek removed? when are all the new models coming zzzz

pretty unethnical to rename Deepseek to perplexity reasoning model and name it 1776... yea Deepseek US edition. wtf is this.

[light veldt]

they are just saying exactly what it is

[short narwhal]

Revolut would work too

Usable in other places other than the us

[knotty ocean]

unbias? thats impossible its in human's nature to be bias. Deepseek US bias edition.

[grand wharf]

bro wym its not necessary

perplexity is unusable for me rn

[light veldt]

when it says unbiased its talking about removing the CCP censorship

[distant aurora]

Lmao

Claude shouldn't even be there.

[left stream]

The forest, it created a new member for that specialty

[zinc quartz]

Unfortunately this isn't something that support nor I can help you with. We will need credit card info of some sort. Without it, there's too high likelihood of fraud.

[zinc quartz]

Look under Infra challenges

[storm pollen]

FYI to anyone interested in manus (I know this is perplexity, but I'm sure most of us are enthusiasts). It appears they may have had a data breach, I was sent a very sophisticated phish email using data that I'd submitted to them. They verified to me via twitter that the email did not come from them. I don't want anyone to get got.

[light veldt]

can u send me the phishing site in dms

[barren bough]

Sharing my experience in leveraging Perplexity for low-effort, high-efficiency research.
This may be helpful if you're considering a subscription or unfamiliar with AI-powered search tools.

Perplexity In-Depth Review: Differences, Feature Comparison, and Everyday Applications with ChatGPT

[low spruce]

are y'all actively trying to push users to cheaper models? every day there's some UI change and different ways of adding friction to using claude 😦

just gimme a way to permanently set claude and disable all the other models :p

[wheat pike]

https://www.reddit.com/r/perplexity_ai/comments/1jm2ekd/message_from_aravind_cofounder_and_ceo_of/

this might answer some questions

[wide stratus]

Ah ok, I see, they finally prioritize stability over new features 👌

[low needle]

Hello

[elfin portal]

didnt know they had a model back then, no wonder they were able to turn back the british

[austere granite]

Why does perplexity need claude? is it better at hallucinating?

[elfin portal]

put all the ais in, then you cant ever fail

covers all bases

[inner quail]

Are you the Hawthorn Fan?

[elfin portal]

no I dont feel strongly about them either way

I had to look them up some american football team

I hope they win the big game when its next on

[tardy junco]

Enjoy yall! Really hope you all enjoy this read.

[frank bay]

when will comet release?

[tropic relic]

Just realised pdf export has been removed, any idea why?

[livid aurora]

It seems Perplexity is faster now 🙂

[tame current]

no idea, no news

[brazen idol]

Hello to everyone, I'm new here, is there a place where I can troubleshoot

and ask questions?

[tame current]

here ig

[frank bay]

[frank bay]

AFAIK they run Sonar on Cerebras now so the Sonar model should be faster

[livid aurora]

But I think even sonnet 3.7 with pro is fast

[brazen idol]

I can't understand why do I get this error from perplexity in the middle of the proccess? its not at the begining before i started everything, it just happened out of nowhere in the middle of it.. I really cant comprehened why

[livid aurora]

I dont see any quick and fast mode anymor

[tranquil orchid]

Hey @livid aurora!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[frank bay]

Seems the same to me

[hollow junco]

Does anyone know the context window on the reasoning models? Is it still 32k?

[cold berry]

Anyone facing issues with answer generation?

[serene sand]

https://discord.com/channels/1047197230748151888/1118264005207793674

[hollow junco]

For the DeepSeek Model, it's 128k. Can't find anything on the o3 mini or Claude 3.7 thinking.

[bleak hull]

that doesnt mean unbiased

[tame current]

yeah happened with me yesterday

[copper rune]

is this the context window on pplx

cuz it def feels smaller than that

[brave vigil]

??

[midnight sandal]

I felt that the context increased with Reasoning with Claude, I send legal documents of 20, 30 pages, pplx analyzes the facts and arguments and generates a very detailed sentence, the prompt itself in which I write is long, and I establish a control for it to put an emoji at the end of the generation, and despite using several follow-ups, it maintains the original instruction, not sure, it seems to me that pplx increased the context window, or they did some kind of juggling to get around it

[copper rune]

hmm. that does sound interesting. are u using any search or is search off

[midnight sandal]

Yes, but it is not the focus of my prompt. For example, I want it to be based almost entirely on the documents I sent, HOWEVER, there is a part of the prompt execution in which I ask it to attach case law, judgments, precedents from higher courts and citations from books to support the thesis, at which point it does a quick search (never more than 30 or 40 sources) and returns interesting results, despite sometimes inventing facts. Then I have another follow-up prompt that I like to use for it to perform URL checks, citations, etc. with the aim of searching for false and invented information and it does very well there too.

[copper rune]

interesting thank you i’ll try optimising my prompts

[midnight sandal]

Maybe it won't be 100k tokens but perhaps it will at least double to 60k? Who knows? I'll try to check by token count sometime

[tame current]

yay for PPLX / PPLX Pro. Gemini... meh.

[distant aurora]

@left stream I'm nearly convinced that og website/app will always be better than third party
https://chatgpt.com/share/67ea7a00-079c-8013-949a-d22da6c36573
https://www.getmerlin.in/chat/share/47ebc788-d134-4019-9650-171aa42fc3ef

https://www.perplexity.ai/search/give-me-an-elaborate-descripti-FpkEC90yQYKQEIsuEOJp3Q#0

[dense crest]

How many threads can we have in the library?

is it limited?

[wraith briar]

Hello everyone, has Perplexity updated to GPT-4o as of March 26th, 2025? I find this latest version of GPT-4o quite impressive!

[hybrid oyster]

Does perplexity support the new image creation with 4o, or is that not available? Also, I see that on the web it says that perplexity has image generation, but I have never been able to use it before. How do I use it?

[vapid crag]

hello, i want to ask if the model GPT-4o in perplexity is actually "GPT-4o" or its "chatgpt4o-latest" because there is a difference, the latest version is updated and can code better

[placid coyote]

I dont think pplx supports new 4o images (is even api available? I think it depends on OpenAI if/when they release api, similar to grok 3)
img gen ui was bad, recently got even worse. workaround/current use with Images tab: #1351614405833527487 message is available only on web, not in mobile apps

[vapid crag]

can you generate images in perplexity at all?

images on a custom prompt