💬│general

[naive sage]

I think he's right about the 20/day limit in actuality, though, despite how the article sounds. There's screenshots on reddit showing 'nearing the 20 limit' or something like that, from within the tool

[dry kestrel]

Tysm

[naive sage]

I think they shy away from posting the actual limits in their documentation b/c they're subject to change. But in that case one way to go about it is just to use a variable in a content management system that you can update whenever the limit changes, then all articles get updated with the new value (and a new 'effective date' or 'as of' date in the articles)

[tropic crystal]

App down in IOS? 👀

[tame current]

' claude website isn't straight from model to your screen
did not know that... thanks!

[ornate night]

is there a way to set default focuses for perplexity? Like set social and web searches as default?

[tame current]

Man Cluade is absolutely terrible at debugging and analyzing code huh.

making up documents. Linking to files that don't exist and gaslighting user. Weird behavior for an A.I.

[crystal cosmos]

yes perplexity is just using movies nothing is real

[forest aspen]

hey, i want to build a ai mentor where i want to implement deep research facility, resume analysis and a career guide bot, also want to integrate with mcp. for which i want to fine tune any llm. this for a hackathon. thing is i have no guide to help me out with this. one of the help i need is to choose where to fine tune as i don't have hardware what model will be free and best or dirt cheap and best for this
some general recommendations around this will be of great help, thank you
also i am very new to it, though i have high level knowledge of llms and have heard terms like langchain but don't exact;y what need to be done.

[lament spear]

when comet

[light snow]

how can I use perplexity as my copywriting assistant?

[dire vault]

Hello, can I use Chat gpt 4.5 unlimited now. With perplexity pro?

[cobalt monolith]

gpt 4.5 is only 5 queries per day with perplexity pro

[sturdy skiff]

4.5 is expensive, even chatgpt plus is giving only 50 per week.

And what is special about 4.5 i dont see any value for my usage and use cases, so i just ignored gpt4.5

[queen girder]

[rose musk]

Hey can any body help me out ?
I have been constatnly using one thread of chat for a very long time prompting back and forth (it is for a project / assignment) that involves a lot of ideation , debugging and test-case generation
I didn't want browser to eat up in my RAM , specifically because I am running WSL which itself takes out 1-1.5 Gb along with VSCode

So I was happy when I found out that perplexity finally launched a desktop version , but now after a long time I am beginning to feel the slow-down's quite a lot (After the chat has become lengthy)

Why is this happening and what are some quick and long-term fixes to avoid this ?

[tranquil orchid]

Hey @rose musk!

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.

[gritty siren]

I went back to the web version HAHAHA

[rose musk]

Is it better ? I mean I don't know what the offline version is doing that it is taking up so much memory ?
Is it storing the prompts and generated data / cached data locally and loading it all at once ? Because it feels quite awful after a while

[gritty siren]

No idea but it's slowing down my laptop

In browser i tink most things are more server sided

[strong pond]

mods?

[rose musk]

Also I got this (weird hallucination twice or thrice today while working on my coding assignment) , Is this because I have set the mode to Auto for this prompt ?

[dim quarry]

Is there a way to have a nice LaTeX front-end just as in ChatGPT ?

kind annoying that he doesn't parse the $$ output.

[white loom]

Hi

[placid coyote]

you can instruct the llm to use format pplx supports, eg https://www.perplexity.ai/search/replace-with-and-R.38s8AKRUqcvnCA5UQJbA

this prompt seems to work too: https://www.perplexity.ai/search/double-all-dollars-dont-use-co-P6JsbNOTQiabBoawKRZHWw?0=d&1=d

btw it feels related to this issue https://discord.com/channels/1047197230748151888/1341734228072730645

[zinc quartz]

Any queries relating to coding we don’t share images; were there any queries that obviously needed images and you couldn’t find it?

[zinc quartz]

This is a bug; mind sending the query URL?

[distant aurora]

Am I the only one here who thinks that gpt 4o is dumber than 3.5 and 4 at grabbing context?

[sterile barn]

Maybe not sure

Explain to me why you came to this conclusion

[distant aurora]

Try this prompt: How high does the xp of top 5000 in codm reach by the end of a season

[sterile barn]

What does that prompt do

[distant aurora]

At this point you are trolling.

It's a simple question.

Which has a direct answer on a reddit post.

[sterile barn]

Yes I understand

[indigo valve]

No ability for extensions and just a web wrapper

[raven night]

Is comet coming out today??

[indigo valve]

Said who

[raven night]

Theres a leak about something on their instagram check it out

[indigo valve]

It's just a pinned post

[raven night]

I meant look at the ad

[raven night]

https://www.instagram.com/reel/DHToBOix-iB/?igsh=MXFpcHBzcDZodG92cw==

perplexity.ai

When you need to get it right, ask Perplexity.

Likes

2114

Instagram

[sterile barn]

Whats in the ad

[gray prawn]

Ads

[sterile barn]

Awsome

[hollow jacinth]

I have been on intercom with perplexity and haven't received an answer in over three days. Previously, I startted a support email, which is also not getting a response in over a week. What is the best way I can reach to someone for an overbilling?

[indigo valve]

What specifically teased it?

[sick arch]

what do u use the most?

[distant aurora]

Most?

R1

O3 mini high is usable.

[lone hound]

Claude, it's more powerful.

[lament spear]

yea bro and the ad tells us nothing, the announcement day was yesterday and they js made an extended version of the ad

i want comet now

:(((

[long pasture]

@signal hamlet Any chance VODs will be saved for the workshop?

[hollow moss]

Gt

[signal hamlet]

I'll check with the marketing team.

[tame current]

Honestly, it feels like it is dumber than Claude 3.5 ... Now it is often lost in context and makes more bad abstracts for me

[distant aurora]

@left stream https://www.instagramez.com/reel/DF-WSwSxF0G

[sterile barn]

Are there private/incognito features for Perplexity?

[vague pagoda]

Yes! You can use incognito mode by clicking on your profile name in the bottom left corner of the interface and selecting "Incognito" from the dropdown

There's also an option in the apps when you tap your profile

[glacial schooner]

Oh that’s cool

[halcyon coral]

Hey @sterile barn!
You can also deactivate AI data retention in the settings.

[sterile barn]

Yes if you click the toggle button

Right?

[halcyon coral]

Yes, the switch in my screenshot shows AI Data Retention disabled.

[sterile barn]

Is this a joke 😂

Regardless of the data retention being on or off the Chat you are in will store chat specified data to assist with the current tasks/ prompts from the first chat to the last. Creating a new one will allow for a clean slate and a fresh quicker response to your direct questions and prompts.

[leaden lagoon]

They are doing anything but releasing comet

[gray prawn]

I swear “Vibe Code” gotta be the modern tech equivalent of “Hippie” in the ‘70s-80s

Just looking at the website that (no offense) slop is not worth $50/mo lmao

Looks like someone just got a bunch of scratch games, “ai-ified “ it, and slapped a huge price tag and called it a day

https://tenor.com/view/south-park-its-gone-gif-4104229

[ruby heart]

🤷‍♂️

for free worth playing around with

[narrow drift]

Hey all!

[vague pagoda]

Hiya!

[lone hound]

For my part I still prefer Claude in reasoning mode, than the r1 and o3-mini models.

[obsidian hare]

i'm a little fustrated. perplexity doesn't seem to be remembering my conversations properlly and it doesn't appear to have any system prompt option or proper memories. Beisde the little blurb you can put about yourself.

it keeps truncating code outputs and isn't really performing as I want. In contrast to openai. Can someone tell me if there is some things I can do to add system prompt or memories or have it actually remember not to truncat code outputs. Its not even large outputs. Just a small configuration file

I am a new but pro user but maybe I am not aware of some options for this. Like on the desktop version or something. Just using web version currently.

[thick dust]

Web is the best version, always have been; you don’t miss any option by using the web version

[obsidian hare]

ok any plans to allow system prompt or memories?

[thick dust]

You have to tag the admins and ask them about that

[willow heart]

gm everyone~

[thorny compass]

The new limit pro search for free tier it's 3 per day now or it's bug ?

[left stream]

I just ate a scorpion lol

[left stream]

They have to make o1 pro join in the fun

[fair solstice]

how to get pro role?

[vague pagoda]

Check https://www.perplexity.ai/settings/account

[fair solstice]

thank u, i will try

[vapid onyx]

If you have issues let one of the moderators know

[sterile barn]

Whats one cool feature you guys really like about perplexity

[green crystal]

anyone notice that deep research is now using less sources?

[sterile barn]

Yep

Thoughts on this?

[unique umbra]

Jfl files released

[sterile barn]

Yep thats right

[unique umbra]

Anyone process the documents with perplexity 💀

[fleet magnet]

hello

[sterile barn]

Nope, have you?

[unique umbra]

Nope

[distant aurora]

Yeah. Apparently we overestimated O3 mini and underestimated o1 and o1 pro.

O1 hype didn't last cuz of paywall.

[frail ridge]

anyone notice this?
API for Sonar Deep Research is not as good??!?!

I was using the API for sonar deep research a couple weeks ago and it worked just as the deep research worked on the website with FULL reports that were very very long and in-depth. Now, however, the reports it gives are only a couple words and aren't like the full reports that were 1k+ words that it gave before and that it gives on the website.

And.. it's still as expensive or even more expensive for requests...

[sharp coral]

https://discord.com/oauth2/authorize?client_id=1085343396173991946&permissions=274877908992&scope=applications.commands+bot

[kind shadow]

hi everyone is anyone in nigeria?
i need to ask something

[sudden raptor]

IMO threads and spaces have a dreadful user experience

[naive sage]

(It doesn't work)

[frank bay]

R1 types really nicely though

it listens to the personalization way more

[crisp roost]

i’d love to see o1 back

[desert hinge]

I think they have abandoned the Perplexity project, they simply continue to add models but there have been the same bugs for months and they don't seem to have any intention of fixing them.
And noone from supports responds

[sturdy skiff]

I dont know what is perplexity project, but one thing is true none from support responds. Thats really bad. I have three support tickets without response from them. Thats truely bad.

[desert hinge]

Me too, is just a joke

[plush helm]

R1 is useless most of the time... all the "reasoning" about stuff that leads nowhere and isn't even remotely part of the issue.... Claude is weird "The query" asks ... I have never been addressed as "The Query" before. o3-mini yields better results overall at least for me when it comes to debugging js code, or maybe there is a better option for code debugging?

[sturdy skiff]

Yes even i am having good help only from o3 mini in perlexity orelse the normal deep research option. These two are beneficial for me

[lone hound]

I'm not a programmer, so I won't be able to help you in that area. Claude 3.7 a improved in the developer session, you could try and see if it works for you.

That's why I recommend Claude 3.7 of Reasoning in Perplexity.

[tame current]

tried but not worked for me

anyone have compared sonar model with any other model ?

[sterile barn]

Nope not yet

[green crystal]

I hope perplexity improves deep research. Even grok deep research is pulling ahead...

[sterile barn]

Yep Im sure it will

I personally really like using Grok, what about you guys?

[fair solstice]

Guys, for your choice, what is the best solution now to choose in perplexity, i am thinking between using 3.7 sonnet with pro mode or using o-3 mini with reasoning, also when 3.7 sonnet with reasoning comes out on iPhone, should i choose it?

[naive sage]

It depends.

[lone hound]

I am going to leave a message of reflection: we brought nothing to this world and we will not be able to take anything with us when we die, so if someone covets money, he goes down a bad path and hurts himself.

[plush helm]

I tried Claude last week and it did not provide any quality code. Would repeatedly create non working code and insisting it works, then again they all do that 😅

[lone hound]

And did you try using on Claude's official website?.

[plush helm]

No only within perplexity (pro)

[lone hound]

Did you try using space mode and give him personalized instructions, and then in the son the message you want to do?.

[plush helm]

Yeah, I only use space. Even with instructions all models sometimes forget it’s JS and start producing Python. Real dumb

[lone hound]

Say that it should only be JS and no other code, without leaving the base.

[vocal crypt]

Did the UX for removing sources from a chat change? Like is there a hidden button to remove them or can you just not do it anymore?

[subtle topaz]

Hi

[vocal crypt]

Correct, perplexity app natively can’t answer my question that’s why I’m asking here.

[subtle topaz]

What question did you ask him?

[vocal crypt]

"Previously, in a Perplexity chat I could access and remove sources from a conversation. That appears to have been disabled. Was there a recent update that did this, or is there hidden functionality that would enable me to remove a source mid-chat like the previous experience allowed?"

He's responding about Jan 2025 when this functionality was working a few days ago for me 💀

Happy to be proven wrong about this 🙂

[vague jackal]

Hey Everyone. I am working on Oblix Project -- seamless orchestration between local models vs Cloud models. Agents monitor system resources and takes this decision. Here is the demo video https://youtu.be/j0dOVWWzBrE
You can see oblix orchestration between ollama and openai.
would love to get feedback on this. Check https://oblix.ai/ My next integration with Perplexity

[clever galleon]

Hi everyone
I am looking for a client who can provide game source of RPG or Strike games that are currently running online.
That must be popular.
I will pay good for that source project.
If interested please DM me.

[silk field]

me too!

[indigo valve]

What am I looking at?

[gaunt onyx]

Why perplexity doesn't work on locked screen?

[sterile barn]

Does it not work on lock screen?

[gaunt onyx]

Not at all bro

[sterile barn]

Ok I see

[gaunt onyx]

It should have been awake in background like chatgpt does

[sterile barn]

Yea I wouldnt know

[gaunt onyx]

Could you please clarify which earbuds integrated with perplexity please

[sterile barn]

Earbuds?

[green valve]

Hello, why can't I choose a yearly tariff on the website?

[devout sonnet]

Is anyone else noticing Perplexity use significantly less sources? Like 8-16, maybe 25 max.

It used to be at least 40 ish before

[lyric stream]

is there a way to look at someone's shared thread and know what model they used for each question on that thread

[naive sage]

It might depend on the question, but yeah - I see about 5-10 for Pro searches now.

If you use Deep Research I would expect more sources, maybe that'd be worth comparing

[frank bay]

you do it on the stripe billing page

"annual (20% off)" or something like that

its a small checkbox

[crisp roost]

try using markdown headings or xml tags

it would be like

# Instructions
You’re a helpful assistant that only uses JS.

or

<instructions>
You’re a helpful assistant that only uses JS.
</instructions>

it might be better to use a specialized tool (cursor, github copilot) for coding tasks

[short narwhal]

o1 pro in perplexity when 😭

[arctic idol]

sonnet 3.7 technical, coding and writing
other then that o4 for others. sometimes sonnet give short answers r1 is also nice for philosophy/psychology

[sturdy skiff]

Perplexity is 20USD per Month and you are expecting o1 pro in Perplexity? 😂 Common be some kind.

[devout sonnet]

The hallucination rate is unbearable if you're using "deep research"

[balmy widget]

Hi G's, quick question, why should I use perplexity instead of open AI for example?

[devout sonnet]

Because you can get a year for like 20 bucks off some sites...

[fair solstice]

I'm really looking forward to when the update on perplexity apps with claude 3.7 with reasoning comes out

[fickle sorrel]

1. when i switch to my PP tab and a thread is open, its not on my most recent prompt, but at the first one. wtf?
2. visiting PP only once a day: u go to the tab, then u wanna write, and suddenly it hits u like a bomb, it reloads some javascript frames or so. this is so disgusting.

[fair solstice]

I will only use claude, so far I've been using r1, it seems to be leading the way with the other models available right now

[jovial kayak]

Did anyone else just get signed out of their account?

[arctic idol]

I had good experience with GROK 3 and Gemini in DeepThinking. Try to check them out

[fair solstice]

to be honest, I don't use grok because I don't like that name 💀 🤠

[patent cargo]

Why not? 🤣 Give us 1 query a day - 30 a month. If.they can give 10 /day for 4.5 with $150/M then 30-50 is reasonable . Good.marketing as well

[naive sage]

I'm not sure whether one query per day would really be useful or not. As soon as you have to refine something... eh.... 🤔

[short narwhal]

1 query a month take it or leave it

[dry kestrel]

never

gpt4.5 being much cheaper is already so heavily limitedf

[ruby cove]

Why don’t I have pro channel

[rare gyro]

same

[tribal skiff]

Hey folks,

I am a Python coder and a pentester. I created a Python program that uses the perplexity API for a variety of functions. It's an OSINT tool, aimed at gathering information about a subject. Some features don't require API's, but the advanced features use RapidAPI programs and Perplexity API calls.

It's called ClatScope Info Tool.

Check it out and let me know your thoughts. A new version will be released soon with 79 functions. My latest release has 67.

I was featured in Linux Magazine and Dark Web Informer (the latter essentially being a scrape of my Github) and have received tons of positive feedback.

The piece of feedback that made me very happy was from a jury consultant. He said there were 2 jurors who weren't in the states system, but my tool was able to identify them which led the jury consultant to discover that the juror posted racist remarks on his Twitter. So his client (the lawyer) didn't have to use one of his 8 peremptory challenges, he simply showed the judge the laptop that had the racist remarks and he was struck.

If anyone here wants to try it out here's my Github. You'll need to open the script in Visual Studio Code (or notepad) and enter your api keys.

Here's my Github.

Github.com/clats97/ClatScope

[brittle jacinth]

[green crystal]

indeed and its skipping information due to it

gotta give credit to grok 3
its looking to be better than perplexity for now

[sleek coral]

Hello ! Is there any affiliate program with Perplexity ? Have quite a lot of Recommendation already so I was wondering if it could go further, anyone knows ?

[harsh plover]

I could not find GPT 4.5 in the dropdown menu. Does anyone encounter this issue? Is this a bug?

[tame current]

yeah its removed for sometime ig

[sturdy skiff]

yes its dissappeared, may be they underwent loss and removed it

[cobalt vale]

thanks 👍

[lone hound]

Is Perplexity going to integrate the o1-pro model?

It is very expensive because what I see.

[paper violet]

why GPT 4.5 disappear from Pro account?

[crisp roost]

they always remove expensive models

[sturdy skiff]

Someone asked o1 pro, so they removed 4.5 also 😂

[ember notch]

Guys I need help like ASAP. I was trying to buy some credits on Perplexity API. I gave all my credit card info and everything. It somehow just entered $1000 to buy and not it just says pending next to it. I thought this was going to be a $3 purchase. I NEED TO CANCEL this like right now. I have tried looking everywhere but im not able to cancel this. Someone please help!!

Can someone please let me know where to raise a ticket? I need help

[split dome]

report to this channel bug-reports

[ember notch]

Ok thanks

[oak tulip]

Ho

How

[wise edge]

o1 > 4.5 , hahaha

That's OK, after all, o1 is more expensive than 4.5~

[verbal verge]

Anyone else notice that when you use deep research 2-3 times or 3.7 Sonnet Thinking, the model just automatically cancels and switches to auto mode?

[devout sonnet]

I'm just gonna switch to you dotcom. Too many bugs. Models don't work and sources are like twice as fewer compared to a few days ago. Unusable. Even grok is better.

[verbal verge]

I support you

[queen girder]

[lyric stream]

hoqwww

[lone hound]

It seems that gpt-4.5 still works on Android

[mortal talon]

It’s been a while I haven’t generated images on perplexity. Since then, the UI might have changed, but can we still do this on perplexity?

Now that I check on setting menu, I figure out there isn’t place where we could select image generation models

[lone hound]

It works for me and I can select the model

[mortal talon]

Thank you for checking! I’m on mobile now and I’m going to check it out on my own later

[lone hound]

Fine. Use the browser and put in desktop mode, no need to have a computer nearby.

[mortal talon]

Oh wow it works! Thank you 🙏

[digital tartan]

I have a problem that I can not see "Generate image" on the right side, and I am a Pro user. Is anyone else having the same problem?

[pine estuary]

hi

are there competitive programmers?

[fickle plume]

Guys i need some Help, Just a Small Question.
Is it normal that i dont have deepseek in Perplexity Pro?

[wraith crow]

Hi, please have a look here https://discord.com/channels/1047197230748151888/1225557012532695060

The "generate image" button will only show at the latest question in a thread.

[wind bramble]

my library threads have disappeared Has it happened to anyone else?

[wraith crow]

You can find the R1 model in the dropdown when choosing "Reasoning"

[sick arch]

claude 3.7 sonnet is better than deepseek

[fickle plume]

Its not there

[sick arch]

share ur ss here

[wraith crow]

Please refresh the page and make sure you're using the right account. Also make sure you don't have an extension installed that might interfere with the website.

[verbal verge]

@wraith crowI need help! When using Claude 3.7 (both reasoning and deep research modes) and Pro Claude 3.7, it keeps automatically switching back to auto mode after every single query. So annoying!

[ivory bramble]

Yeah sameeee

[torpid garden]

lets not even dream about it

can we use mcp servers in perplexity?

[wind bramble]

Thanks for the reply. This is going for 24 hours. Yesterday, I had the right threads in the iOS app, but not on the web. I logged out of everything, and now I don't have the old threads anywhere. I don't use any extensions, and I browse the web on Safari - never had issues before

[shut dagger]

same issue

I see no generate image option

[naive sage]

What search type are you performing? I just tested and these still show up for Auto and Pro searches (using web interface)

[raw prism]

Perplexity can't access airbnb. I've tried all sorts. Any ideas?

[pallid fossil]

Whats wrong with the copy icon the copy icon gets scrolled up previous one was better to copy when generating multiple codes

[sturdy skiff]

send your query here email as well, [e-mail]

[pastel ruin]

Why does perplexity keep defaulting to auto even on PRO?

[pastel ruin]

use a space

and add the link

restrict it to a space

[raw prism]

Yeah, tried that 😦

[pastel ruin]

😦

[raw prism]

tried

[pastel ruin]

😦

[raw prism]

😭

[pastel ruin]

Is it a bug? @sturdy skiff do you know something

[sturdy skiff]

i dont feel its a bug, i feel they did a change to behave in that way. i am also manually always changing it for every message

may be they are getting a loss for 20USD,, if people are using those deep research, o3 reasoning, and claude reasining models so much.

[raw prism]

Is there a free trial of pro avail?

[junior quarry]

Hi, is the T-mobile pro subscription eligible for pro channels?

[cursive jacinth]

is the team in charge in updating the ios app dead or what sir?

[vague mountain]

Save the site as a pdf and upload the pdf

[cursive jacinth]

or have they forgot that ios has no claude reasoning sir?

[pastel ruin]

There are services like t3chat and global gpt which cost 8USD and have different models like Claude Sonnet 3.7, o1, o3-mini

[sturdy skiff]

Can you please send me those urls over personal chat. Really in search of o1 model for less price.

[granite drift]

Hello, I really enjoy using this app and I'm a heavy user. However, over the past couple of days, the app has been experiencing interface lag. I don't want to clear my search history because it's important to me. Do you guys have any plans to optimize this issue? Thank you.

[pastel ruin]

sure bro

[granite drift]

Has anyone else ever encountered this problem? 😫

[cobalt vale]

Pretty sure t3 added a limit of 100 Claude messages per MONTH a while back, and only announced it on social media.

[cobalt vale]

Shows up on Windows app only

[cobalt vale]

Shows up on Windows app only

[sturdy skiff]

Does perpplexity have option to query a question another space for another space , reason is sometimes we need some other expert AI to speak on current topic happenign in this space

[junior quarry]

@slow lantern @grizzled ledge did you get access to pro support channels?

[old blade]

did Perpelxity just got the worst update? on every response the model goes again to "auto" which really sucks for UX

[junior adder]

It keeps alternating between auto and pro, even if you're using auto/pro specifically

[old blade]

for me automaticly switches to Auto. After Pro message the next is defautl AUTO. This is garbage...

[fervent berry]

yea, you need to join via pro invite from settings

first leave server, then use settings pro invite

[junior quarry]

I did that twice to no avail

[sturdy skiff]

now there are two options for Deep research, i really loved the result of High mode ❤️

[fervent berry]

• image and graph output

[green crystal]

Where?

[sturdy skiff]

can you please share english versio, to help me understand

[fervent berry]

sec

[sturdy skiff]

in perplexity web app

[green crystal]

Hmm I'll check

Because the search depth has been getting worse and worse

I only see the deep research button and that's it

[sturdy skiff]

the output structure and detailed content of high mode impressed me in first attempt

[fervent berry]

it doesn't work via auto for example translating, but here's what suppose it say

[sturdy skiff]

ok this is what result?

[green crystal]

Can you show me where of the UI this option is?

[fervent berry]

https://www.perplexity.ai/search/ktory-bukmacher-z-wylaczeniem-.JxTr6qJRYS3Fy8hT_Jxjg

[green crystal]

Ah guess it's not rolled out entirely

Because I only see the deep research button

[sturdy skiff]

yeah i felt there is so huge improvement

[green crystal]

Well I guess I'm using the standard version because the search is so shallow at the moment

[fervent berry]

they should add new voice mode for android

[old blade]

same for me the new Deepresearch is very nice

[old blade]

U have still old Voice on android? oh gosh im jealous. The new one is completly garbagbe. Its just Faster and you can interupt but the Anwsers are 10x worse and also the Voices are very bad a big downgrade.

[junior quarry]

Did it work instantly for you? I've tried like 5 times already

[fervent berry]

yea

[lone lagoon]

If you have a problem joining the Pro Lounge, and you have Perplexity Pro, try to reach out to one of the staff here on Discord.

Saying that you have a pro account, and you havent got access to pro lounge will be enough

[cunning ridge]

Just when I purchased Perplexity Pro, it just automatically improved the Deep Research through an update a few days later lmao

Is that a coincidence or just me hallucinating?

Perplexity is even programming a code to gather information. 💀

And yeah, I am trying to make a full ahh book with AI

And omg this is wild lmfao

[runic raptor]

Why is perplexity forcing auto on me?

[sturdy skiff]

one of perplexity customer told, 4.5 is present under rewrite option 😂. i am not kidding, whoever is sad about missing 4.5, still it is present under rewrite

[whole flame]

Some folks are giving perplexity at $10 for a year. How are these folks being able to do so ?

[earnest tundra]

The new DR is pretty great... are there any usage limitations for pro users?

[serene sand]

ＰＬΞＸ
ＵＰ
ΥＯＵＲ ＰＨＯΝΞ

[bold imp]

Feedback / Question regarding Deep Research modes

Could we get an official announcement or clarification about the new "Standard/High" selector in Deep Research?

Specifically:

• How does the new "High" mode differ from the previous mode?
• Are there any usage limits on the "High" setting?
• Is the "Standard" mode still limited to 500 queries per day?
• Could you briefly explain the underlying models or tools combined in each mode?

This information would not only help us provide more accurate and focused feedback, but also ensure we're using the right tool for the right task.

Thanks!

[sturdy skiff]

as per their website, pro members gets unlimited deep research, that is reason entire world has purchased Pro memberships for perplexity. So your question is not valid about message cap limit of deep research. Its unlimited for pro users, thats why we purchased

[old blade]

thats not ture. Its limited to 500 daily. (nobody can use 500 per day so yea basicily its unlimited)

[torn sierra]

If you want this to be changed, react pls 🫣🤗 https://discord.com/channels/1047197230748151888/1345653079185559563

[sturdy skiff]

Check in their website its cleary written as unlimited and even in email they confirmed me its unlimited

[sturdy skiff]

read here
https://www.perplexity.ai/hub/blog/introducing-perplexity-deep-research

[torn sierra]

They use this term many times - “unlimited”. Actually it is limited but the limit is too high to be used up by a common user 😉

[old blade]

https://x.com/AravSrinivas/status/1884801300027589007 look here @sturdy skiff

Aravind Srinivas (@AravSrinivas) on X

All Perplexity Pro users now get 500 daily DeepSeek R1 queries (without censorship and prompts not going to China).

Free users get 5 daily queries.

X · Jan 30, 2025, 3:10 AM

[sturdy skiff]

We are talking about deep research but not deepseek

[sturdy skiff]

Im ok if it is 500. I am satisfied with 500 per day

[old blade]

sorry look here @sturdy skiff https://x.com/AravSrinivas/status/1890464738951233536

Aravind Srinivas (@AravSrinivas) on X

Excited to introduce the Perplexity Deep Research Agent: available for free to all users. Paid users only need to pay $20/mo to access an expert level researcher on any topic for 500 daily queries, and need to wait less than three minutes for getting a full research report.

X · Feb 14, 2025, 6:14 PM

[sturdy skiff]

Thanks for sharing.

Yeah still good with 500 ❤️

To be frank perplexity is the best AI platform, and its underrated many people do not know about its features

[old blade]

yes and no. Many small issues. Like today. You ask something in pro mode or else. After respond it now switches back to "auto" which is horible.. Also image generation could be better be implemented. And the Voice mode got a horrible downgrade. only positive its faster and you can "iterupt" but everythiong else is worse. Worse audio quality and worse responds

[sturdy skiff]

Yeah that auto issue is irritating me also.

[hybrid zodiac]

hi i have a question i use perplexity on web ios and android on IOS (ipad) i have the new talking thing (it works better and in czech) Android rn dont have the new speeking thing i like the new one becouse it works in my language and i wanna ask if it is available on android too

[distant zephyr]

How do you guys like the new "native" Perplexity app on Windows versus the PWA app that you could install from Microsoft Edge?

[sturdy skiff]

I don't use windows app, because i need multiple perplexity tabs open, browser is convinient forever for me

[fathom moss]

anyone on else on desktop browser(windows) have to keep choosing the same model because it switches back to auto when working in one thread?

[sick arch]

yes, it keep switching to auto

[fathom moss]

okay, it's driving me crazy because i keep forgetting to switch it back. At least I'm not alone.

[sick arch]

yes, i have to manually change it to claude every time page refreshes

[fathom moss]

Same I love Claude It's super helpful for what I'm doing.

[sick arch]

yes, much better than r1

[fathom moss]

Deep Research Set to High is pretty good if you are looking for another option but it's still a little cut and dry but works great if you're not getting the response you need for Claude 3.7

[sick arch]

set to high? what is this?

[fathom moss]

this

It should say standard but you can switch it to high

takes longer but is sooo great

[sick arch]

ok got it, i was using complexity extension, thats why it was not showing

[fathom moss]

I live in the desktop many tabs open situation

[sick arch]

do u use complexity for perplexity, it will enhance ur experience

[fathom moss]

I haven't is it secure?

[sick arch]

yes, its open source

[fathom moss]

I'll check it out!

[fathom moss]

It's awesome! Tysm.

[sick arch]

currently some of its feature not working due to pplx ui change

[distant zephyr]

That's an interesting usage. I, on the other hand, have dozens of browser tabs open at any given time and like the convenience of quickly finding Perplexity when I need it. So I like to have the standalone app, be it the "native" one or the PWA. If Perplexity added built-in tabs, would you switch to the dedicated app?

[sick arch]

u should join their discord channel and talk with their members, they r super chill

[indigo valve]

I've been in love with this app

[zinc solstice]

what app is that?

[indigo valve]

Weather by Google

[zinc solstice]

nice!

[indigo valve]

Reads the weather and summarizes it in an easily digestable way@zinc solstice

[zinc solstice]

i see, isee

i mainly use apple weather

or i just look outside tbh

but, its useful when you're planning holidays imo

[indigo valve]

It's cool before bed because it will tell you how it is now and how it's gonna be the next day without having to look at the 7 day

[zinc solstice]

ah that's a handy feature to have

so ive been trying to get pplx to be my default search engine and i was hoping you could help me with that

i want it to use r1 by default

but it doesn't for some reason. i use edge (please don't judge) and i've inserted this into the "engine" settings:

https://www.perplexity.ai/search?mode=r1&q=%s

however it still does not seem to use r1 unfortunately

[indigo valve]

try this? https://www.perplexity.ai/search?q=%s&model=r1

@zinc solstice

[warm sapphire]

@indigo valve no beuno

[indigo valve]

Well shoot lol

[craggy geode]

found it interesting to learn the concept that steel made making steel more efficient

so without steel it would be more difficult to make steel

a bit of a feedback loop

making the thing made it easier to make the thing

more specifically, using the thing itself to make the thing makes it more efficient to make the thing

kind of what's happening with AI right now

[cobalt vale]

Unlimited Claude might be more important than unlimited deep research for some

[dry kestrel]

nah theres a deep research cap

nothing is unlimited

its 500 a day

[lone hound]

For personal use, the limit set for both models is fine. I don't think anyone is constantly using Perplexity all day long constant tasks..., unless the account is shared with multiple people...

[steady bluff]

hi guys, best tutorial to learn perplexity?

[dry kestrel]

nah ive hit the limit a couple times after intense usage

gone through the limit in like 3-4 hours

its pretty good though considering that they are using api

[alpine siren]

someone report it man

[fervent star]

From the research that I've done, I think of these token limits more like a bathtub filling with water. If you feed it a bunch of stuff, it's gonna run out of space, but if you give it time to drain out the "old" water, then your good. The problem is that we are dealing with data so when that data is gone, it can effect the overall result. So the only way to really deal with that is to change the size of the bathtub or change how the water is flushed. So if you can find a way to drain the tub at the same rate that stuff goes in while retaining context, then that's how you would probably adjust that.

Then you have your token limits that are hard set for other reasons. What I do with the deep research is I do "normal" research with another engine. Then once I have that boiled down. I put it all into a text file, add it to perplexity and then if it's a legal document, for example, I'll be like, pretend your a lawyer and tell me what you think of this document and how can we improve and expand it. I get a lot more out of it that way.

Large Language Model engines have a vast array of information. It's sorta like that saying.. "if you hit an infinite number of signs with an infinite number of BB's, you'll eventually reproduce the entire worlds literary works in braille" It's vast array of data nd LLM's like perplexity use your input to better find where to look in that mountain of information. So when you say things like "pretend your a doctor" or "your a doctor" that will narrow it's knowledge base and your less likely to get irrelevant information.

[fathom moss]

I think the fixed the switch glitch.

[sterile barn]

Do most people here have subscriptions for perplexity

[dry kestrel]

people with the "Pro" role have subscription

[sterile barn]

Yea

[dry kestrel]

its a good deal

[verbal verge]

I can make it very clear that the IQ and function of deep research have been emasculated from today, Now he's split into two functions, and his IQ isn't what deep research used to be

[nova mauve]

Hi

[sterile barn]

How so, elaborate

[verbal verge]

My prompts aren't working anymore 😩 Before, Perplexity's deep research could execute my prompts and complete tasks properly. But now, both models in deep research won't follow my prompts and just generate wrong logic and answers. So frustrating tbh!

I believe perplexity's top people made castrating IQ choices to reduce overhead

[sterile barn]

Ok I see

[low spruce]

is 4.5 gone now D:

[indigo valve]

I don't know the usage limits between Standard and High in Deep research but it looks like High's thought process looks more "Thought out" for a lack of better terms

[solar bison]

are there different limits at all?

[sturdy skiff]

Under the rewrite option it is there.

[indigo valve]

There was no limit on DR and it would make sense for it to cost more

[dry kestrel]

500 daily limit for pro

[solar bison]

500 for both standard and high research, correct?

[dry kestrel]

I think its merged together

[desert night]

how to make perplexity to following all knowledge and information in the thread?

[indigo valve]

There is a limit to how much context is retains

[dry kestrel]

There is 32k context length

[tame current]

This is the Pro discord, yes?

[tame current]

How do I get the Pro role? I have Perplexity pro

[hollow widget]

yo guys, so i recently switched from chatgpt to grok, but grok's features are really buggy. so my friend recommended me perplexity. can i ask - my MAIN motive is math + coding games, can i ask if perplexity is made for coding?

Like, if i understand perplexity right its more of its own thing + uses other models to help it? but in the list i dont see the big models like sonnet 3.7 thinking, or open ai o3 mini high.

so i just wanna ask for yall's opinions on using perplexity to code

[sturdy skiff]

You can see other models only in pro subscription. But lets see if someone answers your coding queries.

[waxen lion]

never

do not use it to code

it sucks

[hollow widget]

ahh i see. what ai would u recommend for coding then?

[exotic flame]

Claude 3.5/3.7 sonnet

[hollow widget]

i heard claude is very stingy for its usages though

[tranquil orchid]

@gkishorek banned

Reason: Fake cracked software

[hidden knot]

does anyone know why my perplexity is always searching random sources (most commonly "Apple CEO 2023" and "famous birthdays"? I had sent a prompt very unrelated to what it searched (for a fish project). It usually still gives a good answer but why is it listing these as its sources

[zinc solstice]

Any help with this?

#💬│general message

[somber lintel]

Any solution to "something went wrong please try again" error when attaching doc or photo?

[wispy zephyr]

Is it just me or gpt45 has been silently reduced to 3 uses per day?

[eager latch]

Is it happening to anyone that perplexity is using regular search after selecting pro search?

And a small text on the bottom appeared saying I only have 2 pro searches left, even tho this is the first time I opened it today

My pro subscription is still active

[hoary phoenix]

I don’t think I’ve ever had that, and I must ask questions along with attachments multiple times every day. I’m sure you’ve considered all these, but the only things I can think that might have an impact are the file format, file size, stability of your connection (VPN? WiFi vs Ethernet? Mobile Data if on a handheld?) or maybe server-side issues.

[gleaming robin]

may i know that why i cannot choose AI models??

[crisp roost]

switch to pro or reasoning mode first

[gleaming robin]

i found it.. the model selection in secondary menu..

[sturdy skiff]

In both pro and reasoning section you can find

[flint steppe]

Hello

[hybrid zodiac]

why you have more ai models than i

[green crystal]

hah that didnt last long. seems like deep research now has its search depth reduced again. guess we're going to have to wait for deep research ultra mode

[verbal verge]

It seems like Deep Research's functionality and intelligence have been nerfed by Perplexity. From my experience, only the first Deep Research query follows my specialized prompts for enhanced thinking and high-quality responses. But for the second and third questions? The thinking process completely ignores my prompt instructions. It responds super quickly with half-baked answers - doesn't matter if I'm using high or standard mode.

The first week I used Deep Research was actually great though! The experience was so much better compared to now.

[green crystal]

It seems like a lot of features from Perplexity are being reduced, and the sub value is degrading over time. I wonder if they're taking their operational costs into account before releasing new features, because promising users features and then silently removing/downgrading them is not a good practice.

and they just massively upped their valuation so i dont think there isnt any money...

[queen girder]

[green crystal]

leave the reduced source count aside. the reasoning capability in deep research high seems to be reduced aswell.

[lone hound]

Yes

It is a very expensive model

About the use of search with the deep research model at a high level, when you search for something, add the following words: with extended details and extensive information, well detailed.

[hoary phoenix]

When a Thread gets very long with multiple questions and answers, it starts to run very slowly (answers’ text stutters as it appears on-screen, scrolling through the previous answers and questions isn’t smooth, questions take longer to answer etc). However, I still have more questions and need the answers to take into consideration the current context provided by the rest of the Thread. How can I start a new Thread (to get around the slowness), but give the AI access to all the context from the old Thread? Is it just a case of creating a new Space and adding the old Thread to it, then continuing to ask further questions in a new Thread in that Space? Will the new Thread be fully aware of the context in the old Thread?

[glad otter]

Agree its getting stupid. Yet I keep paying.

[wind bramble]

OpenAI livestream introduces three new audio models and tools for building voice agents. Video to text thread here: https://www.cofyt.app/search/audio-models-in-the-api-Ko6CApwf8D9D7l-VLOudyM

[river bronze]

Why not summarize it with perplexity instead 😄
https://www.perplexity.ai/search/summarize-this-video-https-www-NU3tTmQeSGCAMVjXIN1.8w#0

[wind bramble]

Excelent question! This is a perplexity inspired tool and while perplexity as improved youtube support i still trust this one to be more trustworthy (grounded in videos). No disrespect meant and its good to have choices. This tool says it can´t answer if it is not in the video and i like that.

Maybe you can try it / test it your self? throw a big yt video or podcast at both.

I still love perplexity tough!

[ivory bramble]

Its Rolling out mh

[modern echo]

Hi

[uncut tusk]

Guys, why is the "high" deep research not available on desktop? Sometimes it feels like both web and desktop versions are very different and away from each other.

[river bronze]

The apps always end up really lagging behind, same goes for mobile. I find it best to just use the website on all platforms

[uncut tusk]

Sadly, looks like this is the only way! Desperately need COMET browser.

[latent acorn]

GPT 4.5 gone?

[short narwhal]

Wonder what comet will have and how that’ll compare to dia in the long run

Although if it’s electron that’s gonna suck…

Don’t think so

[plucky bison]

How do you navigate to it?

[ivory bramble]

U click on pro search or Auto search and then navigate to Deep Research

[midnight belfry]

What are you using for your initial prompt?

[modern sinew]

@quiet dragon When is voice mode coming on windows?

[rare echo]

which is the best mode for coding in perplexity ?

[modern sinew]

Claude?

[rare echo]

in pro or reasoning ?

I'm so confused all the time to select the right mode !

[modern sinew]

I use reasoning for small tasks

[grand wharf]

how you got that model selection window

[grand wharf]

use 3.7 for frontend, r1 best for backend

[rare echo]

it was actually for some outliner detection , data cleaning , and plot generation !

[grand wharf]

what

[indigo valve]

Complexity Extension

They just updated it with the latest UI change

[grand wharf]

thanks

[primal raft]

I am in an enterprise plan unable to access API keys
Anything i am missing

[warped marsh]

@signal hamlet @halcyon coral hey is o3 mini in pplx pro o3-mini high or low?

[shell pewter]

does anybody have any idea when gpt 4.5 model will be available in perplexity pro again?

[lament turtle]

Reseting everyone on mobile to Auto model instead of whatever model they had selected before is a little bit shady in my opinion

[shell pewter]

im also in an enterprise one and 4.5 is gone for like 2 days now

[serene sand]

ＰＨＯΝΞ
ＵＰ
ΥＯＵＲ ＰＬΞＸ

[verbal verge]

Workplace Multi-Round Deep Thinking Framework
turn0 (Initial Analysis)
Analyze key points of work requirements/tasks

Clarify project objectives and delivery standards

Identify key stakeholders and their expectations

Determine timeline, resource constraints, and priorities

Integrate company background and strategic goals

Establish project mainline or core issues

turn1 (In-depth Analysis)
Summarize relevant information and resources needed for the project

Build logical connections between tasks

Consider the business purpose behind requirements

Utilize relevant professional knowledge and best practices

Design preliminary work plans and solutions

Review recent related project experiences and lessons learned

Analyze project risks and challenges

Evaluate alignment between proposed solutions and company strategy

turn2 (Solution Optimization)
Evaluate previous round of thinking against industry standards

Identify and improve weak areas of the solution

Reconsider client/supervisor specific requirements

Supplement missing important information or steps

Consider adding innovation points or efficiency improvements

Assess cost-effectiveness of the solution

turn3 (Detail Refinement)
Re-evaluate and improve work approach

Analyze needs of various stakeholders in depth

Focus on overlooked process details

Adjust communication strategies and expression methods

Prepare initial documentation and presentation materials

Consider potential obstacles and solutions

turn4 (Final Decision)
Compare strengths and weaknesses of all rounds of thinking

Integrate optimal solutions and alternatives

Determine final reporting style and professional tone

Conduct comprehensive document polishing and proofreading

Check if necessary data and supporting materials are complete

Prepare for possible questions and challenges

Organize final professional report or proposal

[indigo valve]

This is what I got from Google Deep Research. Let me know if any of these work for you

@zinc solstice sorry it looks like it didn't copy right lemme see

[old blade]

why the hell are they deleting 4.5???????

[dusk kelp]

Hello

[south shoal]

does everyone see the same news stories, or is there an algorithmic aspect to it, or based on our selections across interests

[umbral salmon]

Why does my AI randomly stop generating in the middle of my script

[tame current]

WAS GPT 4.5 REMOVED???

first it limited me to only 5 uses a day

and now it’s gone

it hasn’t even been that long since it released

[old blade]

yea frustarates me too. It was very good at specific tasks like writing. ....

It feels like they released it to gain new subs for "cheap" subsciption compared to openAI and get 4.5 but after gaining new subs they just removed it.

[indigo valve]

Was 4.5 really that big of a deal? It was slow and didn't really deliver any insightful output but it cost Perplexity a pretty penny.

[old blade]

Ofc. It was SOTA for writing text. Not even comparable close to other LLM. beside of that it was garbage

[grand bison]

Agreed. Upvote and bump the Pro feedback request here https://discord.com/channels/1047197230748151888/1352672220949778553

[dry kestrel]

gpt4.5 disappeared?

[sonic idol]

hello - currently trying to summarize a youtube video using perplexity but facing issue.
looks quite fine on the web app, but as soon as I try to do it through the api, it tells me it cannot fetch the video.

am I doing something wrong ?

[dim quarry]

same here

It's kinda annoying to see models removed without a warning. Especially at this price.

[dry kestrel]

yeah

[safe socket]

Hey everyone, I am non-stop getting API errors lately outo f perplexity, whether it's 204, 503, 500, you name it. I've changed absolutely nothing in my script. Is anyone else having this issue?

[pastel ruin]

Why don't I have the pro role 😢

[grand bison]

You have to join the discord server from the Perplexity mobile app, while logged into a pro account. I was in the same boat and left the discord server and rejoined that way

[indigo valve]

Do any of you guys invest online? If so, what do you guys use and why? I want to get into investing but I have no idea where to start

[drowsy oxide]

eh so, the desktop app doesnt have image generation thus i am required to use the browser for it yes?

[serene sand]

#1351405371411005450 message

please add your report here

[pastel ruin]

Their site link doesn't work?

[fervent star]

I'm not on pro and I'm here. So not sure what the deal is with that. 500 is that something asked the server to do something and the server had an issue but didn't know why it wasn't working, so it returend a 500. So it's either the server is being asked to do something it's not designed to do, or something on the server isn't responding properly to the request. The engineers will probably have to take a loook.

[pastel ruin]

/:

[halcyon marsh]

Hey everyone. I'm relatively new to Perplexity. Is there the ability to save prompts so you don't have to retype or copy and paste commonly used ones? Or is there an alternative solution for this?

[placid coyote]

look at "spaces". though they dont usually work that great with web search enabled. another option could be tags from Perplexity Helper

[halcyon marsh]

Oooh nice one!

Thank you!

[sterile barn]

Perplexity is by far the best for searching capibilites.

[distant aurora]

Real.

Good for research.

[silent pier]

code?

[distant aurora]

Inconsistent for problem solving.

[silent pier]

wht bout grok

[distant aurora]

Never tried grok for coding.

It's was fine for complex math I think.

For coding I would stick to O3 mini high or Claude.

O1 is fine too. Sometimes misses the context of original prompt but you can always remind it.

[silent pier]

so for students overall what subscription would be of most value? Perplex claude or gpt or grok

[distant aurora]

What student are you?

The only model consistent on perplexity with the official websites is R1 I think.

There are some people here who speak in favour of you.com

Tbh, you only need grok for quality deepsearch.

200usd of chat gpt is too high and perplexity deepsearch, though value for money considering the limits, can miss sometimes.

You should be fine with R1 for coding/problem solving I think.

I'll give you an example

If you ask

How many r are there in terrrrific

The O3 mini in perplexity trips.

[silent pier]

Im looking for creative writing, making essays sometimes coding diagrams or learning HTML or python and helping make revision sheets and analyze documents well and create images and ive been struggling to find an option from those

[distant aurora]

What have you tried so far?

[silent pier]

only chatgpt

[distant aurora]

Try notebook lm for documents.

R1 for coding ⬇️

chat.minimax.io

[silent pier]

i dont mind paying if there are better tools availible

but these i will def try

[distant aurora]

For images, idk tbh. I have heard flux is good.

The problem with Claude is
A. No web access
B. You'll hit the rate limits too soon

[dry kestrel]

claude just released webaccess

agree with the rate limits

its pretty atrocious

[distant aurora]

Yeah like you get hardly 50 prompts a day or something?

They over exaggerate the limits. Actual limits are way worse.

[silent pier]

apprently thats only if the responses are really long

what about gemini?

it doesnt even have limits

[distant aurora]

Divide those limits by 4 or 5 to get actual limits.

[distant aurora]

Okayish for coding. Nowhere near chatgot or Claude or R1.

[silent pier]

alright

thanks

[distant aurora]

@silent pier https://artificialanalysis.ai/leaderboards/models

Use the filters to compare performance.

[sterile barn]

When is perplexity going to release a text to speech feature

[distant aurora]

They don't?

Wait. I'm not sure. Lemme check.

[sterile barn]

They dont

[distant aurora]

@sterile barn is that not tts?

[sterile barn]

I dont think so

Where you can actually talk with the LLM

Like chatgpt and Grok

[distant aurora]

Oh. That.

[sterile barn]

Yea

[distant aurora]

It has an assistant mode.

@sterile barn

Here.

[sterile barn]

How did you find that

[distant aurora]

Need to sign in for that though.

[sterile barn]

Only on moble then it looks like?

[distant aurora]

Probably.

[sterile barn]

Im asking for Macos

Hey I just thought about it I can create my own LLM agent using Llama

Who wants to work on a LLm agent with me

[novel mirage]

How did you get them back?

[floral harbor]

hello? why gpt 4.5 disappeared????

[old blade]

use complexity newest version. There you have gpt 4.5 still. But Perplexcity reduced to only 3 request per day...

[floral harbor]

complexity?

i have neweest version

i dont have it on web as well

omg 3 uses

are you kidding

[old blade]

yea fr...

[floral harbor]

how do you access complexity?

[old blade]

what do yoe mean? im on version 1.9.3.0 and enabled all plugins 🙂 you should be good to go

[floral harbor]

i didnt know what is complexity

[sick arch]

google chrome extension

[old blade]

ahhh sorry. Yea its a chrome extension. Its opensource and defintly safe to use

[lone hound]

Let's wait if Perplexity implements Grok 3 in reasoning mode, hopefully it won't be expensive like gpt4.5 for example.

[floral harbor]

i used complexity

and no gpt4.5

[strong pond]

Did they take away math mode?

[old blade]

which version?

[old blade]

yea but u cant compare 4.5 with Grok3. 4.5 is really really really amazing at writing. nothing comes close

[cobalt vale]

voice mode is available on the mac app

[floral harbor]

newest

[harsh verge]

Unless you have downloaded the Early Build for Chrome from the Complexity Discord, you do not have the latest version. It has not yet been released on the Chrome Web Store.

So, head over to the CPLX Discord, go to the updates channel, and download the CRX file (the link is available there). If you don’t know the procedure, check the how-to channel for instructions.

The latest version is v1.9.3.

(And for the CPLX Discord, you can find the link here: https://www.cplx.app/ )

[tame current]

Do you guys think 4.5 will ever come back?

mods, wbu?

it was helping me so much with learning songwriting 🥲

[strong pond]

I don't think so

OpenAI will probably release a better model before gpt4.5 becomes cheap enough

[lone hound]

If you want to use gpt4.5 without problems, you will have to go to the official chatgpt website and buy a subscription.

On Android, GPT 4.5 is no longer available, I don't know if they have permanently removed it or not.

[strong pond]

You can still use it via the rewrite menu

At least on the windows app

It also says 3 but it lets me use it as many times as I want- at least seemingly

[elfin gate]

I don't have it in rewrite options

[lone hound]

Have you tried restarting the application?, if the response is slow as such, it's probably still working.

[faint river]

HoniQKekzZ

[green crystal]

why did deep research options disappear?

[lone hound]

It is no longer available.

[tepid laurel]

why cant I change the model anymore after the update ?

[torn ravine]

Now it seems high mode has become the default

[old blade]

i can use it without problems in Perplexity. With Complexity 1.9.3 as i told you.

[spiral python]

There seems to be an issue with high deep research

I had over 300 sources yesterday and now it can't do more than 100.

[sterile barn]

Ok awesome they should still have it for browser though

[old blade]

Also here. With Complexity you can still select normal or High

Best thing it doesnt switch back to Automode after a request.

[spiral python]

thanks

which plugin is this

[old blade]

Complexity. Get it from github if you are on chrome. Firefox has the newest version in the extension store. Chrome takes some hours/days so you better install it manually

[spiral python]

https://github.com/pnd280/complexity

this one?

"Local development (only works on Chromium-based browsers)
pnpm dev" do I just paste this on cmd?

[old blade]

https://github.com/pnd280/complexity/releases/tag/v1.9.3

download the crx file and drag and drop to your chrome extensions.

[spiral python]

oh cool thanks!

A quick question, do I use pro infinite times a day or just 300?

This looks awesome, what does it do?

[green crystal]

with the amount of sources and search steps being way less?

[green crystal]

its sitting around max 60 for me with way less output

[drowsy harness]

Bro what happened to the new high deep research? I loved it 😦

[fathom moss]

they removed it i guess

[drowsy harness]

No reason?

[fathom moss]

I just noticed I took a break came back and it was gone

Maybe High is the default now? Who knows.

I'm switching to R1 and will finish with Deep reasoning to get the results I need for now.

[drowsy harness]

OK it's most likely not default and they actually remove it. For context the 250+ sources is the perplexity deep research high before they removed it vs the one now.

[spiral python]

exactly

if they want to step up their game they should allow 300 sources minimum, gemini can go up to 1400

[drowsy harness]

It was possible with perplexity deep research high before they removed it. Some people reported perplexity high researching over 400 web pages.

[spiral python]

I use deep research for work and I could tell you with certainty that gemini and grok deep research is the current best, perplexity gotta do more

[drowsy harness]

Grok is pretty good and free. Problem is it always follows a certain format for its answers, which makes it pretty generic.

Overall it's good, especially the new deepersearch

[spiral python]

I think with current deep research they chose a middle ground

I'm getting 200 sources now

for tasks I used to get 300

[drowsy harness]

Did you use very specific prompt?

[spiral python]

160 sources now actually

yes

[fathom moss]

What Engine would you use for formal writing just interested in what you think? I'm looking for an alternative now that i got used to Deep research High is gone.

[drowsy harness]

If you are specifically asking for writing, Claude is the best imo. But if you want to cross the line and go to NSFW story writing or stuff like that then id go for grok.

[spiral python]

I think she meant academic writing no?

[cobalt vale]

I learned that AI companies only communicate impactful changes to their service when they feel like it

[spiral python]

Then grok for sure

[drowsy harness]

Claude is good too

[spiral python]

I'll be telling you if they indeed removed access to high or not, I can still have access with complexity extension but I'm not sure if it's defaulted or not

No they didn't remove access to high, it's just worse than what it used to be

[fathom moss]

yes it is a extremely difficult topic

[cobalt vale]

uh I ran a deep research high and got 66 sources, then I ran a standard deep research using the same prompt and ended up with 157 sources?

[thorny grotto]

Hi! Is anyone working in a customer support job and using AI to prepare docs for LLMs and RAG? I'd love to know what your experience has been like with regards to structuring, providing context, metadata, and what LLMs you use? Thanks!

[left stream]

Time to find out which is the best of the best

[indigo valve]

Is perp's on "high?"

[shell pewter]

lmk too gng

[verbal verge]

Awesome, the original Deep Research is back!

[indigo valve]

Damn that's a lot of sources

[paper violet]

how to use Claude opus in Perplexity? who know?

[verbal verge]

Alright, it's still the nerfed version from yesterday.

[spiral python]

Yeah lol

[left stream]

Perplexity, gemini, and ari all done. Waiting on chatgpt’s deep research

[left stream]

I’m using your question as an example

[indigo valve]

Lemme know what ya get

[left stream]

All of them are long asf 😭 how should I do this?

Want the pdf too?

[indigo valve]

nonono that's fine you didn't need to actually supply them to me lmao

I was just curious on which one you liked more

[left stream]

Gonna test them out

Was surprised ari had the most sources and was the fastest

Chatgpt’s the one that takes the longest

Google definitely has a bunch of context memory and with upgrades gemini 2.0

[rich vortex]

blah

[indigo valve]

Just because it used a lot of sources doesn't mean it can articulate the information I've learned

[indigo valve]

I love the cover page

Damn this looks clean with a ToC

It's very businessy tho

[left stream]

This is gemini’s deep research

[random ocean]

Where’s Deep Research High?

[left stream]

Chatgpt’s deep research

[random ocean]

So, which one gave the best answer according to you?

[tame current]

what happened to the high setting for deep research?

[drowsy harness]

Grok?

[torn ravine]

Uh no. I started a new thread and the deep research seems to be the standard one, not the multistep one we had a while ago😭

[random ocean]

That multi-step deep research feature accurately answered some queries that standard pro modes couldn’t handle correctly. They should definitely bring it back.

I meant the high one that used several steps and several minutes to answer.

[torn ravine]

yeah that’s a killer feature that brings pplx to the top tier

[distant aurora]

Give me a list of how many subscriptions you got.

[left stream]

all of them

[silent pier]

whats ur fav one Baldwin

[left stream]

I have a feeling high deep research by perplexity would’ve been my favorite but I was late to the party this time 😭

I didn’t even know it existed till now and people are saying it got nerfed already lol

[left stream]

Hard to decide between ari and chatgpt’s deep research ngl

[distant aurora]

What's Ari?

[distant aurora]

Bro is batman irl.

[distant aurora]

Have you compared chatgpt deepsearch with grok deepsearch?

[left stream]

Oh shoot forgot about grok will do that rn

[distant aurora]

Ngl but R1>>standard pro search.

[cobalt vale]

I heard that chatLLM also provides deep research?

[distant aurora]

Wait

Chat llm is abacus ai?

[silent pier]

tea

ye

[latent acorn]

GPT 4.5 gone?

[random ocean]

Yup

[green crystal]

Told you

Perplexity subscription depreciates like a car

Grok so far has been good
But it personally doesn't sit well with me to purchase a subscription there and grok isn't great with math at all

[green crystal]

and now another version of deep research is out.
i'd honestly prefer some more transparency from the team...

[random ocean]

Bring back Deep Research High or else! ______ You all can fill in the blank.

[crisp roost]

what happened to perplexity’s deep research

[random ocean]

NERFED

[green crystal]

its changing allot but they're not mentioning it

[crisp roost]

were we previously using deep research high and have now been downgraded

[green crystal]

the new version is where the reasoning and steps are much better
you notice it when it doesnt always start with "i found"

and instead actually thinks after each search

the nerfed version is almost as shallow as just normal reasoning search while being slower

[lone trellis]

why they removed research HIGH and GPT4.5? and android isnt getting the new voice

[random ocean]

Two days ago, the "High" setting for deep research was excellent—it consistently ran for about 10-12 steps, took roughly 5-11 minutes, and produced detailed, high-quality answers. Some responses even exceeded 6,000 tokens in length without citations.

However, yesterday, the "High" deep research setting was noticeably nerfed. It became similar to today's "Standard" setting, although it still managed to produce answers around 3,000 tokens long. Now, we're back to the standard deep research, which is limited to approximately 3 steps and doesn't achieve the depth or quality of the original "High" setting.

If the issue is that Perplexity can't sustain 500 queries per day at the original, resource-intensive "High" setting, perhaps consider reducing the daily limit to 50 or 100 queries instead of downgrading the quality of the feature itself. Very few users actually reach 500 queries per day, so limiting the quantity rather than the quality would be a better solution.

Please bring back the original, high-quality deep research setting!

[green crystal]

for me it seems to be back "for now"

seriously some transparency would be nice

when you count on it the new updates just vanish

[random ocean]

Not for me, can you share a recent deep research answer?

[green crystal]

218 sources

also its annoying that it reverts to auto mode every time....

[random ocean]

Can you check the length of the answer? Use https://platform.openai.com/tokenizer

@green crystal

[green crystal]

8381 tokens

well i wonder if i'll still have this version of deep research at the end of the day lmao

ah think i already lost it

[random ocean]

It’s 218 sources for you, 20 sources for me. Also, length of answer is 1537 tokens against 8381 for you. @green crystal

[grand wharf]

is deepseek-r1 currently not available?

self-resetting

[random ocean]

That sums up the difference, I guess

[tribal ridge]

thats deep research (high) ?

[green crystal]

yeah

but the button wasnt there

just deep research

i think they're randomly testing

but smh transparency would be nice

[tribal ridge]

test it on sports betting

[random ocean]

They probably doing A-B testing

[tribal ridge]

lets see if its accurate

[green crystal]

yeah because now its gone for me lol

[random ocean]

Jinxed

[tribal ridge]

regular deep research hallucinated a ton

when tested on sports

[green crystal]

the search depth is really shallow too

[random ocean]

Standard current one has improved, but it still hallucinates. I couldn’t find any obvious hallucinations for high one though.

[tribal ridge]

when did they improve it

[random ocean]

Like in one of the answers, it contradicts itself at start and end of the query.

[random ocean]

I think they moved away from R1 1776

But not sure

Or changed something internally.

[tribal ridge]

deep research high must beat open ai's right?

[random ocean]

It doesn’t start with “okay, the user wants me….”

[tribal ridge]

on humanity's last exam

[random ocean]

It's significantly more cost-effective and supports a higher volume of daily queries, so even in close comparisons, it will stand out.

[torn sierra]

Haha. Open AI is something that hast been replicated yet. It’s really amazing compared to what we have in Perplexity. Even Gemini Deep Research with new Gemini thinkining model is much better than Deep Research High in Perplexity. At least what I think. You can check it for free in Gemini and compare

[indigo garden]

what about gpt 4.5???

grok3

[full fern]

Hello, could you please explain why the DR HIGH and STANDARD options were removed from Perplexity? Any idea?

[naive venture]

[tough estuary]

[native maple]

Hello! I have a free year of Perplexity Pro code, where can I redeem it? I've already have one moth pro sub. Thank you!

[valid charm]

Sucks to have features disappear. It's not that hard to post text only updates on changes like OpenAI does...

[green crystal]

Hah deep research now only does 1 to 3 steps now

Search steps that is

Counting in reasoning steps is a bit misleading imo

[lone trellis]

I used complexity to still use the un-nerfed high model but now it was completely removed, I saw that now it says "I found, I found I found..." Which suggests that it's using the standard nerfed model 💩

High is definitely removed

[green crystal]

Yeah high was reasoning after each search

https://x.com/AravSrinivas/status/1902876897773760577?t=ACQbrxMr_A0zGRY3v4TuzA&s=19

Aravind Srinivas (@AravSrinivas) on X

We're working on an updated version of Deep Research (shipping next week, hopefully). This will throw even more compute, think longer, present more detailed answers, use code execution, and render in-line charts. Example: Asking to reason how AWS can hit a $10T market cap.

X · Mar 21, 2025, 12:16 AM

I think we're really seeing A/B testing

[void sorrel]

here it still works

[green crystal]

Now it would be helpful if all these announcements were made on their site or here in the announcement channel

[vestal cove]

hello my fellow ai users!

[sturdy skiff]

ok whats next

[random ocean]

It says 3 enhanced queries remaining today in this X post by CEO. So, the limit might be low.

Or maybe CEO was using the free version of perplexity.

[green crystal]

Free version most likely

[modern crown]

Why did they take away GPT 4.5 lol

[verbal verge]

GPT-4.5 isn't really that good tbh. I'd actually prefer if they enhanced AI's memory capacity instead of just 32K. Or maybe bring back the old Deep Research feature that we used to have.

[green crystal]

We'll get the better deep research back "hopefully" next week

[hoary phoenix]

Is the ability to convert a Thread into a Page still available/available in the iOS app?

[torn sierra]

I guess it’s called cheating. I’m not participating in something like this. Bye. https://x.com/aravsrinivas/status/1903238228771213803?s=46

Aravind Srinivas (@AravSrinivas) on X

Should Comet disable Youtube ads by default for all?

X · Mar 22, 2025, 12:12 AM

[dry kestrel]

why on earth do followups revert to using auto

whats the point of buying pro then lmao

[placid coyote]

I found 4.5 really good for writing (stories, but also greentext). btw I am pretty sure the context window is ~64k. at least numerous experiments with file RAG suggest that

[harsh verge]

It is, to say the least, completely unacceptable for a CEO to highlight this point while his company imposes unsolicited advertisements on us despite our status as paying users. Moreover, this is not the first time his statements have fallen short of the standards expected for such a position.

[sturdy skiff]

i see the same Deep Research as before, what is the latest update?

[limber monolith]

i assume to train the perplexity model and dont waste too much capacity. people tend to overestimate their prompts

[distant aurora]

Ikr

[dry kestrel]

i chose to use reasoning because I want a high quality answer

[distant aurora]

Try this

[dry kestrel]

I dont want some ai slop

[distant aurora]

How many r are there in terrrrific

Lol.

[limber monolith]

I prefer it as well.
or Deep Research (high) for multi step tasks

[distant aurora]

@dry kestrel rather just stop it in the middle of generating response > regenerate > pick model

[dry kestrel]

Yeah they just made it harder to use pplx

[sturdy skiff]

someone help me understand please @dry kestrel @distant aurora

[distant aurora]

Copilot has this issue too. You have to select thinking model with every prompt.

Looks like a subtle way to discourage pushing rate limits.

[dry kestrel]

very "subtle"

[distant aurora]

More like a downgrade. Can't even pick between O3 mini and R1 anymore.

[dry kestrel]

Apparently its intentional https://www.reddit.com/r/perplexity_ai/comments/1jgmcor/why_follow_up_question_automatically_switches_to/

[limber monolith]

damn you are right there is no deep reserach high anymore 😄

[dry kestrel]

official dev got 27 downvotes lol

[sturdy skiff]

i can still pick

but not understanding what people are discussing about deep research as i can see old Deep research still available

[distant aurora]

Deep research had 2 variants earlier apparently.

[sturdy skiff]

yeah they did silent release and also removed silently

[distant aurora]

@dry kestrel bro, just use chat.minimax.io

[limber monolith]

because deep research high was awesome, captured like 50+ very structured sources and provided amazing results

[distant aurora]

They offer Deepseek R1 for free.

[dry kestrel]

good stuff

[limber monolith]

i need DeepSikh for software testing

[distant aurora]

The regular one captures 50+ sources too.

Deepseek is goat.

[limber monolith]

actually it is 😄

[dry kestrel]

yeah the high/standard thing wasnt working

it was just using standard

[limber monolith]

some uses Grok2 in daily work / tasks?

i just ignored it

[limber monolith]

true.. but using the high model was still not comparable..
What i generelly miss is something like "branch out" the chat into a new one, with more context. Like openweb ui does

[distant aurora]

Ah

I have never used open ai. I like grok deepsearch.

Even Google doesn't compare.

[dry kestrel]

i feel like perplexity deep search is still a bit better than grok deeper search

with less hallucinations and more detail

[distant aurora]

I thought it's the other way around.

Or more like perplexity gives a bird's eye view.

[dry kestrel]

Let me try to get an example

[distant aurora]

Sure

[cursive jacinth]

too bad not only they removed deep research high but also nerfed claude reasoning sir, it is not following instruction good as before sir. Sad sir. it sound also like deepseek sir.

[dry kestrel]

deep research high never worked

dont think it was intended

[valid charm]

Are there admins from Perplexity here? I'm wondering if they realize how disruptive their lack of communication is?

I get more info from this user chat than their webpage lol

If they can no longer afford high, why no add a limit? Or even add a new tier with more uses? Rather than crippling it for everybody

[limber monolith]

i guess there is a limit, you just dont see it 😄 context just decrease, like chatgpt does. until it starts posting weird stuff. Thats why i said: branch out a conversation into a new one, would be good. I know , Spaces are kind of similar to that, but not every request is a full blown project, at least to me

[sick arch]

i asked this question 😅

[valid charm]

At least we have Gemini Deep Research. (kidding, it sucks)

[sturdy skiff]

its useless, it cant deep research on upload documents, it dont have that feature

[valid charm]

Yep lol! I always forget it exists because it never solves a problem for me

Actually, if you are analyzing YouTube performance, Gemini is useful. Probably because it is trained on internal YT documentation.

[sick arch]

gemini is good at youtube video where it can easily analyses what is being discussed in any language.

[rapid bane]

Who to generate image with perplexity ?

[green crystal]

Apparently it's A/B testing
CEO posted that it's all released "hopefully" next week

Ah and he posted that on X

Not in the server or blog...

[distant aurora]

Bro

340 sources already

[serene sand]

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

Tutorial

[distant aurora]

https://grok.com/share/bGVnYWN5_c6e16a38-e91d-4bdb-9970-d1b95d60c5a9

[chrome token]

write a blog about weight loss

[distant aurora]

Calorie deficit

The end.

[sturdy skiff]

Eventually, I hit the context limit with the O3 Mini reasoning model in Perplexity. After a while, it started forgetting the original objective and began bluffing—just repeating what I was asking in each query, without actually following through. I later realized the last three responses were completely off and wrong. I had to switch to deep research mode, and only then did it apologize and correct its earlier answers.

[distant aurora]

I hate it when this happens.

[sturdy skiff]

they are never clarifying above context window sizes of other AI which are used in perplexity, that is too much of a chaos and confusion to a customer

[dry kestrel]

its 32k

[dry kestrel]

i got this after trying again https://grok.com/share/bGVnYWN5_ea4a636b-23f7-4168-a889-7d76c7f0a3ae

[sturdy skiff]

no man, that is file upload limit context window, not chat context window, my tickets with them are still open about this questions, they didnot answer yet

[dry kestrel]

that is the chat context window...

https://www.perplexity.ai/help-center/en/articles/10354924-about-tokens

[distant aurora]

Maybe app and web work differently.

[dry kestrel]

just grok being unstable

[distant aurora]

Inconsistent

[dry kestrel]

but i got around 80 sources each time i tried

80-100

[distant aurora]

Mine was showing 340 sources initially but dialed down to 143.

Maybe it omitted repititions of same info.

[sturdy skiff]

not at all, i already have email confirmation from them, 32000 is upload context window for pro subscribers, for chat context window they confirmed me it is 128000 for sonar and deep research, for other AI models still they didnot confirm me

[dry kestrel]

https://www.perplexity.ai/help-center/en/articles/10354924-about-tokens

deep research is different to reasoning or pro search

all the models in pro search have a context window of 32k

[distant aurora]

Chatllm/abacus ai won't let you use without credit card details.

Glad I made burner account first.

[sturdy skiff]

no

[sturdy skiff]

i got confimation over email deep research is 128000 context window, also for sonar as 128000, for remaining models still they didnot respond

[sturdy skiff]

see part of email which i received

[distant aurora]

Huh

1 million

https://www.perplexity.ai/search/chat-window-context-limits-on-POOgfy4KQ4Gqk3QqJSX2Tg

[sturdy skiff]

sharing one more, this is email received about 32000 context , that is related to file upload for pro users

[night garnet]

Hello! I have a question.I have the Pro version for a year, which I received as part of a promotion from my provider. I need access to the API, but to get it, I have to link a credit card. Will linking the card change the terms of my subscription, result in different charges, or can I rest assured that nothing will change for the next year?

[dry kestrel]

nope a lot of people tested this

gemini does not have 1m context window on pplx

[dry kestrel]

32k context is for pro search that uses claude, gpt4o, etc

thats different to sonar deep research

[sturdy skiff]

that is not confirmed to me over email by them, i am still awaiting official information from them

[sturdy skiff]

gemini doesnot work good in its own website with 1M context, so lets not scold perplexity due to geminis bad performance

[dry kestrel]

its in the documentation here https://www.perplexity.ai/help-center/en/articles/10354924-about-tokens

[dry kestrel]

search for yourself other users using gemini on pplx lol

I am not having this conversation, try it out for yourself or search online

[sturdy skiff]

you have ego problem man, i clearly showed you email proof they said it is upload context window that 32000 and not chat context window, and still you are sharing link to prove your point, and in the link also it is clearly mentioned it is upload context window 😄

[velvet vault]

How do I get it to generate images?
People are saying there's an option on the right but I can't see it

[dry kestrel]

Its weird how they did it, i often just try asking it to "search for a" and then whatever you want

There should be a button beside the answer with "images"

[lone trellis]

The high mode captured 100-300 sources!

[dry kestrel]

Click that and generate image should appear on the right

[dry kestrel]

Do you have an example, was trying to compare high mode to standard when it was still avaible and I was getting similar results

And I couldn't replicate the chart generation capabilties that was showcased on twitter/x

[velvet vault]

Yeah thanks I've got it now 🙏

Really appreciated

[dry kestrel]

np

[distant aurora]

300 is a lot. Why would they nuke that?

[dry kestrel]

tbh i just think that they accidentaly made it avaible to everyone

and it was never meant to be released until like next week

[limber monolith]

That even happens with openAIs o3 mini, sometimes it just breaks, try to give it reminders, like:
"Please name this created (for example: table): Table123.xlsx, so we can refer to it later on"

[lone trellis]

Because high mode only searches for useful information, if it needs more, it can easily take a big bunch of sources

[serene sand]

#💬│general message @dry kestrel

[sturdy skiff]

ok thank you 🙂

and i just want to confirm everyone, deep research context and sonar model chat context in perplexity is 128000, and for other AI models they didnot confirm me chat windows context, i will share immidiately once they respond me, but the ticket is open from very long

and dont get confused with 32000 context as AI model chat context, that is completely separate feature for file upload for pro subscribers, below is part of emails i received.

[lone trellis]

How they nerfed Claude?

[cobalt vale]

But the supposedly standard response also went through significantly more steps

[left stream]

Deep research by chatgpt. Second place might go to you.com’s ari ngl but it would’ve been most likely different if deep research high wasn’t nerfed

[left stream]

Heard about thay

Is it good?

[proper lodge]

is it just me but is pplx crashing every time i try to use claude?

[cobalt vale]

I don't know, I thought you wanted to test them all

I doubt it is better than perplexity

[placid coyote]

and dont get confused with 32000 context as AI model chat context, that is completely separate feature for file upload for pro subscribers, below is part of emails i received.
yeah, that is minimal guaranteed context for a file. in reality I for months never saw it bellow 62k (mostly used in pro sonnet). I would guess it is similar to that "300+" pro uses daily, which in reality was only shortly down to 450 I think, and in days back to 600

and also it is important to differentiate search and spaces, because spaces use entirely different RAG which is useless for programming and other similar use cases (I would guess math, anything symbol heavy or needeing longer chunks) https://monnef.gitlab.io/by-ai/2025/pplx-tech-props

and lastly, the 1 million ctx window is not working and probably never did. on x ceo and official account never bothered to even respond. same thing in the issue here: https://discord.com/channels/1047197230748151888/1344391265491484733

[sturdy skiff]

yeah we are on same thought of this discussion, let me add two points about RAG and 1M context:

my use case is RAG, but not coding, so perplexity is good for my use case, i mean especially spaces.

now about 1M context, i got earlier response from Perplexity support, they told that 1M context works only with "Auto" mode

[green crystal]

It's A/B testing apparently. Some announcement would clarify things allot

[limber monolith]

and also it is important to differentiate search and spaces, because spaces use entirely different RAG which is useless for programming and other similar use cases (I would guess math, anything symbol heavy or needeing longer chunks) https://monnef.gitlab.io/by-ai/2025/pplx-tech-props

I (more or less) coded a full-blown application with it, using claude 3.7 reasoning and Deep Research (sometimes).
Starting from visioning the complete idea, to getting a React Frontend, ready for N8N workflows, Huggingfaces and NLP ready, tested with dummy data, built different features - and a MVP, created tests, etc.
At the end i moved to CursorAI, but the whole basic stuff was built with SPACE in Perplexity.

BUT: I can not realy say much about the code quality, because it was literally my first AI Project related to coding. I have 0 experience

[sturdy skiff]

how to paste table in perplexity as text ? 😦

when i paste it pastes as headers followed by their values, but not one below other, no identations nothing

[placid coyote]

"Auto" mode, that was the first thing I tried... https://monnef.gitlab.io/by-ai/2025/pplx_M_ctx

[placid coyote]

if you really want to paste INTO pplx, then you probably want the data converted to csv or markdown first.
if it's the other way around, getting from perplexity, then under table should be save as csv, or export button where you can select formats

[sturdy skiff]

not getting from perplexity, i mean to send in chat as table text, because i have multiple to analyse and cant always input as csv for each table

[spiral python]

Does having more than 1 prompt running in perplexity reduce the amount of sources per prompt?

[placid coyote]

it depends heavily on language and how ai-friendly the code is. for example spaces rag (I mean when you upload files to space; not a new thread in a space). in haskell, which heavily uses special symbols and has meaningful spaces (indentation denotes code blocks), it was broken most of the time.
but it took a minute to find out why sonnet seems to suddenly become so bad at haskell. it was becouse he was using crippled code from the RAG, but still in many cases managed to mostly restore the code, before it started workin on it. I was kinda blown away when I discovered Sonnet was soo good at reconstructing such heavily damaged code. After swearing at him why he keeps messing up seemingly simple stuff, I suddenly felt a deep respect. I would take me minutes to tens of minutes to recover code in such state...

[placid coyote]

shift+enter inserts a new line. but better would be to get the data in nicer format

[placid coyote]

or if you have smaller tables, but several, you could try to preprocess them individually. if more tables are confusing ai

[sturdy skiff]

created custom gpt in chatgpt to convert pasted content as markdown, in gpt there is flexibility to paste content as image (by default), so in perplexity we dont have that option to paste as image, instead as text, i mean snipshot image is not possible in my case as table row is very lenghlthy

once custom gpt generates markdown, i will use that in perplexity for my queries

[placid coyote]

you still can do this in a "space" (those instructions to convert it, and without web search) if the tables are not too long (shorter than 8k characters I think, before turning into a file)

those bigger models are pretty good at guessing headers and cells, eg https://www.perplexity.ai/search/convert-to-md-table-pls-castru-AGaCO.FWRA269el0LxpDBQ

[sturdy skiff]

will try 👍 , thank you

[sturdy skiff]

perplexity spaces is really dominating any other AI i used for document RAG till now ❤️

[placid coyote]

btw this is a nice meta-prompt for a space. just copy the block and create new space with that instructions. it is usually quite good converting vague/messy instructions (like those for custom gpt or a space) https://www.perplexity.ai/search/respond-exactly-only-with-this-n1DAY98jTRurUsjPV2k8VA

[little zephyr]

Fr 🔥🔥