💬│general

[tame current]

I ma using brave

[tame current]

Might wanna ask r1 lmao

[soft compass]

Yeah i haven't used brave, but if you go to the extensions page, there should be an update option somewhere

[hasty coyote]

shangus

[empty lotus]

Now settings is not showing you have X number of reasoning left for today field.
I think it is unlimited

[olive solstice]

Can we now use deepseek r1 and o1 on iOS?

[livid aurora]

https://discord.com/channels/1047197230748151888/1333489839127134240

wtf

It's not searching the web anymore

[empty lotus]

deepseek runs this as their side project with spare GPUs. Perplexity could give better limits imo.

[tame current]

"DeepSeek R1 is now available on Perplexity to support deep web research. There's a new Pro Search reasoning mode selector, along with OpenAI o1, with transparent chain of thought into model's reasoning. We're increasing the number of daily uses for both free and paid as add more capacity during the day. Stay tuned!" We stayed tuned and got nothing

where s the official comment on rate limits?

[zenith berry]

@everyone hi

[tired heath]

Hello, why did you ping everyone tho? You need the perms to do it

[harsh stag]

the version 1.4 of the extension on chromium browsers is currently being reviewed by the chrome store, it should be updated within the next day

[torn sparrow]

is it just me or perplexity's o1 feels worse than the o1 on ChatGPT site

[split dome]

Hey, does the O1 model have no limits anymore ?

[mint hatch]

It's much worse. They've just updated it yesterday and somehow managed to broke it

[split dome]

Hey, does the O1 model have no limits anymore ?

[green helm]

Its 15x per day I believe still

[empty lotus]

The only use of perplexity is citation. If someone implements that, I'll moving the same day.

[torn sparrow]

😭

[stark matrix]

it says 10 for me 😭

to me it's still crazy that we are being limited on the reasoning models if we're already paying for the Pro subscription

[buoyant raft]

what kind of watered down R1 is that? and a limit of 15??? i didnt even pay for my perplexity pro and still want my money back! give us the REAL deepseek and keep your OpenAI tokens!!!!

[proper mural]

15 per day? Woah… i currently dont see any “uses left” but i knew it would be too good to be true😔

[buoyant raft]

also the answers are worse than on the free deepseek app, thought process is difficult to access and probaily incomplete. watered down for sure! tinfoil hat mode they dont want us to see the real power of deepseek !!!

[proper mural]

Honestly might just switch to the deepseek app, i only see complaints here on perplexity 😵‍💫

[buoyant raft]

the best model on Perplexity will forget what you asked about after 2 replies. Deppseek remembers stuff after hours that you completely forgot you talked about, its friggn wild!

[wheat dagger]

True , O1 was far better when it was an actual model instead of this cheap reasoning version they implemented on despite having 10 daily usages
Man i miss O1 model so bad

[proper mural]

Same

[wheat dagger]

Probably i will change my subscription to chatgpt after the end of this month

[proper mural]

I mean i got Perplexity for free so i can’t really complain

For a year

[buoyant raft]

Probably cuz all of OpenAI computing power is needed by Israel / IDF for … you know … the things they are up to nowadays.

[alpine flicker]

And here I am still waiting for true AI to have robot maid :/

[halcyon coral]

Hello! Please avoid discussing political topics, thanks!

[buoyant raft]

just wanted to state my dissapointment, but yes, will avoid it and going back to "chat about math, coding, and logic problems instead" lol, good luck perplexity

[hushed plaza]

so now pplx is useless

[proper mural]

Technically yes

[kind rampart]

Allowing Perplexity subscribers to integrate custom Deepseek API keys would be an excellent solution, as the current daily limit of 10-15 requests underutilizes R1's premium model despite its advantageous pricing structure.

[sacred sand]

I've reported this twice never got any answer

[tame current]

even the r1

[sly ibex]

I am not able to access the API can someone help me with it

[acoustic salmon]

Having both atm, I don't find o1 as useful on ChatGPT's site as it can't do even basic searches and has an old information cutoff date.

[torn sparrow]

o1's main capability lies in its reasoning, i feel like im using a heavily nerfed version on perplexity

[dreamy kraken]

are we getting regular deepseek too anytime soon?

[torn sparrow]

i suspect perplexity is running a distilled version of r1, which is fair, since the real version requires a lot of computation
but then i dont see any perks of choosing this over just using the actual deepseek

[acoustic salmon]

I haven't seen any reduced capability in Perplexity's o1, just the limits on usage

[acoustic salmon]

your data is a big one

[torn sparrow]

that's not necessarily a big concern for me, might be different with others

[acoustic salmon]

it scares me that people say that lol

[torn sparrow]

i just dont give out any important data in my prompt, thats all

[acoustic salmon]

your data not only affects you, but anyone you contact or are even near. It's the big reason national security wants data housed in the US, because we have 4th amendment rights and such.

you don't have to actively give out any information is the issue

[torn sparrow]

¯_(ツ)_/¯

[tame current]

they again reduced r1 from 600 uses to 15

combined with o1

but r1 responses have improved

[proper mural]

😵‍💫

[tame current]

i dont know how to feel about this

[acoustic salmon]

I don't currently have any limits

[tame current]

u do u cant see them

[acoustic salmon]

it seems to come and go as their rolling it out

[tame current]

unless you install complexity

[proper mural]

It said that for me too, but once i used 5 it told me i had 10 left

[shut dagger]

SAMEE!!

[tame current]

we're cooked

[shut dagger]

what are they smoking?

[tame current]

not sure

[acoustic salmon]

just let it roll out lol, it changes constantly

[tame current]

yeah

they're experimenting

[shut dagger]

like a new developer's website

who changes stuff constantly

[acoustic salmon]

it's brand new so expect growing pains until they land on what they're comfortable with, and then we can come in and give feedbacks

I would hope R1 has no usage limits in the end given the supposed cost efficiency

but we'll see

[tame current]

i think something like 400 per day

is okay

[acoustic salmon]

idk, if it's 90% redcued cost to current models, models that we use unlimited atm, then I don't personally see why it wouldn't just become another standard model to choose

[tame current]

um

it has become standard model

[acoustic salmon]

but yeah, they'll find out the actual compute requirements and then we'll see what happens

[tame current]

but it has 15 limits

with o1

and other models like sonnet also have limits

i believe its 600

[acoustic salmon]

right, I don't consider those standard models, those are "pro" models

[tame current]

oh i see

[acoustic salmon]

but you're right, if Sonnet has 600 uses (I obviously just won't hit that) it's not unlimited

[tame current]

mhm

so something like 400 is possible ig

lets just wait hm

also janus dropped

i hope they add it in image gen models

also a seperate image gen tab istg

[acoustic salmon]

yeah sounds interesting

I would also like an "image" focus in our list

[tame current]

works

[vapid onyx]

im glad i can test deepseek without having to download another app thats for sure

[swift sable]

Yes

[livid aurora]

How many sources do you get approx for your Pro Searches?

[indigo valve]

I think up to 20 depending

[cursive jacinth]

R1 is 10 only sir?

[woven cosmos]

15

[cursive jacinth]

why mine shows only 10 sir?

[livid aurora]

I only get 12

[woven cosmos]

I just used a query so 15-1 = 14 x)

[cursive jacinth]

ok when one use R1 your o1 is used up yo sir???

[woven cosmos]

Yes sirrrr

[cursive jacinth]

is R1 that expensive sir???

[woven cosmos]

Nop

[cursive jacinth]

why is it 10 to 15 sir?

[cursive jacinth]

ok its super cheap sir, does anyone know why perplexity made it just 10 to 15 sir? cant they afford that cheap sir?

or cheap is the new expensive thing in this world sir..

[halcyon zodiac]

Yea, I’m curious why they’re limiting r1 queries as well since it should save them money over api calls to the other premium models

[vapid onyx]

Curiosity, for those of you who have ChatGPT and use perplexity, why?

[sick arch]

r1 search limit should be a minimum 100, 15 is very less

[echo swift]

Maybe because Perplexity is hosting R1 on their server, and don't want to blow up the server by the massive incoming traffic atm. I think they'll increase limits soon as soon as it stabilizes

Just waiting for R1 available on mobile

[magic talon]

Can't they technically have something such a cache version or offline backup

[cursive jacinth]

one thing is fact in this server sir, they love not to answer important questions sir. 😂

anyone noticed that sir? for a long time sir.

[queen valley]

are pro searches disabled right now? I almost never hit my limit.

[acoustic salmon]

they stated it's limited during rollout and will increase in usage allowance

[queen valley]

and now they are back.

whacky.

[cursive jacinth]

where sir?

[echo swift]

X sir

[wheat elbow]

Guys, can you contact me who has information about how I can translate the text on the image into English without distorting the image, using artificial intelligence?

[unborn flicker]

r1 is trash anyway...

[cursive jacinth]

overated?

[unborn flicker]

yeah just a ripoff

[acoustic salmon]

R1 performs at the level of GPT-o1 and in some cases better. To call it trash would be incorrect lol.

[prisma crown]

Uno reverse: build an ai bot that talks to customer support on my behalf

[unborn flicker]

nope, ive tested them both. R1 is slow as hell and sometimes changes the whole content or context

o1 is fast, always provides full responses and is more accurate

but r1 is great yeah

if o1 is 10/10, r1 is 8-9/10

[acoustic salmon]

8-9/10 is very good lol

[unborn flicker]

yep..

minus the slow responses because of traffic

[acoustic salmon]

so trash would be incorrect

it definitely takes a while for it to process its reasoning

[unborn flicker]

yeah that was an overstatement

[acoustic salmon]

I do prefer o1 but I think it's very interesting how R1 processes

like it sits there and talks to itself

[empty lotus]

You work at Google really.
Not sure why losing image is an issue when you can keep a copy of original image before you pass it.

[wheat elbow]

Yes, this is already a task given to me by google devs and I have to find it lol

[solemn ferry]

hi guys quick question: has ChatGPT-o1 been removed from Perplexity Pro? It was on a few days ago for me but it isn't there in the drop down anymore for me

[unborn flicker]

ocr + add the translated text to the place where the previous text was, python

[unborn flicker]

i just used it..

[wheat elbow]

We are working on a new project, I told my PM what you said...

[solemn ferry]

hmm is it a weekly limit or a daily limit on Perplexity ?

[unborn flicker]

Daily.

[wheat elbow]

I did this, the translations on the image are either moving out of the box in the new img But the AI should be able to adjust it.

[acoustic salmon]

o1 isn't in the model selection drop down anymore, it's in the Pro Search Reasoning dropdown if that's what you mean

[solemn ferry]

oh i see it now! it isnt in the settings any more its now on the pro slider

[hasty basin]

What's the best AI model for science based assignments

[solemn ferry]

ye thanks dude

[solemn ferry]

just saw it 🙂

[acoustic salmon]

np, stuff keeps changing with the new rollout

[solemn ferry]

ooo and r1 is on there yay

[unborn flicker]

Cool

[unborn flicker]

Not that experienced yet, sorry.

brainstorm using AI

[wheat zodiac]

Yesterday I saw a limit of 10 for o1 and r1, today I don't show that on my drop down did something change?

[halcyon glacier]

Imo Claude is still better than DeepSeek, DeepSeek codes very badly sometimes and stops to write sometimes. It leaves you with a script half written.

Deepseek is overrated

And it's slooooowww

[echo swift]

Maybe it will show if the remaining goes down to 10

[halcyon glacier]

oh my god

[frail storm]

No

[halcyon glacier]

Yes

[frail storm]

L opinion

[halcyon glacier]

you're new, don't say that

[frail storm]

Ratio

[halcyon glacier]

I have been using Claude for 1 year

[unborn flicker]

maybe because of the current traffic

[halcyon glacier]

DeepSeek is so bad

for 3 specifics reasons

Code synthaxe looks like a beginner

[unborn flicker]

yeah, and deepseek is weeks old

[halcyon glacier]

It writes slower

[frail storm]

Hey, but is pretty good at maths tho, and I'm saying it as a maths major

[halcyon glacier]

And forget to finish their sentence sometimes

[rustic harness]

What's the best way to know if Perplexity can access/scrape a link?

[rocky ginkgo]

yess

[frail storm]

But not as good as perplexity imho

[halcyon glacier]

Claude never cuts while writing

thats a good boi

[rocky ginkgo]

cuz perplexity directly attribute the source??

[frail storm]

Yep

[rocky ginkgo]

yeah agreed

[frail storm]

It's pro search is pretty cool, it even solves problems from our textbooks

[rocky ginkgo]

dayum, I only have the free version

[frail storm]

But sometimes it still does that AI thingy where it just bluffs out related terms that it finds on the net

[unborn flicker]

I know you seen that

[rocky ginkgo]

still does the job

[rocky ginkgo]

hmm... chatgpt also sometimes do that ngl

like when I change my prompt, it still gives answers relating to my previous prompt or something similar to that

[frail storm]

Chatgpt is badddd at math

Like bad bad

Which makes sense given its just a lang model

[barren saddle]

what happend to the other models lol

[solemn ferry]

its on the slider now

[earnest oasis]

Hey so im tryna use perplexity assistant and it just doesnt work on my device? it's really laggy and it takes a while for the audio recognition to start. im using a samsung galaxy a10e. is it just that my phone hardware is bad or is there an issue with the app?

[tame current]

what the limit for R1?

[earnest oasis]

10 messages, just like o1

[tame current]

😭

Sad very sad

anyways

[earnest oasis]

yeah honestly i feel like it should be more tbh

but idk lol there's a lot of factors

[tepid schooner]

They said it will increase daily but yeah pretty limited right now

[earnest oasis]

hype

[iron junco]

[1326655467577147412](#📢│announcements message)

I cant get it to work? Tried the MacOS App

[vapid onyx]

I know Alex mentioned mobile apps will be updated. Hope it’s sooner rather than later

[green crystal]

Why can't I choose r1 for pro search in the mobile app?

It's only visible on desktop

[tawdry egret]

I have a code for perplexity pro but I can´t use it, anyone can help me, please?

[serene sand]

please contact [e-mail]

[serene sand]

#🍏│feedback-ios message

[green crystal]

I forgot to mention I'm on Android
I guess the update is in the works for mobile

[weary crescent]

How can I use deepseek with pplx?

[earnest oasis]

click on the dropdown icon next to pro search

the options for o1 and deepseek will be there

[weary crescent]

of course when I go to the main page it shows a tooltip smh

there's no way to set as default though?

[earnest oasis]

lmao

[earnest oasis]

i guess not

theres only a limit of 10 rpd so it wouldnt make sense to set as default

[weary crescent]

oh what the hell

[earnest oasis]

ceo says he wants to update the limits

[weary crescent]

might be time for me to selfhost perplexica

[earnest oasis]

lmaooo

[ember wasp]

how can you opt in for the web beta access?

[earnest oasis]

the web version is already open, what specifically are you talking about?

[elfin ferry]

theres nothing there afaik
last time i remember there was something in there (and the last time the opened beta?) was when pages were coming out

[ember wasp]

i was just wondering why some users have the web beta tester role

[elfin ferry]

yeah thats from pages

[ember wasp]

pages?

[elfin ferry]

before it came out they tested it in beta

they sent a google forms were you could request to join the beta

[ember wasp]

ah i see, thx

[sharp tendon]

Can anyone please explain to me where Deepseek R1 is? I purchased the pro version of perplexity after reading this article becasue it had deepseek R1 but I do not see it in the settings https://www.searchenginejournal.com/perplexity-ai-deploys-chinese-deepseek-ai-model/538452/?utm_source=chatgpt.com

[tardy swan]

U need to press the little arrow on the text box when using pro search

[sharp tendon]

thank you

[cedar latch]

How do I get deepseek on ios

[sharp tendon]

can I replace my chatgpt pro suscription or will I be missing anything?

[cedar latch]

Delete that

I’m gonna cancel mine too now

[sharp tendon]

any plans for operator?

[ember wasp]

you are missing the custom GPTs, but if you are not using them, you don't lose anything ig

[sharp tendon]

what about if I want to use more than 100 chats with deepseek R1, it looks like there is a limit..

[ember wasp]

idk, but R1 has also nothing to do with your chatgpt subscription

[green crystal]

Hope so because I was enjoying search being so good I quickly ran into the limit

[sharp tendon]

I know that. I am talking about limits within Perplexity

[lapis briar]

Are there any code specific spaces for perplexity?

[ember wasp]

currently, you can do 15 R1 requests per day, but the limit will increase this limit soon if the first R1 hype wave is over

[sharp tendon]

can I pay to have this limit increased?

[green crystal]

Gonna take awhile since main deepseek site is still getting hammered

[ember wasp]

currently not

[green crystal]

It's how I ended up subscribing

[ember wasp]

[hollow pumice]

The memory imo should be mentioned especially if you use it at all for conversational/personal purposes or advice.

[strange walrus]

Noob Pro user here. How do I get access to Factset and Crunchbase data?

[green crystal]

I'm curious if numbers on how much cheaper deepseek r1 is to host vs O1 get posted

[acoustic salmon]

you'll be missing Tasks and Operator

Perplexity has a working version of what tasks does but it's not fully fleshed out

Canvas is kinda Pages

and Pages is a very strong feature imo

[ember wasp]

what are pages?

[acoustic salmon]

Pages is a section where you can write out prompts and it essentially makes an standalone page for you. Like this
https://www.perplexity.ai/page/packers-latest-news-updates-1DhGTE8nRICBvzT_4XuXCQ

You can edit the text directly. It's not a 1:1 match to Canvas, but I've used it like it was lol.

[ember wasp]

ah crazy, thx

[acoustic salmon]

Spaces is kind of a mix between Custom GPTs and Projects

[harsh verge]

It's for Enterprise Pro customers.

[magic talon]

Rather use perplexity itself

[cunning cave]

which version of deepseek-R1 did perplexity deploy/use ? distilled models? or the 671B model ?

[hazy moth]

Is Deepseek R1 unlimited now? I noticed the 10 message limit message is gone? I've been using the deepseek website lately as it allows tons of free usage, but if it's unlimited on Perplexity, perhaps I don't need to use the Deepseek website anymore.

[earnest oasis]

oooh hype

[harsh verge]

You can track your usage and quota here: https://www.perplexity.ai/rest/user/settings

[hazy moth]

So only 15 then?

[elfin ferry]

for now, they're planning on increasing it

[hazy moth]

Okay. Hopefully soon, since Deepseek offers 50 for free.

The downside to the Deepseek website is that it has a moderation bot that may delete messages critical of the CCP

The moderation bot isn't strict, but it's still better to not have it

Hmm... it seems to skip reasoning when asked about April 15, 1989 when using Perplexity's R1 model, whereas the official Deepseek's website does do the reasoning but ends up getting cut off by the other moderation bot. I wonder if Perplexity is using a distilled version of R1?

Distilled versions of R1 have more censorship than the full model. The distills tend to skip the reasoning for these types of questions, while the full model does not.

[supple isle]

I take it DeepSeek and reasoning is web only right now? Any plans for the iOS app?

[real wagon]

guys have you noticed any issues regarding the ai models? for me i lost most of the models and the i have only 4 options that appear for me

[hazy moth]

Here's what happens when you ask that question on the Deepseek app

Yes, it is well aware of that event

[acoustic salmon]

Perplexity is running R1 on their servers in the US

[hazy moth]

I know, but is it running the distilled version or full version?

The distilled versions are less compute heavy, but more censored

[acoustic salmon]

R1 is not censored in the US

well, on Perplexity

[hazy moth]

So why does it skip reasoning when I ask about April 15 1989?

[acoustic salmon]

it doesn't afaik

[hazy moth]

Switch it to writing mode to talk directly to the model

[acoustic salmon]

same result

[mental lake]

what does that mean?

query_count":1086,"query_count_copilot":939,"query_count_mobile":124

this is per month, right?

[acoustic salmon]

per day

the 15 that is

[hazy moth]

Oh nevermind, it may have been a glitch.

[acoustic salmon]

definitely possible, it's been through various changes since rollout

[hazy moth]

I was on my laptop before, but on my phone it's now working

[mental lake]

I do not get it, how can I have 1086 and 939 per day when i only have 600 available?

[acoustic salmon]

may take some time to be consistent

[acoustic salmon]

sorry I meant your 15 limit

[hazy moth]

Maybe I didn't re select r1, since it turns off on it's own

[acoustic salmon]

I haven't had it for a month yet to determine what the query_count adds up

[acoustic salmon]

I run into this sometimes, too. Doesn't seem to persist in choices sometimes

[hazy moth]

Lmao, it has a thought to avoid mentioning the crackdown

[acoustic salmon]

I like going through its reasoning, it's interesting

[hazy moth]

It did in fact mention the crackdown in the end

[acoustic salmon]

First o3 benchmark I've seen

[ember wasp]

@halcyon coral

[hazy moth]

I'm more interested in o3 mini. I heard o3 is hella expensive to run

I heard o3 brute forces its quality like crazy

[craggy vine]

Wow, R1 day on Perplexity, kind of nuts. I've been using this model for a almost two weeks now, from their site, it's crazy how it's exploded.

[gusty wasp]

R1 for non-reasoning tasks?

[hazy moth]

Uncensored r1 from the looks of it too

https://www.reddit.com/r/LocalLLaMA/comments/1i8rujw/notes_on_deepseek_r1_just_how_good_it_is_compared/

Apparently, it's the best NSFW roleplay AI to exist

So while OpenAI is trying to control what its model says, Deepseek gave everyone a frontier model that is truly unbias and gives absolute freedom

[spare comet]

Can we pleeeeease get past this Chinese censorship - we get it

[earnest oasis]

Oh lord the fanfic subreddits are gonna have a field day with this one

[small tree]

just dont criticize the CCP then

simple

lmaooooooooooooo

[floral parcel]

Is perplexity pro getting more queries for the new DeepThink R1 model? I didn't even realize i used them all up today while I was playing around

[gilded finch]

What’s the model do that’s so innovative

[gilded finch]

Yes, it was at 600 last night

[floral parcel]

Really? It stopped for me after like 15

[ocean gorge]

Any idea why I have pro but dont see r1 in my settings?

[gilded finch]

There’s a drop down menu

[floral parcel]

is it still being rolled out to all accounts?

[gilded finch]

Queries keep changing because r1 is getting a lot of use

[floral parcel]

so it's not 600 like i saw

[dense kelp]

Is it 8 remaining on both R1 and o1 or 8 per each?

[lyric jasper]

I asked Perplexity to do some processing on a 526-line HTML file, and its initial go was impressive. I did some tuning on it and then saw it was missing the last two sections, and despite repeated questions, it was making up a little data and not using mine. So I signed up for Pro. The setup looked for a model, so having asked perplexity what was best for HTML/CSS, I set it to Claude 3.5 Sonnet. I queried again, and it was worse; now it will not even do anything on the file. What is going on? How can I get it to do all I ask?

[floral parcel]

it was both for me lol

[hazy moth]

8 total it seems.

[floral parcel]

not alot

[dense kelp]

that's a shame that its limited so much per day for pro

[hazy moth]

Yeah, also, r1 costs a lot less, so it's odd for it to share a limit with o1

[supple isle]

No R1 in app yet tho right?

[median summit]

Yep, it costs 25x less. It should have at least a limit 250 queries then.

[hazy moth]

It only uses 37b active parameters

At scale, it should cost very little per token

[median summit]

$2.19 per million output tokens vs $60 per million output tokens (for o1)

[shut brook]

anyone else notice DeepSeek (on its native website) will at a certain point show "Thinking..." and then 'server busy' and will not do anything further, but you can start a new thread and it'll immediately churn something out?

[hazy moth]

Claude Sonnet is $15 per million output tokens, which is far more than r1

And Claude Sonnet is unlimited

[median summit]

yeah, I think they will soon set that to be unlimited too, since there is no financial reason to maintain up to 10 queries

[hazy moth]

Even if you assume a 4x increase due to the thinking tokens, it's still less than Claude

[hazy moth]

Probably they have to scale up their infrastructure first to host their own r1, given that they can't use the API

So they don't have the scale to give unlimited yet

[floral parcel]

that makes sense, from my limited use, the source gathering from perplexity with the deepthink R1 is such a good combo

i been using it to help me study for my PMP exam lol

[cloud marsh]

I heard R1 cot + claude 3.5 produce better answer. why isnt perplexity using that combo

[gilded finch]

15 in total

[gilded finch]

How would you use both of these ais combined

[floral parcel]

my question

[hazy moth]

I've mostly been using the Deepseek app due to the less limited use. While it does censor blatant criticism of the CCP, it's surprisingly unbiased when discussing China politics

It even mentioned "Uyghur internment" without getting deleted

[cloud marsh]

https://aider.chat/2025/01/24/r1-sonnet.html according to this they use R1 thinking output feed into sonnet

[hazy moth]

I think the fact that it discusses such topics openly and critically from both sides earned it my trust

The Deepseek app appears to have 2 layers of moderation. One layer checks the output in realtime to find blatant cases, and the other layer checks it more thoroughly after it's done writing. You can bypass the second filter by turning on airplane mode as soon as it finishes its response

[cloud marsh]

if enough people post this on social media it will be fixed

[hazy moth]

Fair. I guess enjoy it while it lasts.

And by then, there will be more Western providers to use it from

I think Deepseek r1 Zero might be even more uncensored, since there was even less human involvement in the model training

But I'm not sure who's hosting that one as a serverless api

It seems the t/s on non-Deepseek platforms is trash rn

https://artificialanalysis.ai/models/deepseek-r1/providers?utm_source=perplexity source

[hazy moth]

Also, the official Deepseek discord has these images in it. I doubt they care about controlling things that much.

[spare comet]

Wake me up when everyone stops talking about T Square

Massive brainwash happening across the world, it’s quite hypocritical

Also this DDOS only makes them look better and us worse

Talk about supression…..

[cloud marsh]

They probably don't care. But if this caught the attention of the Chinese government's censorship department, they would give DeepSeek a call and ask them to censor it.

[hazy moth]

Possibly, they thought Discord would take care of it for them

They don't realize that Discord doesn't come with free moderation

Since the platforms they're used to take care of that automatically

Also, time zone differences

Though I think their discord is a lot cleaner now than it was before. It was being flooded with spam

Though that's a very very low bar, as it was really really bad earlier

[olive ermine]

o1 also doesn't seem to be in the app (on android) nor is the new 'Recency' parameter you can set

would love to see the app get the pro + reasoning modification

[past basin]

i lost o1 and no r1. weird

[worldly kestrel]

Hey is there any example of sonar new updated api use case for research marketing assistant ?

[night falcon]

i miss sonar huge, it was so reliable when i didn't make it assume details about my queries

bring it back as an option

pls

[exotic flame]

How can I use reasoning mode in app? Or it can be used in website only?

[olive wave]

hello just watched a video on how to use it! Looks pretty awesome !

[unique umbra]

Deepseek heh

[cedar latch]

i dont see deepseek in my app

help

[cedar latch]

they definetly care, the ccp controls deepseek i dont think theyll let it be like this for long

[empty lotus]

is sonet 3.5 (new) or sonet 3.5 used.
claude website shows they released a better one

[open dune]

Is deep seek available on the app yet?

[empty lotus]

now only 100 Pro search perday ?

[torpid marten]

Anyone knows how to open incognito in web interface without using keyboard shortcuts?

[short narwhal]

Not on iOS yet at least, don’t know about android

[whole tree]

are they bringing back O1 model?

[solemn trench]

is it possible to use deepseek offline via sonar? in other words, i dont want to pay the bravesearch call

in fact one of my applications is stock classification, so i would prefer not an online model

[worldly kestrel]

What perplexity playground does it use sonar? Can you give it research questions and it goes and fetches ?

I swear it worked for me on a data set of 10 . Then next time I tried it it said it can’t do search . I’m so confused. Is anyone on their team in this discord ?

They api route burned like 22cent. I don’t want to test it with fees like that …

[plain nacelle]

What R1 parameter model is this using?

[solemn trench]

is the API down?

[harsh verge]

It works perfectly on my end. (API)

[plain nacelle]

Seems the app doesn't have R1 logic but desktop does, am I missing something?

[serene sand]

https://x.com/aravsrinivas/status/1884492340657606896?s=46

Aravind Srinivas (@AravSrinivas) on X

The daily limit has been increased from 10 to 25 DeepSeek R1 queries a day for Perplexity Pro users. Goal is to keep increasing this as we add more capacity! Enjoy.

X · Jan 29, 2025, 6:42 AM

[naive roost]

PS re DeepSeek.
https://x.com/johncoogan/status/1884127587292832005

[fair pelican]

I've seen the tweet on using DeepSeek R1 + FactSet or Crunchbase - but I cannot see it yet on my options even if I am a pro user. Anyone?

[wise mirage]

does anyone know, where i can redeem codes? do i have to contact customer support?

[graceful iris]

I am from India and I see O1 model is removed from the list of models?

[harsh verge]

It's for Enterprise Pro customers.

[alpine flicker]

hover over the Pro R1 near the checkbox and click

I was confused too

Is it me or did complexity addon got worse after recent updates?

Ahhh nevermind I do see that after update options did reset and it is necessary to turn then on once more

[elfin ferry]

nice! they upped the limit from 15 to 25

[mint hatch]

They didn't return the previous o1 though

Still gloomy for me

[elfin ferry]

also for o1 since they're sharing the same "pro_reasoning_limit" i believe

[elfin ferry]

what previous o1

[serene sand]

https://x.com/aravsrinivas/status/1884563740575551915?s=61

Aravind Srinivas (@AravSrinivas) on X

Well, even if you don’t care about your data going to China, I think it’s worth caring about not using a censored model that the DeepSeek app serves. And that’s why it’s worth using the R1 model on Perplexity with web search grounding and reasoning traces.

X · Jan 29, 2025, 11:26 AM

[inland stump]

Thanks for all that info. I'm curious about DeepSeek usage pricing markup. I think it's ~5x on Together.ai. You say "effectively free." Do you know the exact pricing for using DeepSeek on Perplexity (and how much it's marked up over direct usage of DeepSeek)?

[narrow tide]

Nvm, now it stopped doing it

[naive roost]

By "effectively free" I meant for consumers using it for queries at like 5x/day. At that rate the tokens are trivial. As in you can't possibly spend $1/mo.

[mint hatch]

Previous o1 before this newest update. O1 this update is much different in story writing, much much different how hard I tried to make it work the same.

[stone axle]

Can someone explain, how is DeepSeekR1 used for reasoning, while I have Claude selected as the model for example?

[naive roost]

But yeah the effective token markup for consumers using the genpop app / website vs. API / playground is absurd.

• Even more so with Google AI Studio vs. Gemini Advanced or any other Google consumer option.
• AI Studio tokens are acbtually free and unlimited there + web search + 66k token output (flash thinking) + stream realtime (AI interactive video of computer screen or outside world via camera). And PS you can use with desktop or mobile browser via website.

[young niche]

being a pro user why cant i use deepseek unlimited

[naive roost]

Great question.

• I also use 3.5 Sonnet selected as my default model.
• Pluis I have some detailed custom instructions input to the profile boxes (where they work perfectly btw).
• Seems like when you click the blue "Pro Search" toggle button , which then reveals the options to add r1 / o1 reasoning. And then chose either r1 or o1,
• The r1 / o1 replaces 3.5 Sonnet as the brains, and your profile / custom instructions are ignored.
• Ignoring the custom profile instructions is actually a bonus because you want the minimum possible effective instructions when using thinking models.

[naive roost]

Take 30 seconds to get an API key and then use it via the Playground interface. Unlimited usage and effectively free if you choose the "sonar-reasoning" model.

[stone axle]

Interesting, I have custom instructions in a "space" and R1 seems to follow it just fine

[naive roost]

How do you know?

• With my custom instructions, I tell it to use "thinking" XML tags to start and a confidence score to end. Those don't appear with o1 / r1 reasoning but do appear with every other model, and I don't toggle them off for the reasoning models.

[stone axle]

I have custom instructions that ensure I receive answers in a very specific format that's slightly modified but meant to imitate the "UniversalPrimer" GPT available on ChatGPT. The format in which it provides me is unique and very specific to that space. Plus, I can see where it reasons and decided to follow my instructions.

[stone axle]

Here's another chat example. I can even see where it debates on whether to follow my instructions or not when I asked it "Is this chat using Claude 3.5 or R1?" Funny because the reasoning model is making a good point about whether or not the format is appropriate with the given question but it still followed it anyways.

[naive roost]

Holy cow - you are right.

• I didn't realize you had to click on the "researching" section to display the thinking part.
  But my ending "confidence score" is there, so it absoultely must be reading the profile stuff.
• Am kinda wondering if I should remove that when using r1/o1, when tbh I use r1/o1 for every Perplexity query now. Bc that's the only way it matches DeepSeek-R1 via their website (reasoning + web search) and AI Studio Gemini Experimental 1206 (second-best lmsys model + web search).
• That said, the results I get from Perplexity + r1 with the instructions included are fantastic, so maybe don't fix what ain't broke?
  https://www.perplexity.ai/search/research-and-e-reference-text-EQ7yIJFhQ8.V3l0Zz0qPlw

Sry - wrong link. Perplexity keeps trying to shift me back to non-reasoning pro search if I don't always make sure it's toggled on. Here's the reasoning version.
https://www.perplexity.ai/search/reference-text-regarding-the-t-bwmmT1aDSzGbK1ey1eRG_w

[keen lodge]

I'm sure this has been asked to death so I apologize, but as a Pro user, how do I use the DeepSeek R1 AI model?

[serene sand]

#👑│pro-lounge message

[brazen rapids]

is that perplexity?

[elfin ferry]

with the complexity extension

https://www.cplx.app/

[brazen rapids]

thx

[tight harbor]

can i swear

[foggy tapir]

But O1 and haiku are removed from perplexity right?

[brazen delta]

Yo dudes! How to toggle r1 on pro perplexity app on IOS?

[elfin ferry]

they're only removed from the native model list, they're still there though
complexity doesnt use models that arent in perplexity at that time

oh o1 isnt removed even natively idk what you mean

[elfin ferry]

@harsh stag can explain it

[foggy tapir]

Atleast on web and android it's not visible.

[harsh stag]

o1 is available via the pro toggle ^

[foggy tapir]

I see another user stating the same.

[foggy tapir]

Ooooh

Thanks to both of you.

[harsh stag]

or you can install and use the extension, more streamlined and intuitive ui for switching models

[foggy tapir]

Yes trying it out 😁

[weary crescent]

which extension?

[harsh stag]

#💬│general message

[weary crescent]

oh yeah, complexity is awesome

[worldly dirge]

Hey folks, Toven from OpenRouter - When Perplexity says powered by R1, do we mean the full 700b param model or one of the distillations? Thanks in advance!

[livid aurora]

How can we use the three Sonar Models in writing and web search mode?

[livid aurora]

People dont know yet

[rose musk]

Context : I just got access to Perplexity Pro after it was made free for many Campuses in India : https://www.linkedin.com/posts/pbshi_seeing-high-demand-on-the-free-perplexity-activity-7288610004033581057-JsbB/

Where do I find how many "pro prompt" credits have I exhausted for a given day ? ( Because I was not able to find them as mentioned on the site ) ?

[buoyant wagon]

"you want the minimum possible effective instructions when using thinking models", curious as to why?

[livid aurora]

https://www.perplexity.ai/rest/user/settings

[lusty sentinel]

r1

[worldly dirge]

https://x.com/denisyarats/status/1883763037875978669

Denis Yarats (@denisyarats) on X

@D_Twitt3r yup, this is a true R1, coming out soon

X · Jan 27, 2025, 6:24 AM

[rose musk]

Thanks !

[cursive jacinth]

i dont see limits on r1 sir

what is the limit now sir?

[icy cloak]

daily query limits from 10 to 25 for Pro users

https://www.businesstoday.in/technology/news/story/perplexity-ai-increases-limit-for-daily-deepseek-r1-queries-on-its-platform-462421-2025-01-29

[magic leaf]

Hey, can i know limit of o1 r1 and pro search?

[minor shard]

25 daily

[hallow tulip]

I'm fine with 25 (for now) though 30 would be awesome

[half owl]

r1 answer is not good

but the reasoning is good

the result is bad

[summer meadow]

Does anyone know how Pro search works? For example, if I have a space where I selected the GPT-4o model, but when asking my question, I choose Pro R1, does the response include the reasoning from R1 while the answer itself comes from GPT-4o? Or is the entire response generated solely by R1?

[wheat dagger]

It's because the answer is provided by the writing models (gpt 4o , sonnet ....) , R1 only helps in reasoning , even if the reasoning is good you won't get the result you perceive due to the weakness of writing models
It's a similar effect to what's called bottle neck in gaming
That's why O1 was better in perplexity when it was an actual writing model instead of being limited to reasoning only (probably they had done this to lower costs and increase the ammount of tries , yet the consequence was the efficiency).

[summer meadow]

Is it really like that? only the reasoning part is done by those models, and the writing part is the model of our choices?

[half owl]

I don't think so, we can't change the writing model

[summer meadow]

you can have a space with model selected like claude 3.5, and then use pro r1. But im not sure if the writing is done by claude then or just solely by r1

[nimble slate]

What model does “Pro search”, at the top of the drop down menu use? There’s no specifications for it. How does it compare to the other models?

[serene sand]

https://x.com/aravsrinivas/status/1884509590684934211?s=46

Aravind Srinivas (@AravSrinivas) on X

@julheetel8890 Full.

X · Jan 29, 2025, 7:51 AM

[worldly dirge]

Thanks!

[rain lodge]

Oh heck yes r1 reasoning now in android that was fast

I very rarely need to reprompt for complex questions now

[livid aurora]

https://x.com/testingcatalog/status/1884652009938178115

TestingCatalog News 🗞 (@testingcatalog) on X

BREAKING 🚨: Grok 3 will support reasoning! It will be able to expose its "thinking" process to the UI as well 👀

X · Jan 29, 2025, 5:17 PM

[lethal cloud]

Hi

[dawn badge]

Ello peoples

[odd osprey]

I've requested access to the beta for "return_images" in the api response. Anyone else have access?

[astral nymph]

perplexity has some kind of different privacy policy compared to other chatgpt companies ?

[dawn badge]

I honestly had no idea that an AI company had a discord

[peak mantle]

I want to ask that which is best pro search or pro search with R1? please answer if anyone knows about this?

[dense kelp]

has anyone used qwen so far? I found it interesting but seems to be lagging, cosndiering they claimed they were better than deepseek

[fast meteor]

🤷‍♂️ you can read yourself https://www.perplexity.ai/hub/legal/privacy-policy

[ivory mantle]

Wow

[empty lotus]

deepseek effect

[weary crescent]

o1 effect

[floral parcel]

Which model does the AI ASSISTANT on android use?

if it's default then i'm just going to keep using app only lol

[torn moon]

U can choose

Anyone knows if there is possibility to activate perplexity assistant with voice? Something as hey google for google's?

[floral parcel]

how do I choose? Does it just take the one i selected in settings of the app?

[torn moon]

Activate assistant and u have settings button, there is assistant model optoon within

[floral parcel]

ah i see ok

these are the only options

[earnest oasis]

a lot of em do it seems, groq, perplexity, anthropic, openai

its good for getting a good community and fanbase

[jaunty raven]

Hello

[astral nymph]

I asked chat, openai and anthropic are more privacy friendly

[naive roost]

https://x.com/itsPaulAi/status/1870892042769039598

Paul Couvert (@itsPaulAi) on X

Reasoning models are becoming the norm

So a friendly reminder that OpenAI has released a guide to writing prompts for them.

Few important points and example:

X · Dec 22, 2024, 5:59 PM

[woeful widget]

Hey

[naive roost]

This 1hr course by OpenAI’s Colin Jarvis covers reasoning models EXTREMELY well. Highly recommended. https://www.deeplearning.ai/short-courses/reasoning-with-o1/

[small tree]

why on earth would they do that?

r1 on deepseek provides very good answers after the reasoning lol

is it because it requires less compute?

[naive roost]

Click on the Pro Search toggle button, a hidden menu pops up.

[tame current]

Is o3 launching tomorrow?

[small tree]

yes

will perplexity get o3?

@indianceo

Aravind

[small tree]

Sams probs buying a bunch of koenigseggs

with the $500bill

[naive roost]

Note that the sole head of the Department of Government Efficiency just dropped $4bn on a 100k GPU supercluster and got promptly smoked by a handful of Chinese with $6mn.

[still needle]

I miss my Perplexity 😭

[tame current]

Why what happened?

[small tree]

his 1 month trial ran out

[dawn badge]

Night chat

[tame current]

Damn sounds heavy

[tame current]

As a good gesture here is a good luck 🪨 for you 🙂

[small tree]

🪨

[naive roost]

Use Perplexity Playground site with API key. It's dirt cheap, borderline free for casual like 5x/day. Very tough to spend even $1 if you're just doing standard queries. You get the API from the account page and then start using the models like "sonar-reasoning" which is Pro Web Search with Deepseek-r1. And usage is unlimited. https://labs.perplexity.ai/ Search my comments on this site from the last 24 hours.

[left stream]

They increased the number?

[ionic stone]

I'm trying to understand what the API key is needed here. I can use labs for free right now on the playground already so why do I need an API key?

[sleek topaz]

what parameter size r1 is perplexity using? I dont think its the full 671B one?

[ionic stone]

#💬│general message
"Full"

[serene sand]

https://x.com/aravsrinivas/status/1884708550795985382?s=61

Aravind Srinivas (@AravSrinivas) on X

Number of DeepSeek R1 daily queries on Perplexity for Pro users has been increased to 50 a day, free to 5 a day. More updates coming shortly. Enjoy!

X · Jan 29, 2025, 9:01 PM

[sleek topaz]

ty

[tribal skiff]

Can someone please help me figure out the new API call, I followed the docs to a T and I cant get it working. Python

[stoic crater]

Is there a limit for how much you can use R1 a day within Perplexity if you have a pro subscription?

[rich bane]

yes, 100 at the moment

[stoic crater]

Not bad at all actually!

[serene sand]

https://x.com/aravsrinivas/status/1884718399797821642?s=61

𝒱𝑜𝓍 𝒫𝑜𝓅𝓊𝓁𝒾, 𝒱𝑜𝓍 𝒟𝑒𝒾

Aravind Srinivas (@AravSrinivas) on X

Should Perplexity make the default model DeepSeek R1?

X · Jan 29, 2025, 9:40 PM

[grim raven]

This might be bundled with the android assistant feature but does anyone know when web/iOS can expect the “agent”/cron functionality? Looked around a bit and wasn’t finding anything since the initial posts about it.

[ionic stone]

wow, it seems PP is going all in on DeepSeek R1

[small tree]

Yea, it's the only way it'll survive

for now that is

the only way to survive is to adapt

[stoic crater]

I personally find myself using perplexity way more than chatgpt

Only been using chatgpt for the advanced voice mode

[small tree]

Same, now that it has R1

[stoic crater]

It's cool but I would still use the regular pro mode since it takes more time for it generate a response. for decision making questions at least...

[thorn osprey]

i use perplexity as my default search engine and not all my searches require R1 (sometimes I just want to navigate or get an instant answer, like what you get from google search preview)

[stoic crater]

Exactly what I said there

[small tree]

ofc, it's cheap to run

cheaper than all the other models

lmao

[thorn osprey]

it would be awesome if perplexity could know when more thinking is required

[acoustic salmon]

cheap is great but performance is key

[small tree]

they can increase performance, just limit its thinking tokens

[acoustic salmon]

R1 doesn't perform any better, it's just that it performs almost as well as o1, which is expensive

but o3 is already hitting some benchmarks that are well above

[cursive jacinth]

is o3 mini available now in chatgpt sir?

[small tree]

tomorrow

[harsh verge]

You don't need the API to use Perplexity Labs' Playground—it's free...The platform allows users to experiment without any cost

[cursive jacinth]

ok sir

[acoustic salmon]

I have GPT Plus still for a bit, so I'll be testing o3 as it drops

but one thing about GPT, is even o1 doesn't have web search and such yet

so o1 on Perplexity with their addition of web and other functions is pretty nice, even if it's limited

[stoic crater]

For me, OpenAI gives us a few hours a day of advanced voice mode and I'm good 😭

For plus users

I don't care about o3 or o3 mini

[acoustic salmon]

o3 gonna be expensive I know that

[tepid sequoia]

people i don't understand here everyone is either rejoicing or crying, how great Perplexity gave you thinking but the answers are maybe a little better, but still concise it just takes longer 😄 ... like yeah the chat.deepseek.com page is also limited and sometimes throws censorship on something, but that's what all these services do, but try Deepseek how it answers purely through the API when it's not censored those are different answers, and the search on Perplexity is terrible for some things, look at this

[stoic crater]

And then a year later R2 will drop and be just as good 😂

[acoustic salmon]

and Claude 4 will be out

and Gemini 2.5

[stoic crater]

I mean it will be way cheaper

[acoustic salmon]

it's a race

meh, cheaper. It really depends on your wants and needs

[stoic crater]

I'm only happy about it because now we'll see more features and better models reaching more people and becoming more widely available...

If things actually gets cheaper

[lethal cargo]

hello guys

i have qusitions about perplexity pro plan can someone help me please

[serene sand]

https://x.com/aravsrinivas/status/1884727961263214672?s=61

Aravind Srinivas (@AravSrinivas) on X

100 daily DeepSeek R1 queries now for Perplexity Pro users!

X · Jan 29, 2025, 10:18 PM

[minor shard]

bout TIME

[stoic crater]

But if it's that cheap, couldn't they just extend it to even 300 or more?

But I never even reach that number of queries anyway

[lethal cargo]

and what about chatgpt models and claude limits ?

[minor shard]

They just dont want "abuse". Idk why they are still experiencing issues with abuse that they need limits like this

[stoic crater]

Really? 🤔

320K$

Used pro search with R1 to generate it by the way

[orchid blade]

Anyone know when r1 is made available in iOS app?

[silk grotto]

how perplexity labs are hiding all the data from the networks tab, i can't see their api calls, any web dev ?

[orchid blade]

How do you set the pro model to use r1 tho

[lone valve]

I have the latest version but it doesn’t work

[kind relic]

I would guess they're using sockets instead of individual requests and then keeping the connection open

[silk grotto]

yeah, I saw some sockets requests during the start up of the website

cool way to hide the api calls 🤔

[orchid blade]

This is all I see, how did you get the drawer to open with the options?

[tepid sequoia]

lol refresh page 😄 😄 😄

[orchid blade]

Ahh wait, there was an app update my bad

[serene sand]

update your app

[lethal cargo]

guys how many queries i can use in all the availble models in pro plan ._,

[silk grotto]

No, I mean all the api calls are made during the startup , when you are interacting with the llm, you cant see anything.

[tepid sequoia]

Oh my God, open a blank page, turn on inspect, and then load Perplexity.

[tribal skiff]

I pay for chatgpt pro, 280 CAD per month, but totally worth it

[serene sand]

https://www.perplexity.ai/help-center/en/articles/10352901-what-is-perplexity-pro

[tribal skiff]

Seriously tho can someone please help me figure out how to call the new API, I've been trying for days with no luck

[hazy moth]

100 R1 calls per day in Perplexity is a true game changer. I've been using R1 in Perplexity all day today and still have plenty of requests left

[hallow tulip]

Damn 100 is crazy time I'm thoroughly impressed by the change now

You can already imagine that they are adding in o3-mini tomorrow.

[hazy moth]

Usually, when coding, AI gets worse and worse at making meaningful changes, but R1 is so much better at avoiding that trap. For a really complex task involving tampermonkey reading API calls, I only had to do one code reset and I was rolling

[hallow tulip]

I'm really hoping that the US companies can step it up r1 is really a different sort of model man it thinks in a way that I've never seen from an LLM before its pretty scary

[hazy moth]

Yeah, it helped me bypass CORS restrictions and everything

[jagged epoch]

I have Perplexity Pro. Today logged in on the MacOS app, I searched for "substitute for pandan". I did the same search with the AI model set to Claude and then set to ChatGPT-4o. I expected different results but got the same text, virtually verbatim, on both searches. Shouldn't I get unique search results when using different models?

[hazy moth]

Actually, I just checked with R1, and the script actually does not bypass CORS restrictions. It just passively "observes existing API calls rather than creating new ones that would require CORS negotiation."

[cerulean turret]

Is perplexity Stock research live?

[visual olive]

Hello, is the o1 limits removed?
Also I'm trying to use the android assistant. Can I call it like I'm calling Google?
Like "hey Google"

[hazy moth]

Hmm. I guess they are. I thought 100 only applied to R1. I'd still advise that you use R1 as it's better and more efficient than o1.

[hexed shell]

Why Deepseek-R1 is pro only? It is free because of MIT lincese

[visual olive]

is it? will do, thanks

[hazy moth]

It's amazing how fast R1 is on Perplexity too. I think they have the best tokens per second out of all API providers

I wonder if Sonar reasoning is R1 or o1?

[mossy parcel]

How to access R1 on iPhone? I just got the pro subscription but these are the only model options they give me:

[hazy moth]

Safari version

[hazy moth]

Just checked. Sonar-reasoning is indeed based on R1, so if you want to integrate R1 as an API, Perplexity is the way to go

[serene sand]

#💬│general message

[small tree]

1 year?

3 months max

lol

they can just use the same method, use o3 for synthetic data

😂

[hazy moth]

Yeah right. OpenAI doesn't expose it's reasoning process

[naive roost]

Call the DeepSeek API? If so maybe part of them shutting new customer access via the web and app. They got flooded. Instead call the Perplexity API using sonar-reasoning, which gets you r1 with Perplexity pro web search.

[mossy parcel]

Well interesting, this is all I get

[hazy moth]

https://sonar.perplexity.ai/

[tribal skiff]

I've been trying this for literally 3 days

[hazy moth]

Not sure then. I haven't tried it yet

[small tree]

Doesn't matter

[serene sand]

check app version, should be 2.37.2

[naive roost]

Stop using iphone apps to access Perplexity and all others. Use mobile browser set to desktop view if mobile view doesn't give you everything.

[tepid sequoia]

They must have changed something there 😄 because when you turn off web search, it uses pure DeepSeek and it's working pretty well right now, I'm curious how long it will last 😄

[naive roost]

^^^ Listen to my man.

• The whole point of playgrounds was to give developers cheap access to test and improve their prompts, without offering the candy tools you get with consumer versions (voice, code interpreter, canvas, artifacts, etc.)
• That's still the case with Anthropic and OpenAI.
• But with Google especially and now also Perplexity, the playground / studio / labs / console sites are built to entice enterprise developers while avoiding cannibalizing the consumer access cash cows.
• The hurdle for consumers to access FAR better everything at Google AI studio and Perplexity playground is trivial, but very sufficient to stop the flow.
• Everyone on this board is savvy enough to be here, and easily top 0.1% of users in savviness.
• Google and Perplexity have effectively ceded the free candy to the teensy % of consumers that bother to look.
• Take the much higher quality freebies, my friends. We've earned it by following this stuff carefully over the last six months.

[serene sand]

https://tenor.com/view/synths-and-sounds-ecstacy-ecstatic-climax-peak-gif-24235783

[tepid sequoia]

DeepSeek no longer has a limit, we'll see how long it lasts. If it continues like this, I'll stop complaining about Perplexity.

[hallow tulip]

They just upped it again to like 500 so basically limitless

Scratch that its back down to 100

[hazy moth]

Dang! That's exciting! That probably means we're getting 500 soon

Today, the limit seems to have gone from 10 to 25 to 50 to 100

All in one day

[past basin]

what am i missing in perplexity. i dont see 1o or deepseek r1 in the mac client.

[halcyon zodiac]

I wonder what’s their stack across iOS, Android, web, and macOS. The delay across platforms for different features makes me believe they may be native heavy

[mild valve]

This is such an unnecessarily disrespectful reply man, the app is convenient and it’s perfectly legitimate to wonder if this feature’s gonna make it there at some point.

[small tree]

perplexities o1 and r1 is handicapped

they're so fkn bad

the outputs are subpar

[mint hatch]

Welp it's been a while, since the last update for my case.

[odd lichen]

I just signed up for a free trial for Pro, and it’s available for me

[small tree]

what do you mean?

[mint hatch]

O1 is changed from the last update.

Has

[harsh elm]

Perplexity is probably the best Ai ive used

[mint hatch]

I use o1 for story writing. It can literally follow my long promt precisely and produce much longer result in each prompt compared to other models. 1 prompt can give around 5k to 9k+ word length results. Now? I wish. Not only shorter, but also dumber.

[small tree]

Exactly

Perplexity is a fad

theyre creating hype

stealing money from new customers, giving them an inferior product

they boosted the quota from 10 to 25, decreased context from 20k to 2k

disgusting

[sage sun]

I got the same problem with o1, it's unusable for me

[void sorrel]

it's because you didn't understand how it works. For creativity Claude is better ^^

to solve problems with scripted surgical precision, that's where O1 and R1 come in

[small tree]

Tony, are you paid by Arvind

you must be a paid shill

how much does he pay you to lie?

[void sorrel]

What

lol

[sage sun]

nah, I'm using it for coding, and it's been nerfed since the end on last week. It used to get long contexts and output a lot of tokens, now it's limited for both context and output tokens. Also, it used to spend minutes thinking, now it can't think for more than 10 seconds

[mint hatch]

For creativity sonnet is much better. But it is AWFUL for writing at length.

Previous o1 can write 5k+ in one go. Sometimes that's what I needed. Sonnet after a while is barely do 1k.

It won't.

[small tree]

You refuse to acknowledge that o1 and r1 have been nerfed on perplexity

it's an inferior product

[void sorrel]

just ask what the system prompt is and request to follow your writing style.

[mint hatch]

1k can easily achieve 4 paragraphs. But o1's 5k can achieve dozens. All I have to do is make the prompt as concise as possible, in which I always did.

[small tree]

Do you understand English

@void sorrel

it musn't be your native tongue

[void sorrel]

sorry

[small tree]

is Perplexity paying you in peanuts?

or Tacos

🌮

[sleek topaz]

could you show an example of this? i initially had the same thoughts but after doing more testing it was fine. I just had to change the "Recency" of the web search to more recent (the web search seemed to have been giving it trash context due to outdated info which made the response less than ideal).

[void sorrel]

he doesn't understand that models go through optimizations. He's too lazy to formulate his prompts then blames the company that has nothing to do with it lol

[sleek topaz]

i just wish there was a way to disable the "Pro" mechanism for r1

so it doesnt need to use web search

[sage sun]

it's not optimizations, it's nerfing. Perplexity is probably setting o1 reasoning to low effort and reducing output tokens limit to cut costs

[small tree]

"optimizations" do a direct comparison of both o1 and r1, o1 via chat GPT interaface and R1 on deepseeks own interface. Then compare the outputs on Perplexity

I wont engage with a paid shill any further

[small tree]

Exactly

[sleek topaz]

will you answer my question then? unless im a shill too

[small tree]

They cut costs by placing an Indian as the CEO, now they cut costs by reducing context length

Next they're going to have Indians typing out the responses for them, AI (Actually Indian)

[void sorrel]

So prove it and do your qualitative prompt tests. I really wish I was paid to tell the truth lol But I'm already quite satisfied with my carrier providing 1 Year of PPX

[small tree]

You use AI to correct your grammar. We use AI for more demanding tasks.

[sage sun]

they are giving it away to everyone, that's probably why they have to cut costs 😂

[void sorrel]

Do you need to criticize someone else's grammar to feel good? Poor boy

[acoustic salmon]

meh, obvious troll

[sleek topaz]

This guy might be the ultimate rage baiter, he can't even provide a singular instance of the model giving a worse output on pplx

[acoustic salmon]

obviously paid by a competitor to go in other Discords spreading junk

[small tree]

Yea just dismiss actual feedback by claiming we're trolls.

Competitor? Perplexity isn't even in the race anymore

It lost its USP (Unique selling point)

[sleek topaz]

sure but you ignored what i asked 3 times now, you cant even substantiate your claims

[small tree]

It oversold its subscriptions, burned through investor money

[void sorrel]

👍

ok little guy, feeling better now?

[small tree]

Tony, how many tacos did Arvind pay you

or was it Vindaloos

[sleek topaz]

most funny troll

[acoustic salmon]

nah, not even good, 0/10 would not troll again

[void sorrel]

.

[acoustic salmon]

any of y'all using VoiceMod for wake words on Android?

was wondering if it working out well before I drop it into Tasker

sorry it's AutoVoice, VoiceMod was something else

[small tree]

@crystal cosmos

hahahahhahha!

that's exactly what perplexity is using

[crystal cosmos]

indians eat tacos?

[small tree]

DeepSeek-R1-Lite-Preview-1B

[crystal cosmos]

yes

[small tree]

😂

[crystal cosmos]

NO

you are wrong

perplexity is using Groq mix with custom chatgpt api that

[small tree]

DeepSeek-R1-Lite-Preview-1B-Taco

[crystal cosmos]

hold on brb

[small tree]

and for the o1 model they're using

o1-alpha-go-200m

[sage sun]

here is an example, I asked if there are japanese non verbs that end with -U:

• Perplexity R1 gives me examples ending with -O and -I, and claims that they end with -U.
• Perplexity o1 thinks for 4 seconds and can't even finish the response.
• R1 on Deepseek platform gives the correct response.

[small tree]

Thanks M

They still wont acknowledge it

they'll tell you you're bad at prompts

even though they're exactly the same

[indigo valve]

500 usage limit?

[small tree]

🤣

[small tree]

it's fake

[void sorrel]

you are dumb and good at grammar . My congratulations

[indigo valve]

So complexity is lying to me?

[small tree]

Yea, complexity

[acoustic salmon]

wrong yet again

like everything else

[small tree]

lol, believable

[indigo valve]

What is that? the api?

[small tree]

ChatGPT Plus itself only offers 50 messages a week

how does perplexity offer 500?

[acoustic salmon]

I use this for language work, and yes you have to be concise in prompting, regardless of platform

[small tree]

[small tree]

Case in point

[acoustic salmon]

also false as I have that too

[small tree]

HAHAHAHAHAHHA

[acoustic salmon]

anything else you'd like to be wrong about?

lmao

[small tree]

"With a ChatGPT Plus or Team account, you have access to 50 messages a week with OpenAI o1 and"

Sams paying you too?

@acoustic salmon

[acoustic salmon]

source? because I've blown that easy

I use o1 for language

[small tree]

You dont use it enough lol

I also have chatgpt plus

[acoustic salmon]

132 message today

[indigo valve]

https://tenor.com/view/whoa-shaking-head-windy-wind-blown-away-gif-15465608

[sleek topaz]

Thats definitely just because of the web search as ive already stated, if they gave you the ability to not use the search it would give a similar response. Also "yesterday" in japanese technically ends in a U "kinou" is the romaji for it

[small tree]

at this point it just makes more sense to use OpenRouter

[sage sun]

yea, I'm using writing mode, it doesn't search the web but still shows Pro Search. I think that's what's ruining it too. Still a nerf imo

[sleek topaz]

Yes its only "nerfed" because of all the preamble stuff they're doing which I dont like

If that can all be disabled, 100% its 1:1 the same r1 as the real deepseek site lol

[small tree]

well right now its not

until it is, dont say it is what it isnt

[acoustic salmon]

the R1 model is the full model, same as the site but hosted in the US

[small tree]

right now it's operating like its been lobotomised

[acoustic salmon]

so it's not nerfed, but depending on the focus you've chosen it adds [insert whatever Perplexity adds for that focus] to your processing

[small tree]

It is nerfed

You need to look at it from a users perspective

I know you're paid

but just try to look at it from our perspective

[acoustic salmon]

I look at it from a perspective of someone who knows what they're doing lol

[small tree]

It has less functionality

therefore it is nerfed

see, 1 reasoning step

didn't even need to think

[acoustic salmon]

we understand you don't like Perplexity yet sit in its Discord asking if people are paid

attention is attenion, I know

[acoustic salmon]

that's unfortunately as far as you can go

[small tree]

unfortunately you can't even reason, you're preprogrammed by perplexity to provide canned responses

no thought

[acoustic salmon]

no thought would be further than you've gone lol

[small tree]

😂

[acoustic salmon]

but yes, if you use the writing focus, it speaks more directly to the model. It's not something I've skipped between with every prompt to test differences with, but seems to be the one that works as closely to the normal model as possible

[small tree]

circular reasoning

😂

[acoustic salmon]

otherwise you'll be adding on the additional customisations and whatever tweaks have been made for the other focuses

[abstract ibex]

Well it's good to know that it's not only cursor that suffers from this type of people

[wispy zephyr]

i'm impressed with the new r1 on perplexity. For my use cases, it gives better responses than the official deepseek chat

[sage sun]

until last week o1 was working great for me, not that they moved it to "Pro Reasoning" it's not the same. I can try a prompt from last week and it will get a much worse response, for now it no longer fit my needs
And as I showed on the japanese example, even R1 is performing better on Deepseek platform than on Perplexity.
I tried using R1 on perplexity all day for coding and it couldn't output long messages, so I endeded up using it on Cursor, and there it worked just fine

[wispy zephyr]

why are you guys comparing 2 different products lol

[sage sun]

I'm mainly comparing last week's o1 on Perplexity to the current one

[acoustic salmon]

They believe Perplexity's R1 is a nerfed version of R1

[abstract ibex]

They are looking for the cheapest access to the top models

[small tree]

Because we were sold on model selection and Arvind continues to change the way perplexity functions rendering it useless for us users who originally purchased our subscriptions for a use case advertised by perplexity?

[acoustic salmon]

Cheapest access would be the free one of course

[small tree]

You're literally all paid

shills

disgusting

[abstract ibex]

I got gifted a yearly subscription

[sage sun]

same

[acoustic salmon]

ignore him, he's just in it for attention and trolling

lol

[abstract ibex]

But I hear this all the time in every other AI product

You can see on cursor's server

People complain that cursor is tricking them!!!

[dense kelp]

Now we get 500

[hybrid rover]

hi

do we have perplexity a.i bot for discord?

[small tree]

@dense kelp 500 o1 prompts when openai only provide 50 a week on their plus plan

😂

we got such a good deal

[dense kelp]

We just used it at work and it optimized our code

[small tree]

insane

amazing

[acoustic salmon]

like has someone made a Perplexity bot for Discord or?

[hybrid rover]

yeah

i'm starting a web developer project with some friends and we use discord as our main communication method when not on real life meetings, so using perplexity a.i in a server would help us a lot since we use it for the same research

[acoustic salmon]

I haven't seen one, but what specifically from Perplexity's API would you be trying to use besides their Sonar model?

You can create a Space and invite people to collab in it, but not all threads seem to show up for everyone

but everyone can use the same tailored instruction set

[hybrid rover]

well, we use it the most to save our time with text stuff, let's say we are making a website about selling soda, and we need some soda info that is really easier to get with a.i

[hybrid rover]

interesting

[abstract ibex]

What are spaces used for?

[acoustic salmon]

You can give custom instructions to the model that apply to any prompts you use in that space

[hybrid rover]

that's really cool

[acoustic salmon]

Example, I have one for Language. It's instructions are to create a lesson for students based on an article or document I provide.

[hybrid rover]

now we're talking!

[acoustic salmon]

So I upload a doc or link a web article and just say "make a lesson" and it know to make 5 Multiple Choice Questions, 5 fill-in, and go over specific idioms or expressions

[hybrid rover]

alright

[acoustic salmon]

you can also do that with CustomGPTs in ChatGPT and Custom Gems in Gemini

[hybrid rover]

i don't really like those, that's why i use perplexity

[acoustic salmon]

I don't have Claude but assume it has something similar maybe

[hybrid rover]

it's the best free one and has a good privacy policy

[acoustic salmon]

Well Perplexity uses those models, as well as options for others

[hybrid rover]

hmm

[abstract ibex]

oh ok got it

[acoustic salmon]

that's why most people use Perplexity as opposed to the specific model subs. Most of them are $20/mo but you're just getting that set.

[hybrid rover]

it's all about speaking the a.i language

[hybrid rover]

yeah

[acoustic salmon]

prompt engineering is a whole thing aye

you can sometimes get better prompts by asking the AI to make the prompt based on what you're trying to achieve lol

[hybrid rover]

maybe i can use the a.i for some frontend work eh

[acoustic salmon]

little promptception

[acoustic salmon]

I use it to review my JS and Python code

and when I'm stuck ask it to make a specific function

[hybrid rover]

i get it

maybe i can do some basic html5 and javascript commands for it but i don't wanna rely on it for that

or else i'll get lazy

[acoustic salmon]

up to you

but hand jamming that stuff out can be tedious lol

[hybrid rover]

you seem to know a lot of it

[acoustic salmon]

you can always QC its output

[hybrid rover]

yeah, but i'm studying all of these frontend and backend stuff in technical school, so i'll make it worth

[acoustic salmon]

there's some people who can get it to code entire pages but I've never been able to accomplish that much

if you're learning, you can also have it do code challenges with you, giving you problems to solve with code and then it can evaluate how you did

[hybrid rover]

my friend does that

but he's way more advanced than me

i'm stuck on mysql server and php for now

already dominated most of frontend (bootstrap, hosting, css, javascript)

we're probably getting kotlin this year

[acoustic salmon]

now do Angular

and cry deeply

lol

[hybrid rover]

thanks for the chat dude!

[abstract ibex]

just created a space to find bugs in my code lol ty @acoustic salmon

[left solar]

Is there limit to use deepseek-r1? 10 times a day?

[harsh stag]

it's 100, but subjected to change throughout the day (most likely to keep increasing)

no actually it's 500 https://x.com/AravSrinivas/status/1884801300027589007

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

[naive roost]

Apologies if it came across that way. Thread was:

• Anyone know when r1 is made available in iOS app?
• Stop using iphone apps to access Perplexity and all others. Use mobile browser set to desktop view if mobile view doesn't give you everything.
• This is such an unnecessarily disrespectful reply man, the app is convenient and it’s perfectly legitimate to wonder if this feature’s gonna make it there at some point.

What am I missing?

[cobalt monolith]

Hmm interesting is it possible to use this on Android app?

Yes I can

What exactly is the difference between deepseek reasoning and "copilot"

Sorry pro search

[tight prism]

‪The DeepSeek api can answer it but only the desktop and app can’t‬

@signal hamlet o1 is missing in the app
Is that accidental or intentional?

[harsh stag]

2 reasoning models on perplexity work exclusively with Pro search, which is also a reasoning step where separate agent/model figuring out where to search, and the chosen model with take those curated information and synthesize the final answer

[shell jay]

Hi there. I got offered 1 month of Pro , but upon subscribing I got charged for it. Can anyone help?

[cobalt monolith]

Yes but which is better?

Is it specific use case?

[harsh stag]

they are 2 different things, and they work together to give you the answer

[cobalt monolith]

I am a bit confused by your answer lol, you say together but perplexity lets you choose between pro search and r1 reasoning

[harsh stag]

ah so you're referring to the dropdown, "r1 reasoning" option also uses pro search, but the model that synthesizes the answer is r1

if you only choose "pro search", it will use the model that you've chosen on the settings page

[cobalt monolith]

OHHHH

So let's say hypothetically there was a sonnet option it would be pro search and sonnet?

[harsh stag]

yep

[cobalt monolith]

Ah I see thanks

It's incredible I get 5 queries for free per day

And that's like a top model

[tame current]

BRO THE LIMIT IS 500 FOR DEEPSEEK

LESGO

[tame current]

I JUST SAW THEY MADE IT 500

earlier it was 15 for deepseek and o1 combined

then they increased it to 25

[cobalt monolith]

What is the o1 limit?

[tame current]

NOW ITS 500 FOR R1

wait i must be smoking something

why is o1's limit also 500?????

[cobalt monolith]

🤔

[cobalt monolith]

Tf

[cobalt monolith]

Btw sonnet is 300 correct?

[tame current]

wet dreams

i think 600

[cobalt monolith]

Check rq

[tame current]

600

[cobalt monolith]

Huh

Isn't r1 cheaper than sonnet?

[tame current]

i think so yes

[abstract ibex]

way cheaper

[cobalt monolith]

So why is it higher in number of queries compared to r1

Like 600 vs 500

[abstract ibex]

Cause R1 uses more tokens to answer you

Because of the reasoning

[cobalt monolith]

Huh I see

500 is still a lot ngl

[tame current]

i believe this is a bug

theres no way this is real

[abstract ibex]

lol

[tame current]

i just woke up

[abstract ibex]

It's just very cheap to run ig

[cobalt monolith]

Restart the page?

[tame current]

i think im still sleeping

[abstract ibex]

Cursor for example gives you unlimited R1

[tame current]

no change

[abstract ibex]

it's pretty great tbh

[tame current]

yeah thats crazy

[cobalt monolith]

Wtf

[tame current]

what was that alternative for

cursor

uh i forgot

[abstract ibex]

windsurf

probably

[cobalt monolith]

R1 is really disrupting the industry

[tame current]

they really broke the bubble

[abstract ibex]

but they don't give you any model for free except their llama 3.1

[tame current]

they even released janus

and people are saying its better than dall e