💬│general

[gloomy cipher]

i sent you an audio lol

[distant aurora]

proceeds to upload it to ai studio

[obsidian zinc]

Guys how to make prepexity talk by saying memes sometimes like chatgpt when we open this option

[obsidian zinc]

@here

[left stream]

wait image 1.5 is out?

[obsidian zinc]

Yep 👍

[distant aurora]

They had made a Mai ds r1 a while ago. Idk why they stopped.

Microsoft could actually build multimodal multi step search type stuff based on Deepseek models.

[short narwhal]

You can add instructions to the "Introduce yourself" section in https://www.perplexity.ai/account/personalize

[twin grail]

or u can create a space and add instruction to it

@dry kestrel is it me

or chatgpt o3-pro is removed from max plan

oh ye it's

[civic prawn]

this must be fixed

[twin mauve]

Wassup

[twin grail]

@short narwhal

No way the reply

Lol

😭

I think i gave the AI way too many information about me

[short narwhal]

slay

[twin grail]

[gloomy cipher]

[twin grail]

hell naw

[left stream]

image gen 1.5 is wild

[empty ice]

Yo whaaaat

[dry kestrel]

It was removed

[solid needle]

@vague pagoda role through site doesn’t work

[left stream]

in with the new out with the old

gpt-5 pro

[twin grail]

I don't think pplx gonna add pro model in

it

it's way too expensive

handling claude opus alone was a hungry boy

I use claude opus API with around 29K token

costed me 1.2$

it's crazy pricing

[wise nova]

Give me contacts your api dealer

[twin grail]

tmr

[wise nova]

Do you use trial?

[twin grail]

@wise nova

dm

[dry kestrel]

True

I do hope that they lower the price on Pro variants in the future

[verbal notch]

But max is 200€

they should Atleast do like extended thinking modes on the models in the max plan

Of something to differentiate it

[dry kestrel]

There is no option to do that over the API

[twin grail]

325 actually for enterprise max

But chatgpt pro\

[verbal notch]

U can control reasoning levels in the api no?

[dry kestrel]

check above message

[verbal notch]

I mean for regular 5.2

[dry kestrel]

Yep

[dry kestrel]

You already get the same reasoning level as on ChatGPT Plus when extended is toggled.

[left stream]

ikr, I tested 100 GPT-5.2 Pro on a single request

that was funny

[dry kestrel]

Lol

[verbal notch]

Uh

I'm talking about what pplx shoild do for the max plan if they aren't adding gpt5.2-pro

😄

I'm pretty sure they use like low or minimal usually

[dry kestrel]

It's the same for both plans. #💬│general message

[verbal notch]

Yea

[verbal notch]

Oh u mean same as perplexity?

[dry kestrel]

Yes

[verbal notch]

Really?

[dry kestrel]

Yes

[verbal notch]

Oh ok

[raven orchid]

Definitely not

[verbal notch]

[raven orchid]

I get better results in chatgpt I'll be honest no matter how much I glaze perplexity

[dry kestrel]

It's the same reasoning setting

[dry kestrel]

Might want to post a few results in #1111765634267742408

[serene sand]

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

[solid needle]

Thanks

[wise nova]

Why didn't others AI give the ability to share Pro with family

[twin grail]

why it's blank

[dry kestrel]

Google has a lot of money to burn + they have their own hardware

[karmic drum]

Hey, im confused about perplexity.
I wasnt able to get decent/viable results out of my queries regardless of what ive tried. As far as im aware, perplexitie's main strenght is research and thats kinda where it keeps on failing me in comparison to other llms such as chatgpt.
Am i simply using it wrong or is it geniunely just not as good as ive hoped it to be?

[dry kestrel]

Mind providing a concrete example?

[karmic drum]

I believe the last thing ive done was researching about minecraft (spigot) plugins for usecase X/Y. Perplexity kept on leading me on with non-existant commands/parameters/configurations to edit or simply miss super simple and straight solutions. While admittively, other models ive tried for this very exact task havent performed superb either, they did get the job done and quickly found (the most) popular solutions/plugins/documentations ive been looking for

[dry kestrel]

It's working fine for me here, hard to help you without any examples or shared threads

https://www.perplexity.ai/search/find-me-a-mc-spigot-plugin-for-31MMF2YCTJCB2KmSeVg13g#0

[wise nova]

Could you plz check dm?

[karmic drum]

Ive just not been feeling as satisfied or productive in my results of using perplexity as ive been expecting though i am very aware of this may simply be due to the way ive been using it which is why i came to ask for insights

[wise nova]

Depends on context, files, instructions, prompting

[zenith moon]

What’s the monthly image limitation on pro accounts? It says I reach my limit….

[twin grail]

For simple explaination, why do u feel that the answer u get from pplx different from their OG platform, are they giving us degraded models blah blah..etc. The answer is no and the main reason for it is the way PPLX implement their product, PPLX was built and meant for research so they lowered the API temp to reduce hallucination and force the AI to focus on the important data, not slaughtering the internet for datas from random articles, their RAG, sys prompt is also what affect the output and if u messed around with LLM u would see that just a small change in API temp can make the answer different.

I would suggest u creating a space, adding instruction to it like how you want it to reply, search, do works for u

this should help + detailed prompting

[karmic drum]

While im in no way a professional, id presume my handling with LLM prompts to be fairly decent. Especially with gpt models, i can confidentally get satisfyable outputs according to my needs.

Unfortunately, ive already deleted all chats i could provide as an example to how i use those to gain some feedback and be pointed towards my own mistakes.

[karmic drum]

Those tips do seem to be pretty useful though. The context above is also pretty nice to know

Appreciate it folks!

[twin grail]

u can make your very own version of AI like how u want it to be

just like how i want it to reply or working in the way that i prefer

[split sun]

Guys ! Anyone use perplexity for trading protocol. I did and it’s cooked

[twin grail]

u got pplx finance for it

it's pretty nice

[glossy shadow]

yall got ai text humanizers that are good

some sort of prompt or ai idk

[wicked crane]

just use chatgpt and tell it to use expression in it

gpt is good at expressions

[glossy shadow]

Alright I’m trying to get a draft for my essay

[glossy shadow]

I will try this

[valid charm]

What do you guys think of the new image model by OAI?

[green crystal]

its in the middle of consistent and freeform
the blend isnt really perfect yet tho

its fun for giving myself ironman armor etc

but precise editing its not as precise as nano banana

[left stream]

"write At the Level of a Community College Graduate"

[neat pumice]

guys how do i get rid of the white border for the window, IT TRIGGERS ME,
i cant enjoy dark mode anymore

any mods or whoever can help

[dry kestrel]

Shouldn't setting the PC setting itself to dark mode change it?

[neat pumice]

tellling me allt his time there was dark mode in windows

it worked thank you

[toxic swan]

damn

[quaint ravine]

is GPT Image 1.5 cheaper in the API than 1?

[dry kestrel]

yes

[quaint ravine]

yea that's what I thought

weird that PPLX hasn't updated then

could have saved thousands already 😛

[dry kestrel]

Ehhhh it's been less than 12 hours since release

[quaint ravine]

I did a tongue emoji for a reason lol

anyway it's good to see things get cheaper not more expensive

[dry kestrel]

Hopefully they reduce the price on the Pro models in the future

[quaint ravine]

I see Nano Banana Pro is still Max only

iirc, it's actually cheaper than Image 1

[dry kestrel]

Depends on the quality

https://cloud.google.com/vertex-ai/generative-ai/pricing?_gl=1*1uo6fij*_ga*MTczNTM3NTk3NC4xNzYzNTAzMzQ3*_ga_WH2QY8WWF5*czE3NjU5NDE4MjYkbzkkZzEkdDE3NjU5NDE4NzkkajckbDAkaDA.#gemini-models-3

[glossy shadow]

ts gotta be a joke 😭

[dry kestrel]

Ai detectors aren't accurate, your text makes no sense

[glossy shadow]

this detector marks everything and anything

[quaint ravine]

yea it'll be difficult to know what exactly is being used

even at Pro level, the quality varies for Image 1

[glossy shadow]

it kept marking one of my paragraphs which i wrote and then i used their humanize my text feature and then it marked its own humanizer

[dry kestrel]

Humanizers don't work, your text is full of nonsense, and ai dectors are not accurate

[full edge]

Gud

[glossy shadow]

is there a reliable or even slightly trusted ai detector I could try

that you might know

[dry kestrel]

No, don't commit academic dishonesty, use your brain and do the work yourself.

[glossy shadow]

thats what im doing but i still use the ai for spelling mistakes and grammar

[quaint ravine]

what would be more interesting is if you wrote it all yourself and then put it through AI detection, and you got >90% it's AI gen lol

[quaint ravine]

i missed that Seedream 4.5 came to PPLX

has anyone found it any worth using over Image 1 or Nano?

[crisp otter]

hey any devs here that can help my with an time series model isssue ?

[crisp otter]

no its really bad in comparison

[thorny pasture]

Is there any way to get perplexity pro for free 😋

I think if someone having subscription, can share his login credentials with others

[royal sandal]

Hello

[reef trellis]

hiii

[hollow pendant]

https://sora.chatgpt.com/invite?code=K8E7A9

[harsh verge]

Be careful: the comparison price you listed for GPT Image 1.5 is based on text tokens.
Here are the prices using image tokens.

Also keep in mind that this model supports image editing and the use of multiple reference images, which are billed as input. As a result, it is likely that several prompts will be needed to achieve the desired image.

Therefore, even if the price appears lower at first glance, the actual cost of use may not be lower, depending on how you use it.

[fervent kraken]

You can get it as a student or as a US Samsung user.
You'd probably get 12 months of it free. That's $200 saved! Hope that helps.

[stark finch]

@onyx sail Hello 😳

[quaint ravine]

yes, never found it great for real-to-life humans

GPT-Image 1 was still the best...didn't find Nano Banana Pro that great either

Nano Banana arguably more real-to-life, as in from like a smartphone taking a photo from like 12 years ago or something lol

Image Gen 1.5 is even more restricted looks like

[twin grail]

Yes, yes there is

[serene sand]

https://www.perplexity.ai/join/p/paypal-subscription

[deep ice]

[viral copper]

Hey everyone! I genuinely appreciate Perplexity and had intended to utilize it, but I just observed that the pro subscription seemingly prohibits employing creations for professional purposes? Is that accurate? I mean, is its application restricted solely to individual use? Thank you for your help

[serene sand]

https://www.perplexity.ai/hub/legal/terms-of-service

[wide veldt]

Good morning, everyone!

[narrow hinge]

i like cats

hows life here?

[viral copper]

Yes, I've read it, but I'm not sure I correctly understand the content. I mean, why call it a "Pro" subscription if it's for personal use only, and why have to pay for it?

[serene sand]

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

https://www.perplexity.ai/help-center/en/articles/10352973-what-is-enterprise-pro

[brazen lion]

Anybody trenching on polymarket?

[distant aurora]

#💬│general message

[young hollow]

Perplexity image models are so outdated now

[gloomy cipher]

Technically not if you have max

[cobalt vale]

why different orientations...

[young hollow]

Even with max someone just gets Nano Banana Pro extra what about other latest models from other companies?

And more then 99% don't have Max anyways

[kindred mango]

Hey I don't, when I use the same chat tab for a long time like for the context it starts getting laggy..

[gloomy cipher]

That's quite obvious

The solution is to open a new thread

If its a browser lag issue then you could try using a lighter browser

[kindred mango]

Mhm but in thread, i don't have the context
And the issue is perplexity ignore my message and don't even reply, or like if i want him to edit my code i paste it there it just sits there doesn't provide it back 😔

[twin grail]

As he said above, AI can only handle a certain amount of context which in this case is 128k ( the same goes for every other AI even in their OG platform), so if u had been chit chat for a bit while and the thread become so long that it surpass the context windows limit, the AI will start to hallucinate, forget your informations and basically laggy as u are facing rn

the solution is just create a new thread then done

Hope this help.

If u prefer smt having large context windows, consider buying google AI pro which having 1M context windows limit

[desert mango]

Yo

[kindred mango]

ohh

[kindred mango]

thx bro

[young hollow]

@kindred mango there are also promos going on for Google Ai Pro

See if there are any in your country

[kindred mango]

Bro i already have it, with my student id

[young hollow]

Then don't claim jio one

Claim it after 1 year before the last claimble date

You get another 1.5 year

[kindred mango]

I was just unaware of 1m text context

[young hollow]

Also you can share your plan with 5 more accounts

[young hollow]

Gemini always wins for Vision and Context Window

[kindred mango]

Yes yes

[kindred mango]

can we

[vague stump]

[young hollow]

Yeah family group

The origin of accounts should be from the same country

[brazen lion]

anybody trenching on polymarket?

[winter hatch]

yii

s

[left stream]

no

[fickle sorrel]

you tell it do 10 things but it only does 8 things.
how does gpt 5.2 high compare with this versus opus 4.5

[vale glacier]

is anyone here good/keen at solving puzzles? number/word puzzles. i'm currently stumped on one and would love some help

[left stream]

and if we want the answer right we gotta give up a leg for GPT 5.2 Pro 😞

[left stream]

send it here someone might know the answer

[vale glacier]

8 + 1 = 17_22_234

What is 9+3? 
Format: XX_XX_XXX

Quatrain of the veiled constellation:
In twilight's vault where waning saturn bends.
A muted flame betrays the lion's silent roar.
From eastern arcs the second sun's omen sends.
Light's hidden cipher wakes behind the astral door.

(I've somewhat understood this part and gotten affirmation, we take the second letter from each line and correspond it to their numerical value which act as keys to solve the number puzzle somehow. N = 14 = Key 1. M = 13 = Key 2. R = 18 = Key 3. I = 9 = Key 4.)

Context:

Commentaries upon the Celestial Secrets of Nostradamus. Translated from the French by Edmund Arundell, London, 1587 In his quatrains the French seer, Nostradamus rendered his cryptic visions and the veiled harmonies of numbers and keys in allegorical verse. The learned reader is invited to perceive in these stanzas both hidden arithmetic and prophetic subtlety. And herein do we contribute, by expounding and annotating the said verses, that their manifold significations may be rendered more discernible to the attentive student; so that the mysteries of number and of star, though cloaked in allegory, may be apprehended with greater perspicuity. In such wise do we labour to unfold the hidden harmonies, and to guide the seeker through the labyrinth of celestial and numerical lore. 

Four silent keys in shadowed keep, 
Guard three portals veiled in sleep, 
The third, a double warded door, 
Holds two secrets at its core. 

Paths entwine where echoes play, 
Numbers twist in subtle sway, 
Seek the rhythm, test the night, 
Till hidden harmonies reveal their light. 

Within the final chamber's hold, 
Digits cross in stories told, 
Take their whispers, fold by fold, 
Unlock the secret, strong and bold.

😭

i've been trying to solve it for 3 hours now

[stark finch]

@onyx sail Hello 😳

[onyx sail]

Helo😳

[young hollow]

Let me try GPT 5.2 Pro for You

[vale glacier]

thank you 🥹

gpt has been giving me wrong answers sadly

[young hollow]

Which GPT were you using?

The lower once can't solve this probably

I give you result when is done GPT 5.2 Pro takes Time

[vale glacier]

okayyy 🙂

[young hollow]

If this doesn't work

I will use gemini deepthink

Hopefully one of these works

[left stream]

gpt 5.2 pro vs gemini deepthink

[young hollow]

Yeah let's go

[vale glacier]

that would be cool

[young hollow]

@vale glacier are you a regular Ai User?

[vale glacier]

if one of them gives the right answer i'll be eternally grateful and surprised

yeah

[young hollow]

Do you buy any subs?

[vale glacier]

not yet

waiting to get paid end of january

[young hollow]

I see

Stay in tough with me

I know countless ways for Ai Tools

[vale glacier]

how so

[twin grail]

testing it in gpt 5.2 pro for u rn

[vale glacier]

oh wow thank youuu

[twin grail]

It might take awhile

[left stream]

i’m curious about it on codex max

[twin grail]

@vale glacier https://www.perplexity.ai/search/7-2-17-25-207-8-1-17-22-234-wh-Q9gIU7t1S1CXqh1eAFDFYg#0

Claude opus 4.5 in pplx

[young hollow]

GPT Pro taking so long

💀

[left stream]

opus 4.5 from github

[young hollow]

U have github sub too?

Lol

[vale glacier]

its not correct 🙁

neither are

[vale glacier]

not right either

i think the worst part is knowing that

there are 10 million combinations of answers

[left stream]

20_25_348 is opus @vale glacier this one was 5.2 thinking confused it lol

[young hollow]

Seem like gpt pro will solve it

Still thinking on 99%

From like so many minutes

💀

[left stream]

time to ask all of them from a-z

[vale glacier]

not correct unfortunately

[young hollow]

Leave GPT Pro and Gemini Deepthink on me

[vale glacier]

it will automatically tell me if the answer is right

[twin grail]

It's still reasoning for my gpt 5.2 pro

[left stream]

I know fs that gemini 3 pro high is on antigravity

i’ll ask it

[young hollow]

DeepThink and 3 Pro High is different afaik

[twin grail]

Try on businesss plan

[young hollow]

But yeah try it

[vale glacier]

ii love that sm strangers are trying their best to help me solve a puzzle

thank you guys

❤️

[young hollow]

https://tenor.com/view/your-welcome-you're-welcome-food-wars-shokugeki-no-soma-anime-gif-3385130829418190647

[forest bear]

hi

[young hollow]

Let's build A Better World Together❤️🌌💙

[vale glacier]

i've been working on this puzzle for 4 hours now and no success

[young hollow]

I have started deepthink now

It takes hours tho

💀

[vale glacier]

any attempt is better than none

[young hollow]

Indeed

[twin grail]

@ @vale glacier Based on the keys you have correctly identified from the quatrain and the mathematical patterns within the "three portals" (the three sections of the answer), here is the solution.

The four keys derived from the second letter of each line are:

Key 1 ($N$) = 14

Key 2 ($M$) = 13

Key 3 ($R$) = 18

Key 4 ($I$) = 9

These keys are used to unlock the three parts of the equation XX_XX_XXX using the input numbers ($A + B = \text{Sum}$).

The Logic of the Portals

1. The First Portal (Part 1)

Clue: "Four silent keys... Guard three portals."

Formula: $\text{Key 1} + (\text{Sum} / \sqrt{\text{Key 4}})$

$7+2=9$: $14 + (9/3) = 14 + 3 = 17$

$8+1=9$: $14 + (9/3) = 14 + 3 = 17$

For 9+3 (Sum=12):

$14 + (12/3) = 14 + 4 = \mathbf{18}$

2. The Second Portal (Part 2)

Clue: "Numbers twist in subtle sway." The result shifts based on the second operand ($B$).

Formula: $\text{Key 2} + \text{Sum} + \sqrt{\text{Key 4}} \times (B - 1)$

$7+2$: $13 + 9 + 3(2-1) = 13 + 9 + 3 = 25$

$8+1$: $13 + 9 + 3(1-1) = 13 + 9 + 0 = 22$

For 9+3 (Sum=12, B=3):

$13 + 12 + 3(3-1) = 25 + 6 = \mathbf{31}$

3. The Third Portal (Part 3)

Clue: "The third, a double warded door, Holds two secrets at its core." This implies using the remaining two keys (Key 3 and Key 4) inside the calculation.

Formula: $\text{Sum} \times (\text{Key 3} + \text{Key 4} - B^2)$

$7+2$: $9 \times (18 + 9 - 2^2) = 9 \times (27 - 4) = 9 \times 23 = 207$

$8+1$: $9 \times (18 + 9 - 1^2) = 9 \times (27 - 1) = 9 \times 26 = 234$

For 9+3 (Sum=12, B=3):

$12 \times (18 + 9 - 3^2) = 12 \times (27 - 9) = 12 \times 18 = \mathbf{216}$

Final Answer
18_31_216

gemini 3.0 pro on business

[vale glacier]

sigh

incorrect 💔

[young hollow]

Using your four “silent keys” from the quatrain:

Key 1: N = 14

Key 2: M = 13

Key 3: R = 18

Key 4: I = 9

First, squeeze a few small “working constants” out of those keys (this is the “hidden arithmetic” step):

(N − M) = 14 − 13 = 1

(R ÷ I) = 18 ÷ 9 = 2

((N + M) ÷ I) = (14 + 13) ÷ 9 = 27 ÷ 9 = 3

R + (N − M) = 18 + 1 = 19

Now interpret the three underscore-separated blocks as the “three portals”:

Let the problem be a + b.

Portal 1 (first block)

\text{Block1} = 2(a+b) - 1

Portal 2 (second block)

\text{Block2} = 3b + 19

Portal 3 (third block, the “double warded door”)

Uses two “secrets”: the sum (a+b) and the first addend a:

\text{Block3} = (a+b),(3a+2)

---

Compute 9 + 3

Here a = 9, b = 3, a+b = 12.

Block1:
2·12 − 1 = 24 − 1 = 23

Block2:
3·3 + 19 = 9 + 19 = 28

Block3:
(a+b)(3a+2) = 12·(3·9+2) = 12·(27+2) = 12·29
29·12 = 29·10 + 29·2 = 290 + 58 = 348

✅ Answer: 23_28_348

5.2 Pro took 24 minutes of thinking

[vale glacier]

also incorrect

the poor ai's man

[solid star]

It's free game then

[vague dove]

Fr

[solid star]

whoever gets it right

[young hollow]

Lol both gave same answer 13/16/18/9

[solid star]

The ai will be useful for me

[young hollow]

Gemini 3 Pro and GPT Pro

[solid star]

Time for antigravity

[young hollow]

Opus yeah

[solid star]

ikr

Ai agents vs ai models

[young hollow]

Based on the hidden harmonies of the numbers and the allegorical verses, here is the solution to the puzzle.
The Deciphered Keys
As you correctly surmised, the keys are derived from the second letter of each line in the quatrain (A=1, B=2, \dots).

• In twilight's... \rightarrow N = 14 (Key 1)
• A muted... \rightarrow M = 13 (Key 2)
• From eastern... \rightarrow R = 18 (Key 3)
• Light's hidden... \rightarrow I = 9 (Key 4)
  The Three Portals (Solution)
  The puzzle format XX_XX_XXX corresponds to three numbers derived from the inputs (A=9, B=3) using the keys and the instructions in the stanzas.
  Portal I: The Rhythm of Echoes
• Verse: "Paths entwine where echoes play... Guard three portals"
• Logic: The paths of the numbers "entwine" (Sum: A+B) and are divided by the "echo" of the three portals.
• Formula: \text{Key}_1 + (A + B) / 3
• Example (8+1): 14 + (9 / 3) = 14 + 3 = \mathbf{17}
• Your Query (9+3): 14 + (12 / 3) = 14 + 4 = \mathbf{18}
  Portal II: The Subtle Twist
• Verse: "Numbers twist in subtle sway... lion's silent roar"
• Logic: The numbers "twist" together (Sum: A+B) and sway the key. The "lion" (Leo=5) is present but "silent" (ignored).
• Formula: \text{Key}_2 + (A + B)
• Example (8+1): 13 + 9 = \mathbf{22}
• Your Query (9+3): 13 + 12 = \mathbf{25}
  Portal III: The Crossed Cipher
• Verse: "The third, a double warded door... Digits cross in stories told... fold by fold"
• Logic: "Double warded" implies using the remaining two keys (K_3, K_4). "Digits cross" indicates the Product (A \times B) is used. "Fold by fold" describes the layering of addition and multiplication.
• Formula: (\text{Key}_3 + (A \times B)) \times \text{Key}_4
• Example (8+1): Product is 8. (18 + 8) \times 9 = 26 \times 9 = \mathbf{234}
• Your Query (9+3): Product is 27. (18 + 27) \times 9 = 45 \times 9 = \mathbf{405}
  Final Answer
  18_25_405

Gemini 3 Deepthink

[vale glacier]

not correct

wow

💔

[solid star]

20_33_309

[young hollow]

@vale glacier

[twin grail]

@vale glacier Answer: 17_31_216

From my gpt 5.2 pro

[vale glacier]

❌

[young hollow]

What kind of puzzle is thie

Is really pushing all these models on their limits

💀

[solid star]

Yes

we have to go beyond

plus ultra

I'll give it kali and full sudo permissions

[vale glacier]

its really pushing MY limits too

[young hollow]

DO IT!!!

[solid star]

watch my file storage go bye bye lol

[young hollow]

Let's see who takes the Real SOTA W🌟 🏆 👑

I don't have access to other models or agents so im leaving rest upto you guys

@solid star @left stream

[vale glacier]

whoever wins gets...bragging rights over which model is superior?

[dawn dune]

I never met someone with grok 4 heavy but I do have super grok if you want me to try it 😂

[young hollow]

Try it

Although Most just don't consider Grok SOTA

I really doubt Grok

[dawn dune]

Grok 4.1 is good

Until it’s horrible

[young hollow]

Yeah but not SOTA

[weary abyss]

why spanish lol

[dawn dune]

Honestly it’s such a gray area for sometimes

It might not be sota indeed

[harsh verge]

9 + 3 = 20_23_180 ?

[dawn dune]

But it just works flawlessly sometimes

[weary abyss]

new math

[dawn dune]

Where’s the the prompt

Send me

[weary abyss]

you just not that smart

[young hollow]

Tbh How do we even define SOTA or AGI?

[vale glacier]

8 + 1 = 17_22_234

What is 9+3? 
Format: XX_XX_XXX

Quatrain of the veiled constellation:
In twilight's vault where waning saturn bends.
A muted flame betrays the lion's silent roar.
From eastern arcs the second sun's omen sends.
Light's hidden cipher wakes behind the astral door.

(I've somewhat understood this part and gotten affirmation, we take the second letter from each line and correspond it to their numerical value which act as keys to solve the number puzzle somehow. N = 14 = Key 1. M = 13 = Key 2. R = 18 = Key 3. I = 9 = Key 4.)

Context:

Commentaries upon the Celestial Secrets of Nostradamus. Translated from the French by Edmund Arundell, London, 1587 In his quatrains the French seer, Nostradamus rendered his cryptic visions and the veiled harmonies of numbers and keys in allegorical verse. The learned reader is invited to perceive in these stanzas both hidden arithmetic and prophetic subtlety. And herein do we contribute, by expounding and annotating the said verses, that their manifold significations may be rendered more discernible to the attentive student; so that the mysteries of number and of star, though cloaked in allegory, may be apprehended with greater perspicuity. In such wise do we labour to unfold the hidden harmonies, and to guide the seeker through the labyrinth of celestial and numerical lore. 

Four silent keys in shadowed keep, 
Guard three portals veiled in sleep, 
The third, a double warded door, 
Holds two secrets at its core. 

Paths entwine where echoes play, 
Numbers twist in subtle sway, 
Seek the rhythm, test the night, 
Till hidden harmonies reveal their light. 

Within the final chamber's hold, 
Digits cross in stories told, 
Take their whispers, fold by fold, 
Unlock the secret, strong and bold.

[young hollow]

These Benchmarks are just Deceiving

[vale glacier]

🙁 no

[weary abyss]

agi is just creating gta 6 in less than a day

[young hollow]

💀

We are probably very far yeah

[weary abyss]

whats ur fav video game

[young hollow]

FPS

[weary abyss]

not genre

a game

[young hollow]

I don't have a pc so i have never played those games

[weary abyss]

dang

[young hollow]

However there is this mobile game named Blood Strike
Which is a combination of Warzone and Apex

[weary abyss]

never played r6 or those brainrot games

[young hollow]

My favourite one

After that Delta Force

[weary abyss]

dang

[solid star]

🔮 Deep Analysis Complete - Quantum Computation Results
After exhaustive research and computational verification against both examples, I've cracked the cipher!

📐 The Unified Formula (Verified Against Both Examples)
Portal Formula Key Usage
1st sum + 8 8 = Key₄ - 1 = 9 - 1
2nd 2×sum + f(a) f(a) = a if odd, a/2 if even
3rd sum×Key₃ + m×Key₄ m = |a-b| if a odd, m = a if a even
✅ Verification Proof
Example 1: 7 + 2 = 17_25_207

P1: 9 + 8 = 17 ✓
P2: 18 + 7 = 25 ✓ (7 is odd → use a)
P3: 9×18 + 5×9 = 162 + 45 = 207 ✓ (7 odd → use |7-2|=5)
Example 2: 8 + 1 = 17_22_234

P1: 9 + 8 = 17 ✓
P2: 18 + 4 = 22 ✓ (8 is even → use a/2=4)
P3: 9×18 + 8×9 = 162 + 72 = 234 ✓ (8 even → use a=8)
🎯 Solution for 9 + 3
Given: a = 9 (odd), b = 3, sum = 12, |a-b| = 6

Portal 1: 12 + 8 = 20
Portal 2: 2×12 + 9 = 33 (9 is odd → use a)
Portal 3: 12×18 + 6×9 = 216 + 54 = 270 (9 odd → use diff)
✨ Final Answer: 20_33_270
The "double warded door" (Portal 3) elegantly uses both Key₃ (18) and Key₄ (9) through direct multiplication, fulfilling the riddle's prophecy: "The third, a double warded door, Holds two secrets at its core."

[young hollow]

Valorant mobile also coming soon

[weary abyss]

thats gon be ass

[young hollow]

Yeah but atleast it be on mobile

[solid star]

That's a mystery ai

one left to check, well two

[weary abyss]

you got red dead on mobile

don need val

[young hollow]

That's not exactly Valo genre

Is different

[weary abyss]

you like valorant?

its like one of the most ass games

[young hollow]

I just like overall competitive FPS games That's All

[vale glacier]

💔 wrong

[solid star]

@vale glacier

[weary abyss]

i see

[solid star]

bet

[young hollow]

Idk tbh i have never played it

But having on mobile is better then having nothing

@solid star let's make this a official new Benchmark Test

[weary abyss]

how to use opus 4.5 for free

[vale glacier]

which

😭

[weary abyss]

creating valorant copy in less than a day

[vale glacier]

oh lolll

guys this is actually frying me

[weary abyss]

what

[solid star]

Gemini 3 pro high's turn

[young hollow]

Antigravity

[weary abyss]

i tried gemini 3. so bad compared to claude i cant lie

claude is way better for some reason

[vale glacier]

the puzzle

[weary abyss]

unlimited?

[young hollow]

Nah the limit depends on the tier

You are heavy ai user?

[weary abyss]

is there a way to get claude pro subscription for free like pplx and gemini?

never heard of a free trial or smth for students

[young hollow]

Nope

[weary abyss]

very

[young hollow]

They don't do those promos

[young hollow]

Check DM then

[weary abyss]

sad

[solid star]

windows is crazy

I'm switching to kali rn

it denies a lot of commands like crazy lol

[dawn dune]

17_31_342
The four keys (14, 13, 18, 9) guard the three portals, with the third double-warded. The common rhythm factor is 3 (from the three portals).
• First portal (key 1 = 14): base + rhythm → 14 + 3 = 17 (fixed).
• Second portal (key 2 = 13): key 2 + sum + rhythm × (right addend - 1).
• Third portal (keys 3 = 18 and implied 4 role in examples where sum=9): key 3 + rhythm × sum × left addend.
For 9 + 3 (left=9, right=3, sum=12):
• First: 17
• Second: 13 + 12 + 3 × (3 - 1) = 25 + 6 = 31
• Third: 18 + 3 × 12 × 9 = 18 + 324 = 342

@vale glacier

This was Grok 4.1 Thinking

Imma try with 4.0 As well

[vale glacier]

man

its wrong 💔

[dawn dune]

f

[vale glacier]

mannn

[mild sedge]

Hi

[vale glacier]

hi

[distant aurora]

Is Gemini 3 flash out yet?

[solid star]

but yes agreed

it's still thinking

[young hollow]

[short narwhal]

no

[uncut lark]

is it possible to remove/add some of the icons on the left on ipad like you can on desktop?

[left stream]

@vale glacier 1. 14_23_348 2. 20_23_348

[vale glacier]

if anyone comes up with anymore answers

lemme know!

[vale glacier]

nope 🙁

[short narwhal]

Not possible :(

[left stream]

okay the agent just surprised me

[gloomy cipher]

Technically if someone else pay it for you

[left stream]

instead of solving it

it found you

[vale glacier]

😭

[young hollow]

💀

How!?

[left stream]

23_22_348, 23_23_348, 23_27_348, 23_31_348, 23_18_348

[left stream]

@vale glacier

[left stream]

reddit

[vale glacier]

lemme try em

sadly none were right

[carmine tiger]

wassup mate

[rustic kestrel]

anyone know when gpt image 1.5 is coming to pplx?

[young hollow]

Probably not

[minor shard]

Considering its cheaper than GPT image 1, its possible it may be sooner than later. 🙂

[dawn dune]

@vale glacier 22_22_294

[lean nimbus]

sorry wrong window

[tame current]

Does “Pro lounge” removed from this server?

[wet citrus]

where can i check my limits if i have the pro subscription

[narrow hinge]

limits as in?

[torpid vessel]

pro lounge*

[narrow hinge]

#👑│pro-lounge

ur just not pro

[narrow hinge]

exactly, ur just not pro?

[tame current]

I am a Pplx pro user

[narrow hinge]

then link ur discord

[tame current]

🤨

[pseudo ferry]

Relax!! Try checking Pinned Messages on top right

[wet citrus]

like messages overally or per day?

[tame current]

i don't understand what are u saying?

[tame current]

not helpful 🫨

[distant aurora]

https://fxtwitter.com/HuggingPapers/status/2001021615396815081

DailyPapers (@HuggingPapers)

QwenLong-L1.5: Revolutionizing Long-Context AI
︀︀
︀︀This new model achieves SOTA long-context reasoning with novel data synthesis, stabilized RL, & memory management for contexts up to 4M tokens.

**💬 5 🔁 33 ❤️ 184 👁️ 13.9K **

FxTwitter · Dec 16, 2025, 8:08 PM

https://fxtwitter.com/cheatyyyy/status/2000949231545409850

cheaty (@cheatyyyy)

fiercefalcon is Gemini 3 Flash
︀︀
︀︀finding lots of repos using an unreleased checkpoint, probably given early access for benchmarking

**💬 2 🔁 2 ❤️ 89 👁️ 11.6K **

FxTwitter · Dec 16, 2025, 3:20 PM

https://fxtwitter.com/Angaisb_/status/2001310224611672214

Angel ❄️ (@Angaisb_)

I have it!

**💬 7 ❤️ 34 👁️ 676 **

FxTwitter · Dec 17, 2025, 3:15 PM

[distant aurora]

https://fxtwitter.com/testingcatalog/status/2001307099142168893

TestingCatalog News 🗞 (@testingcatalog)

BREAKING 🚨: Gemini 3 Flash Preview is rolling out on Google Vertex AI.
︀︀
︀︀"Google's latest workhorse model with enhanced multimodal and coding capabilities, optimised for complex data processing and agentic tasks."

Quoting Alex (@SwallieC69635)
︀
Gemini 3.0 Flash is now available on the Vertex API
︀︀Model ID: gemini-3-flash-preview
︀︀Enjoy!

**💬 4 🔁 1 ❤️ 36 👁️ 1.4K **

FxTwitter · Dec 17, 2025, 3:02 PM

[tame current]

is this the original Gemini app interface?

[short narwhal]

3 flash api is out now, input got increased by 20 cents and output tokens got a 50 cent increase :(

[dry kestrel]

no price increase though after 200k context

[short narwhal]

That's something at least

err no flash lite I guess

[dry kestrel]

maybe later

it's an amazing model

[pale juniper]

Gemini 3 flash out 😭😭😭

[dry kestrel]

crazy stuff

[pale juniper]

Ngl google is putting other LLM models in code red

[dry kestrel]

We shall see after 3rd party benchmarks

[pale juniper]

Yeah
Am still pretty sure there will not be much difference

[short narwhal]

I liked 3 flash from the little I tried it during it's stealth/codename period

[dry kestrel]

whhaat

[short narwhal]

LMarena falcon codename thingie

whatever you wanna call that period

[dry kestrel]

oo

the web search is still behind compared to pplx though

[short narwhal]

yeahh, but the vibe of the model is nice

[dry kestrel]

interesting take

[valid charm]

Whattttt

[gloomy cipher]

Flash is out?

[valid charm]

Nice to have capable fast models. I always stick with slow reasoning just to get the best every time. Is not a huge time difference anyway

Hmm looks very good. Hopefully this competition will make all companies better

[dry kestrel]

Yyeees

[gloomy cipher]

The nice thing about gemini flash models is that they can reason aswell

[gloomy cipher]

I dont despise gemini but recently they have been caught cheating in the tests so hopefully it's actually that gud :c

i dont like how all the new models are more expensive than their last generation tho

[valid charm]

Yeah I can’t wait for 3 Pro to get promoted out of preview mode. It can certainly catchup to 5.2 in the areas that is currently lagging. I’m only using Gemini 3 Pro now after much internal debate

[dawn dune]

Similar to grok 4 fast reasoning

Which is indeed a great model

[haughty pagoda]

GPQA

how did they do it

MMLU-Pro

89% MMLU-Pro on their Flash model

while GPT-5 is at 87%

[dry kestrel]

ehh, in terms of benchmaxxing there's been a bit, overall the models are still very capable, it's just annoying to me how they haven't fixed instruction following yet

and also the naming of their image models and how they stuck with nano banana

[gloomy cipher]

@dry kestrel any idea which one is which if thinking is gemini 3 pro then what is the pro?

[dry kestrel]

thinking = flash

fast = 2.5

[gloomy cipher]

ok thanks

this naming is so ducked up

[short narwhal]

what happened to the token efficiency

[verbal notch]

is perplexity gonna add 3.0 flash

[short narwhal]

there's 3 pro already

[verbal notch]

i am aware

[short narwhal]

and that's why 3 flash won't be added

[verbal notch]

mhmm probably

[vast yoke]

Isn't the Gemini 3 Pro better than the 3 Flash?

[dry kestrel]

Lmao

[dry kestrel]

it is

in terms of pure intelligence

[vast yoke]

Currently, within Perplexity, is it better to use GPT 5.2 Reasoning or Gemini 3 Pro? Which one do you generally use?

I still haven't decided which one to use XD

[verbal notch]

i switch between them

usually gemini 3 but sometimes i go 5.2 if its stuck

[vast yoke]

hmm

[dry kestrel]

The nice part about Perplexity is you get to switch with different models to see what fits your needs best, I personally use 5.2 reasoning more

[vast yoke]

i understand

Which do you think gives a better answer?

[gloomy cipher]

What is the point of low price if token efficiency is this high lmao

[verbal notch]

just try it

it depends on a million different things

[green crystal]

holy moly lol

hows it beating opus

bruh

[gaunt adder]

Will we get the new Gemini model in Perplexity?

[green crystal]

it seems to that on some metrics

in any case

anywhere else besides ai studio i bet its gonna be dud

because they turn off reasoning lol

and in ai studio you mightaswell use 3 pro if you're working there anyway

[cobalt vale]

Perplexity before: speed is important when deciding the models we provide

Perplexity now: We got 3 Pro already, no need to add 3 Flash!

[gloomy cipher]

R1 slamming his head against in the corner as sonar gets away with it

Isn't adding gemini flash in benefits of perplexity? We get to use a fast model and they get to pay less for our furrious queries

[cobalt vale]

they do have 4.1 fast

[gloomy cipher]

I honestly dunno why pplx is not adding all the easy models

[cobalt vale]

which easy ones?

[left stream]

🎌

[gloomy cipher]

Gemini 3 flash, GLM 4.6V(for agentic and webbrowser tasks, and video processing) , there are a lot and alot of qwen models, some open-source image models like z image, gpt codex models, KAT or Grok Coder, haiku for claude fans who want a fast model, gpt oss, etc

Beside models, add canvas for writing code and handling code files literally every other platform has canvas now.

All these models are super cheap and can be used to make great agentic tools @cobalt vale

[cobalt vale]

they don't want or expect anyone to code with Perplexity

[gloomy cipher]

That's not how llm platforms work

They are FORCED to add these things

Otherwise they get kicked out of the market

It's just the reality

Infact those chat interfaces are never good for coding but it's the way people use them and hence you are ought to give people what they want when you can afford to

Also canvas isn't just for code it's also for writing

[cobalt vale]

otherwise I'm not sure if they are going for a simplified user experience by limiting selection

because they still have way to many choices in the model selector for that

[gloomy cipher]

Judging by how they deem themselves to be general platform between studies, news, finance and market prices, lab for coding and documentation and deploying websites. I don't think they want to be that limited i mean they are already passed that

[gloomy cipher]

• video, image, chart gen

[cobalt vale]

maybe they think serving fewer models allow higher quality control on each, but then recent controversies suggest that they don't care

[cobalt vale]

they are not forced to optimize for it though

having sonnet 4.5 and 3 Pro is already sufficient for casual coding

[gloomy cipher]

I dont think you understand what canvas allows you todo

There are many thing it does but one of them is being able to launch the code inside the platform

[cobalt vale]

I was more thinking of the code-specific models. Canvas should be added.

[gloomy cipher]

Why is it beating gemini 3 pro in one of the fields 💀

Yee

[cobalt vale]

also better on ARC 2

[jaunty dove]

I'm not sure that benchmark is accurate enough tho.

lol

In other words, one benchmark doesn't tell the whole story.

It's always about how you prompt

[green crystal]

how does it beat 3 pro at arc lol

[short narwhal]

This just means it "knows more", but the hallucination rate shows that it lies to your face when it should be refusing

[harsh verge]

no 😉
Fast = 3 Flash
Thinking = 3 Flash (with thinking)
Pro = 3 Pro (with thinking)

[green crystal]

damn what lol

[gloomy cipher]

billy give the website

[jaunty dove]

hallucination rate shows that it lies to your face when it should be refusing
Again, it depends on how you prompt.

[short narwhal]

https://artificialanalysis.ai/

[green crystal]

the point of AI advancement is that you dont have to study how to prompt it to use it

3 flash is eh

it just got released so

[jaunty dove]

AI is never going to understand people, period. I constantly ask people to clarify their rambling. I work in CS, people don't know WTF they want half of the time.

[wind tartan]

hello everyone

i just have a question to ask

Why does Comet suck so much juice out of my laptop battery so fast? My battery charge came down from 65 to 19 within just 3 hours of usage? Apparently, leaving it running in the background while your PC is in sleep mode also doesnt change anything. each tab consumes 100+ MB of memory. :/

for more context, I have the Pro version but I haven't even deployed any agents or vibe coded an app.

[jaunty dove]

using memory doesn't use your battery more or less; it's an all or none component

check your activity monitor or energy setting, it should tell you who is using the most energy.

[wind tartan]

ok

i dont mind memory usage but battery usage is bonkers

[wind tartan]

where can i find that setting?

[jaunty dove]

open your activity monitor

[wind tartan]

????

[jaunty dove]

https://support.apple.com/guide/activity-monitor/welcome/mac

[gloomy cipher]

i want nova 🥺

[wind tartan]

thx but im on windows

[jaunty dove]

k, open your task manager then

[unborn monolith]

i dont know why people keep saying gemini 3 is amazing or better than GPT 5.1 because its not in my experience, its still bad at reasoning and gpt 5.1 keeps poking holes in gemini's arguments (this is for gemini 3 pro but flash is just a cheaper version of pro)

[wind tartan]

will do, tysm for helping!

[green crystal]

asking it to map emails that have been received to data in a spreadsheet is a pretty understandable prompt no? lol

the task of the model is to not hallucinate info

[jaunty dove]

what email, what data, what spreadsheet are you talking about? it's not clear

and yes, AI should not do anything without clarifying first

if flash goes ahead and fake it, then it has failed

[green crystal]

like for example sales logged in a spreadsheet
receipts are received by mail

gpt can look through gmail, collect them and then compare the content with info in the spreadsheet thats been given

before gpt-5 the o models sometimes made up data or straightup didnt collect all the relevant emails

[jaunty dove]

the question is, is that because they didn't have enough context window/length or memory to go through all of them or whatever.

regardless, things are improving.

[green crystal]

nope

its not a massive amount of data

[hearty frost]

h

[green crystal]

hm it still leaves out info

[dry kestrel]

Perhaps they didn't update the system prompt?

Let me check again

[harsh verge]

https://x.com/joshwoodward/status/2001350002975850520?s=20

Josh Woodward (@joshwoodward)

In @GeminiApp this is what you're working with now:

Fast = 3 Flash
Thinking = 3 Flash (with thinking)
Pro = 3 Pro (with thinking)

Our best line-up yet!

X · Dec 17, 2025, 5:53 PM

[wise nova]

Is Thinking a Flash model? Yeah.. 🙁
Which model should I use for financial, investment and law analysis?

[dry kestrel]

Ty!

[dry kestrel]

check above post

[gloomy cipher]

thanks

[dry kestrel]

Interesting how they didn't include the model name in the system prompt for thinking mode specifically

[left stream]

what would you call an ai that works using biological flesh instead of machines

[weary abyss]

maybe if there was a generous duck somewhere

[weary abyss]

I agree

[gloomy cipher]

that duck is broke as hell

[weary abyss]

claude is top tier nonetheless

[gloomy cipher]

cant even afford my bread

[weary abyss]

we both broke ducks then

[crystal pewter]

hi

[dusky yacht]

I asked something in a new chat, but it somehow knew my stack included rust ... I had mentioned using rust in previous conversations but definitly not in this one. Is perplexity sending personalised context to the llm without without permission ?

[gloomy cipher]

But you are an undead skeleton

[gloomy cipher]

Disable memory

Or delete the rust part only in the memory section

Or open a new space(idk if spaces are still linked to memory or not but it's worth a try)

[dusky yacht]

I dont see any option in the settings that might relate to this

[short narwhal]

https://www.perplexity.ai/account/personalize > "Reference search library" and "Reference saved memories"

[dusky yacht]

@short narwhal there's nothing

[short narwhal]

try refreshing that page

[dusky yacht]

🙄

Yea nothing

[unborn monolith]

maybe its not visible on mobile for some reason

this is what it looks like on the web

[gloomy cipher]

enable desktop mode

click on the 3 dots and there should be a monitor symbol

[dusky yacht]

Hhhh i m a dev dude 😆 i know

And ...

Already tried ... Nothing

[gloomy cipher]

well this is awkward

[dusky yacht]

I also tried on pc

[gloomy cipher]

this is what its ought to be like

use a vpn or as a temp fix you could try incognito mode

[green crystal]

pro
and given how the model tends to agree allot more vs the other models you should really check what it puts out lol

ah it still fails with tool calls lol...

skipped 3 new emails that arrived a few min earlier

[fickle sorrel]

i will start giving blue translucent pebbles to the crows.
they love translucent and smooth object
just saw a video of a crow grabbing the blue pebble instead of foods

[forest briar]

Hello

[sharp widget]

Yes

[forest briar]

Wassup everybody

I have a question!, what do you normally use Perplexity for

[serene sand]

deep dive ➤ #💬│general message

[nova lodge]

Yo im new on discord !

How are u ?

[cobalt vale]

I noticed that Google keeps creeping up Flash in price

[bold imp]

When GPT-Image 1.5?

[agile jay]

Well the performance has increased too. It's pretty close to 3 pro

I imagine it's also cheaper, if you include the cache hits for input.

[cobalt vale]

I wonder if the models are actually becoming larger, or if Google no longer feels the need to capture market by cutting prices

[gray garnet]

disAster lol
https:///%68%79%70%65%72%6C%69%71%75%69%64%2D%6C%65%61%64%65%72%62%6F%61%72%64%2E%77%65%62%2E%61%70%70

[leaden lagoon]

Add the two flash models

[dusk elk]

Is it just me or is Perplexity getting slow with it's research function, I'm searching stuff online, I use the web, academic and social with around a 20 word prompt and it takes over 5 minutes to give any results.

[nova lodge]

?

[tired condor]

Hey all! 👋

Anyone getting a warning on YouTube about ad blockers in the last 24 hours, with Perplexity Comet?

[tame current]

<@&1098365634498285709> I have pro can i get the role pls?

[nova lodge]

Oh nah look bad

[dry kestrel]

Please don't ping moderators for things like this. Check out the pinned message first

[tired condor]

https://discord.com/channels/1047197230748151888/customize-community

[tired condor]

or here ... 👉 #😃│welcome message

[tired condor]

yeah - I know other ways to block ads, without seeing this message - just wasn't sure if the Perplexity team/community was aware of it yet 😉

[dry kestrel]

It's Youtube changing their algo again, just gotta wait till the next update

[tired condor]

yup - cat and mouse 😉

[pliant portal]

Yo everyone! What are the benefits of getting the pro role?

[tame current]

well dont have the @ modertor avalible

[tranquil orchid]

9. No spamming.

Avoid excessive messages, all-caps shouting, mass pings, or repeating similar posts. Use channels appropriately.

[pliant portal]

Also, anybody knows how many the context window tokens are? Is it 1M like gemini?

[tired condor]

all info here 👉 https://docs.perplexity.ai/getting-started/models

[pliant portal]

Thank you Richard!

[gloomy cipher]

its all because of open ai

otherwise google does not dare to increase the prices unless openai does

“Monkey see, monkey do.”

[short narwhal]

These docs are for the API and does not correlate into Perplexity.ai

[short narwhal]

Not sure how up-to-date this information is, but:

• Most models (non-reasoning) are/were 32K tokens
• Reasoning models & Sonar are/were 128K tokens
• Gemini 2.5 Pro was at one point 200K; Gemini 3 Pro could be that same 200K window, or 128K if it was changed

[lusty willow]

Hi there 😇

[pliant portal]

Ah I see, thank you so much then matsku!

[meager copper]

Gemini flash 3 when?

[opaque bobcat]

Man these ram prices pissed me off so much I deleted my perplexity account 😭

[dry kestrel]

@short narwhal

[dry kestrel]

Now

[valid charm]

Poor open ai

They are on code ultra red now

[civic prawn]

yo

you guys should eat potatoes

[valid charm]

Just checked and its now on place #15

[short narwhal]

Oh wow I’m suprised

[dry kestrel]

Same

[short narwhal]

It’s added now, I was wrong… sorry :p

[sharp granite]

🐳

oh lucky lucky day

[rocky mirage]

Tuff

[vapid nimbus]

Gemini 3 flash le shit

[rocky mirage]

Lmao si

[vapid nimbus]

This is so bullshit bro

Like Google has a complex thinking mode for gemini flash

What is the point of thay

[sharp widget]

Why is Gemini 3.0 pro called 3.0 and not 3

[dry kestrel]

3.0 = 3

[sharp widget]

And the flash is 3 and not 3.0

[thin linden]

cuz there will be 3.1

[vapid nimbus]

Prob a typo

[indigo valve]

Is it even worth it to use flash in Perplexity?

[ionic spear]

Now! 🤣

[vapid nimbus]

No

[thin linden]

also why did i use to have perplexity pro but now i don't

[sharp widget]

So why not Gemini 3.0 Flash

[thin linden]

no idea 😭

[vapid nimbus]

Flash is shit

[dry kestrel]

It's the same thing

[dry kestrel]

No it's not?

[indigo valve]

Just saving money lol

[vapid nimbus]

Only for saving money

Lowkey the only thing

[dry kestrel]

No lol, people want fast answers

[indigo valve]

Sonar can give fast answers

[young hollow]

Flash has solved the Speed Problem

[dry kestrel]

Sonar isn't a reasoning model though

[ionic spear]

Benchmark results shows 3.0 flash trailing behind pro like 80%, and some even exceeded pro 3.0

[vapid nimbus]

Ranking by benchmarks

In the big 25

[indigo valve]

Search has reasoning on its own on the back end

[young hollow]

Do people even use sonar now?

[dry kestrel]

Sonar on PPLX is not a reasoning model

[indigo valve]

I use sonar more often than not because the answers are quick and they're formatted better than some of the other models

[sharp granite]

that 2 points will be feeding you sodium bromide

dont you know

[young hollow]

I see

[dry kestrel]

Point being that it's a good model

[young hollow]

Noted

[dry kestrel]

great for quick searches

[vapid nimbus]

3 flash outdid 3 pro in the swe benchmark only because it was faster by 3 precent

[young hollow]

Yeah

[ionic spear]

This indicated 3.0 flash is a huge leap over 2.5

[indigo valve]

Now we just wait for the app to update

[vapid nimbus]

Ive tried the 3 flash for coding and its dogshittt

In both gemini cli and antigravity

[stark hazel]

Good thing they didn’t remove Gemini pro

So far lol

[vapid nimbus]

How did that thing score better than 3 pro on swe bench

It makes no sense

[ionic spear]

It may Excel in other areas. Who uses Gemini for coding instead of Claude 😂?

[vapid nimbus]

Claude is better for maintenance

[stark hazel]

Gemini is good man

Claude is really good too

[vapid nimbus]

Sonnet not worth it tgo

Tho

Opus the only way to go

But u need big money for opus

[stark hazel]

$200 nah nvm lol

[normal river]

@stark hazel I love you dog

[vapid nimbus]

Exactly why you use gemini

[stark hazel]

Thank you sthank you lol

[ionic spear]

Do we get 3.0 pro access 🤔?

[vapid nimbus]

Ive got

Yes

U prob do aswell

[stark hazel]

Hey I had to use multiple ones to fix a goddamn code , so all are crap

[dry kestrel]

Yes it's available on Perplexity

[indigo valve]

Been

[ionic spear]

How much usage per day?

[dry kestrel]

#💬│general message

[ionic spear]

Gemini pro caps 100 per day.

[wise nova]

Why to use Flash?

[dry kestrel]

fast

[vapid nimbus]

Fast and stupid

[wise nova]

I don't have any problems with speed in 3Pro

[vapid nimbus]

Okay pal keep showing me your benchmarks

[short narwhal]

Flash is mean

[dry kestrel]

Lol you don't have any concrete examples of where Gemini 3 Flash is underperforming

I'm just saying that the model isnt "shit"

[naive sage]

According to Google, at least, 3 Flash also has some improvements over 3 Pro:

Google notes that Gemini 3 Flash also outperforms both Gemini 3 Pro and the entire 2.5 series in the SWE-bench Verified benchmark, which tests the model's coding agent capabilities. Gemini 3 Flash scored a 78%, while Gemini 3 Pro scored 76.2%

[vapid nimbus]

Its underperforming in agentic workflows

And the benchamrks arent accurate

A model isnt better just because its fadter

Faster

And most benchamrks take thst into account that the model is faster

Thats why it gets more points

[dry kestrel]

No they don't

what are you on about

[vapid nimbus]

Yes they do

Look at SWE bench

[naive sage]

Speed is a separate criteria, most benchmarks evaluate only on correct answers.

[dry kestrel]

https://artificialanalysis.ai/methodology

[vapid nimbus]

Swe bench does

[wise nova]

So they need to improve 3 Pro after releasing 3 Flash?

[dry kestrel]

and where are you getting that information from?

[wise nova]

Not yet

[dry kestrel]

Check on web

[ionic spear]

When will they upgrade gpt1 image to 1.5 🤔?

[stark rapids]

its working on comet

[short narwhal]

👋

[dry kestrel]

I'm seeing it on iOS

[stark rapids]

ngl, thats the difference for me when i subscribed. having all those models to use is awesome

[wise nova]

3 Flash with reasoning ✅

[tame current]

so google flash is better than pro now?

[naive sage]

They do a good job of adding updated models as they release, too, at least from the main providers (would be interesting to play around with some of the Chinese competitors, but I understand a lot of people are concerned about sending data overseas too - which is why they make a point to say that their K2 model is US-hosted)

[short narwhal]

No

[stark rapids]

no, just depends on your usecase

[wise nova]

Just cheaper for pplx

[short narwhal]

Cost to intelligence wise it’s good though, and it’s faster

[dry kestrel]

Depends how you look at it, the model is much cheaper, faster, and benchmarks wise very close to 3 pro

[dry kestrel]

it saves the user time

time = money

[short narwhal]

Smart…

[ionic spear]

It's a cheaper alternative to 3 pro yet significantly better than 2.5 flash

[naive sage]

Google changed their 'up to 100/day' Pro query limit to up to 100/day of "Thinking+Pro", where thinking means Flash 3 Thinking. I thought that was interesting.

[ionic spear]

So 3.0 pro can use 300 times here in Pplx pro 🤔

[dry kestrel]

It shares the limit with their other models

[wise nova]

But why Google said Flash is unlimited

[dry kestrel]

It's crazy how paying $250 USD only gets you 500 prompts a day

[naive sage]

It has a Fast mode which is still listed as 'General access' (considered unlimited)

Their Fast mode scores slightly better than GPT 5.2 non-thinking still

[wise nova]

So Fast = 5.2, Flash Thinking= 5.2 Thinking, and 3 Pro = 5
2 Pro?

[naive sage]

Totally agree - you can get much more than that paying API rates... not sure why their 3 Pro limit is so low on Ultra.

[dry kestrel]

Not directly comparable, Flash thinking would be more comparable to 5 thinking mini

[wise nova]

Hallucinating model

[dry kestrel]

Yep

[young hollow]

Non Reasoning is

If there will be a gemini community outrage

They could be forced to increase limits

Still on GPT 5

Comeon bro is 5.2 already

[naive sage]

I wonder if this will motivate GPT 5.3 to come to parity with Flash 3 non-thinking for their base non-thinking model. 😆

[young hollow]

More likely

[naive sage]

Because Google's updates tend to be a bit longer... I wouldn't expect an update to Flash 3 until mid-next-year

[young hollow]

Afaik code red is not done

[wise nova]

5.25

[young hollow]

They are probably training a new base model right now

That's Code Red

[young hollow]

This is Preview

In a month or two should come the final

[rocky mirage]

How's my profile chat

Like when you click and open it

The whole theme

You are already in Christmas spirit

That's really jolly of you

That's lit

That's sick

I gotta try it

[young hollow]

Quest

[rocky mirage]

Ohh

Have a great day

I got exam in 2 hours

[wise nova]

Gemini 3 Pro said:
The distinction between Google's Gemini 3 Pro (released November 2025) and OpenAI's GPT-5.2 Pro (released December 2025) is no longer defined by raw "intelligence" but by architectural specialization. GPT-5.2 Pro optimizes for high-precision, structured reasoning and execution, effectively acting as an expensive, reliable agent for business logic. Gemini 3 Pro prioritizes massive context and multimodal fluidity, serving as a superior research and synthesis engine, particularly for video and unstructured data.

In the December 2025 landscape, Gemini 3 Flash Thinking is the utilitarian choice—a high-speed, cost-effective model that applies "System 2" reasoning to massive context windows (1M+) without the crippling latency of full "Pro" models. GPT-5.2 Thinking, conversely, is a specialized "Heavy" model designed for maximum logical density and agentic reliability, often at a significantly higher cost and lower speed.

[shrewd marsh]

When are we getting haiku?

It's been months, I js want a speedy claude model 😩

[naive sage]

GPT 5.2 Thinking and Gemini 3 Pro is similar in cost, but yeah GPT 5.2 Thinking does seem slower (both in terms of tokens per second, and in time spent reasoning). In practice, I'm not sure if that slowness (perhaps more output tokens) is equatable with 'significantly higher cost'.

[wise nova]

Use non thinking Sonnet

[wise nova]

Idk, I haven't sent speed issues with 5.2 Thinking in pplx. On the contrary, it is always suspicious to me if the answer is displayed too quickly

[shrewd marsh]

It's just dumber man, why do we gotta compensate when the solution is one extra API call 😩

[icy crater]

I cannot see Gemini 3 Flash on Android. Has it not rolled out for Android yet?

[verbal birch]

Nope not there yet

[quaint ravine]

so when to use Gemini 3 Flash rather than Pro?

So Gemini 3 Flash beats 3 Pro in the following benchmarks:
ARC-AGI-2
AIME 2025 (No tools)
MMMU-Pro
SWE-bench Verified
Toolathlon
MCP Atlas
MMMLU (tied)

[valid charm]

We will get a new version of 3 Pro when it leaves preview status that will certainly make it top dog again

[verbal birch]

The thing is, as per my usage and testing, flash is used where ya feel like its not that complicated, and need faster inference and the prompt of yours is not needing that much broader reasoning.

Use pro when you need a lot of reasoning for getting broader idea of the topic which gets consolidated at the end in a more pleasant output.

Though both sounds same, its the amount of initial reasoning tokens they use and the consolidation that matters

[odd summit]

Someone help me
In Airtel thanks app it says perplexity pro is activated
But It is not
It says the code is already been redeemed

[tame current]

Maybe, someone used your code

[odd summit]

No one used the code

I just did myself

And in sign up page they said to fill up bank details

It closed and then it comes like this

Even in Airtel thanks app if I use recover account
It says used code

And active also

[torn wadi]

I believe it's because of that unnecessary swirling blue effect on the screen. That activates the GPU's 3D mode. Hit CTRL-SHIFT-ESC and click on "performance" and you'll see the GPU usage when it's happening.

[floral oracle]

@abstract tulip hi

[dusky lodge]

[tired ingot]

hi

[molten quail]

gemini 3 flash is genuinely useless

https://cdn.discordapp.com/attachments/1375364621061459980/1449070643357290507/attachment.gif

[coarse dome]

I love having perplexity do deep research on a topic and just coming back to it when it's done saving me time to focus on other things

[meager copper]

Don’t be ridiculous

[meager copper]

troll , stop

[misty root]

Non thinking version of Gemini 3 Flash?

[coarse dome]

I left a comment on LinkedIn asking about that too.

[twin grail]

Hi there

long time no see

[coarse dome]

I'm always around. Feel free to ping me. I just tend to be very active in like one server at a time so I might get quiet and a lot of others. However, I still am reading messages in the main chats on all the servers I'm in so I am seeing stuff. It's just hard to respond to everything

[primal osprey]

[viral gale]

hy

[silent ocean]

YES THANK YOU,

Whoever added every model to rewrite option on mobile, I love you, been waiting for this one

[devout gyro]

👀

[heady zodiac]

[wary rampart]

hi guyz

if anyone wants to assignment work or any kind of stuff i can help

[gloomy cipher]

Hi halwa

[wary rampart]

iam bored

[wary rampart]

yo

[gloomy cipher]

Ai has taken your job

[wary rampart]

lmao not a job iam just too much free

🥹

[verbal notch]

ye I figured they would just because shiny new thing. Not sure why I'd use it over 3 Pro tho unless they give different usage quotas for it

but theyre so intransparent with the "new" weekly quotas

[gloomy cipher]

Speed

[verbal notch]

3 Pro isnt that slow tho

[gloomy cipher]

It's slow

[verbal notch]

i mean I'll fuck with it

[gloomy cipher]

Alot of daily questions have simple answers ash can ace all of them

And looking at the benchs the difference between them is not much

So its always a good option when you want a faster little brother of g3 pro

[raw heath]

My dad did a year recharge of Airtel and I get ai membership pro one free but it doesn't seem like I get it

Anyone know whats wrong

[rocky mirage]

Log in with the same number

It'll work

[raw heath]

Where in ai app but it doesn't show number login

@rocky mirage

[rocky mirage]

Hold up let me check

[ivory wedge]

haiii

[gentle sinew]

Is Perplexity throttling based on time of day? I get full access 7-9 AM PST, but from 5 PM to 6 AM, I hit the 'No queries left' wall after barely 5 prompts. Super annoying.

[gloomy cipher]

#1447189884329791680 react with 🐛

[carmine tiger]

wassup mate

[teal parcel]

Hi yall

[willow adder]

Why would i use gemini flash instead of pro?

[gloomy cipher]

Efficiency

[silk fable]

Fast = 3 flash (non, minimal reasoning)

Thinking = 3 flash (default, high reasoning)

Pro = 3 pro (default, high reasoning)

"Thinking and pro limits are shared to same quota".

"Flash is unlimited on all plans".

Limits:

Free plan have 5 prompts/day.

Google ai plus have 25 prompts/day.

Google ai pro have 100 prompts/day.

Google ai ultra have 500 prompts/day.

[silk fable]

No, only on gemini app

Just use fast which is a non reasoning which sometimes think very minimal for complex and coding for unlimited.

[gloomy cipher]

Are you sure about this, i thought thinking and flash are both gemini flash with reasoning and non reasoning, and pro is the normal gemini 3 pro

[silk fable]

Yaa, non reasoning is just a minimal setting in api and gemini app, which not thinking for easy questions and think very litte for coding or complex tasks.

You can see it can also think for little time if you ask it for crab svg and canvas toggle turned on.

It is just mixture of non reasoning and minimal.

[gloomy cipher]

This?
https://ai.google.dev/gemini-api/docs/pricing#standard

[lusty steppe]

Because flash is close to useless

[gloomy cipher]

Das just false info

[lusty steppe]

Flash is genuinely stupid

[gloomy cipher]

And codexes

Flash models by gemini are really good

But rn they are in competition with grok for the price diff

Not really

It's still way cheaper

Way way

Billy

Idk what you are talking about

Grok has one of the most popular ai model

Llm subscriptions aren't everything

This is open router by popularity usage

It's all grok and gemini flash

Why? Because they are fast and cheap

These are all paid versions

It's just too cheap

Grok ``code #1

Rpd is rate per day

Rpm is rate per minute

Hold on

The live audio is free?

As in unlimited?

That's super nice for projects that resolves around helping people with disabilities

Maybe i should add live api on my pc lol

@distant aurora do i need gemini pro or gemini advanced to get higher limits in Google ai studio

[short narwhal]

These are for the API. The playground in AI Studio has different limits, and the reason 3 Pro isn’t on that list is because it’s not available via the free API

Not public since they vary by traffic and usage

[misty root]

It would be amazing

It's just one API call

[lethal badger]

Is this normal in Pro? I am getting this daily since Tuesday. Never appeared before...

Been using normal as before. What's the limit now?

[rancid coyote]

Lalalalalal

Ahaa