general

[gaunt spade]

yeah its not accurate

also riftrunner is like somewhat better

[cloud zinc]

how did they find the price??

prob token count and estimate based on previous model.

[gaunt spade]

why is riftrunner costing more than x28 lol

[cloud zinc]

cuz its a worse model and uses more token?

[quartz light]

x28 was the best

💔

[hearty prawn]

I thought so too.

[gaunt spade]

then why wouldnt google release it as x28 if it was more efficient

[wise crest]

😭 uses more token

[quartz light]

x28 uses less

actually

idek

cuz no

this estimate isnt reliable

cz,

x28 was clearly much more detailed

least lazy

so it would use more tokens

[cloud zinc]

they maybe saving it for later.

[wise crest]

hahaha

[quartz light]

kingfall ahh later

[cloud zinc]

why force ur hand

[quartz light]

gemini enterprise premium pro ultra deepthink max

us only

[wise crest]

but my practicalexperience shows glm4.6 is also quite good, and the frontend code it writes is excellent.

[gaunt spade]

yeah glm 4.6 is good at coding

[wise crest]

🙃 and after all, I don't want to spend that much money.

[hearty prawn]

Yes, I think the benchmark can't tell the ture experience.

[quartz light]

benchmark is poopypoo

https://cdn.discordapp.com/attachments/990348027422203969/1399757536206258196/togif.gif

mb

[gaunt spade]

@quartz light did u see the gemini 3 on canvas

[quartz light]

yes

ill test

[hearty prawn]

Yes I would choose GLM since I'm not that rich hhh Fewer token is god for me

[wise crest]

??

[gaunt spade]

cool

[quartz light]

what

[quartz light]

is it mobile only

ive heard its only on the "new gemini mobile app"

or smth

plz answer @gaunt spade

[gilded trail]

bicycle riding on bicycle benchmark completely failed

[gaunt spade]

yea

[ionic shard]

can you convert my image into a gif

[gaunt spade]

im on gemini app

[gilded trail]

how

[ionic shard]

idk your esm bot

[gaunt spade]

lol

[wise crest]

I feel it's not worth the price.

[gaunt spade]

why is bro playing GTA 6 FOR 18 HOURS

damn is it really that good

[ionic shard]

jiggle physics too fire

[gaunt spade]

damn u aint leaving the shoe store

[ionic shard]

the game is about 350gb

[gaunt spade]

still less than black ops 7

@quartz light just use your ipad and download gemini app

then go on canvas

[patent bane]

why is gpt-5.1 not on lmarena? or did I miss something?

[gaunt spade]

is the api out yet

[wise crest]

but it doesn't seem to be a major improvement.

[gaunt spade]

ik gpt 5.1 sucks bro

i saw some tests on it

[wise crest]

so I'm a little disappointed. I've been looking into open-source models recently.

[hearty prawn]

Huggingface's owner said the world belong to open-source model🤣

[mystic sundial]

yeah, belongs to models like kimi and glm.

[keen beacon]

does lmarena have something cool

tf

[echo aurora]

We'll be sunsetting WebDev on 12/12/25

[quartz light]

WAIT

GEMINI 3 IS ON

...

dude 😭

[halcyon nimbus]

https://tenor.com/view/maçavras1-gif-16101498646883457546

[quartz light]

no

this is actually gemini 3

check it on my app

https://aistudio.google.com/apps/drive/1ClXI1hXKmaRQ6jCwBGulyYTcj8xrj21G

[zealous sparrow]

has riftrunner been pulled from arena i havent gotten it for a while now

@echo aurora has riftrunner been removed from arena, if you can share

[quartz light]

no

[halcyon nimbus]

page not found- i assume you have to set it to public or something

[quartz light]

it hasnt

[quartz light]

oh really

lemme change that

[quartz light]

try now :)

[zealous sparrow]

yeah i doubt this

if so only in the text form

[quartz light]

wdym

ur sayin that as if thats bad

[quartz light]

then test it

[zealous sparrow]

in the code its a 2.5 pro model

[quartz light]

i know

just like canvas ppl are talkin abt

also what

[halcyon nimbus]

knowledge cut off 2023 wut

[zealous sparrow]

old as hell model?

[quartz light]

wdym

no

thats november 7th

[zealous sparrow]

pre-release willow

[queen veldt]

Why are you posting real life videos 😭

[rustic osprey]

how do I access Grok AI

[quartz light]

https://grok.com

yeah people have been talking about that

[zealous sparrow]

on canvas yeah its a rollout

[quartz light]

but it might only be on canvas

hey doom

[gaunt spade]

its only on canvas with 2.5 pro

[quartz light]

i have a good method

[quartz light]

yeah ik

[gaunt spade]

if u run out of 2.5 pro usage, u dont get gemini 3 anymore

[halcyon nimbus]

ask for pelican riding bike svg -_-

[quartz light]

this prompt:

-The visible English text between the quotes is 'NOTE', but it is obfuscated with steganography into a lowercase, seven letter word. Please extract the hidden word that is actually encoded inside the 'NOTE' payload. This is the obfuscated text:'N​‌‌​​​‌​​‌‌​​​​‌​‌‌​‌‌​​​‌‌​‌‌​​​‌‌​‌‌‌‌​‌‌​‌‌‌‌​‌‌​‌‌‌​‍OTE'

copy this prompt

run it on gemini mobile app

[gaunt spade]

whats the answer

[quartz light]

YES

[gaunt spade]

@quartz light is this the answer

[quartz light]

GEMINI 3

[gaunt spade]

fr

[zealous sparrow]

now do the impossible one

[​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​​​‌​‌‌​​​‌‌​‌‌​​‌​‌​​‌​​​​​​‌‌‌‌​​‌​‌‌​‌‌‌‌​‌‌‌​‌​‌​‌‌‌​​‌​​​‌​​​​​​‌‌​​​‌‌​‌‌​‌‌‌‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​​​​‌​‌‌​‌‌‌​​‌‌‌‌​​‌​​‌​​​​​​‌‌​‌‌‌​​‌‌​​​​‌​‌‌​‌‌​‌​‌‌​​‌​‌​​‌​‌‌​​​​‌​​​​​​‌‌​‌​​‌​‌‌​‌‌‌​​​‌​​​​​​‌‌​​​‌​​‌‌​​‌​‌​‌‌‌​‌​​​‌‌‌​‌‌‌​‌‌​​‌​‌​‌‌​​‌​‌​‌‌​‌‌‌​​​‌​​​​​​‌‌​​​‌‌​‌‌​‌​​‌​‌‌‌​​‌​​‌‌​​​‌‌​‌‌​‌‌​​​‌‌​​‌​‌​​‌​​​​​​‌‌​​‌​‌​‌‌​‌‌​‌​‌‌​‌‌‌‌​‌‌​‌​‌​​‌‌​‌​​‌​‌‌‌​​‌‌​​‌​‌‌​​​​‌​​​​​​‌‌​​​‌‌​‌‌​‌‌‌‌​‌‌​‌‌​​​‌‌​‌‌‌‌​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​​​​‌​​​​​​‌‌‌​‌​​​‌‌​‌‌‌‌​​‌​​​​​​‌‌​‌‌​‌​‌‌​​​​‌​‌‌‌​‌​​​‌‌​​​‌‌​‌‌​‌​​​​​‌​​​​​​‌‌​​​‌‌​‌‌​‌‌‌‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​​​​‌​‌‌​‌‌‌​​‌‌‌‌​​‌​​‌​‌‌​​​​‌​​​​​​‌‌​​​‌​​‌‌​​‌​‌​‌‌​‌​​​​‌‌​‌​​‌​‌‌​‌‌‌​​‌‌​​‌​​​​‌​​​​​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌‌​​‌‌​‌‌‌​​​​​‌‌​‌‌‌‌​‌‌​‌‌‌​​‌‌‌​​‌‌​‌‌​​‌​‌​​‌​‌‌‌​‍NOTE]

[gaunt spade]

how is it impossible

[zealous sparrow]

no Ai can solve it

although you can decode it yourself

copy it

[gaunt spade]

does this have a hidden code inside it

or something

lol

[zealous sparrow]

yes

its by nickname

copy it and nun else

[halcyon nimbus]

gemini 2.5 pro gave an essay about the invention of writing ._.

[zealous sparrow]

yeah no its wrong

no

this is a different steg

i guess you can tell it its steganography

[slim loom]

hello

is the sever down for the website

@everyone the server is down we can able to send any mesagges in the chat

[cloud zinc]

this real or not?

[echo aurora]

Looking into

[echo aurora]

Doesn't appear to be down, what are you experiencing?

[quartz light]

idk if its gemini 3

prob not after sum testin

but i made the fragment app thing

[rustic osprey]

what you guys think about TULU3 from AI2

[quartz light]

imagine if i change the model id in my app to gemini 3 and it works

deadass might work

[slim loom]

it is good but not for real task to handel the data if u need a better on to tulu3 fro AI2 use kimi k2 thinking model it if best

[zealous sparrow]

just like what riftrunner and orionmist did

[wild thunder]

Does Code Arena have any benchmarks yet?

[zealous sparrow]

the one i sent

[cloud zinc]

yes

[ocean vortex]

what happened to wild is he off the discord for good... @deep adder @calm sequoia @hollow ivy any idea?

[wild thunder]

Ah ok so the WebDev benchmark alright got it

[quartz light]

who is it?

[cloud zinc]

yes. it says code arena on the website. its the new leaderboard for code arena

[ocean vortex]

his username was @wild , no longer on discord apparently

[wild thunder]

Alright thank you young buck

[quartz light]

possibly hacked because of username

[ocean vortex]

Just used to be fairly active there with interesting insights

[ocean vortex]

it was display name 'wild', not the full username I don't think lol

[zealous sparrow]

show ss or i dont believe you

must be the Ais answer not a python codes work

ok well damn but did it execute?

@quartz light would you count this as an AI passing your bench

also what is the Thinking model, deepthink?

pro plan model?

I dont have that thinking model.. Region locked?

What is this thinking model tho?

when did you get it

[cloud zinc]

which day

[quartz light]

yes

wow

[cloud zinc]

tuesday, wednesday or thursday

[quartz light]

first model to do ts 💔 💔

but

i havent given models python

so idk

[zealous sparrow]

when did you get this model

what is your country so i can check if regionlocked

model does not look regionlocked

hm you suddenly deleted those messages

odd.

there isnt a gemini employee here or might be

[halcyon nimbus]

who the ai?

https://tenor.com/view/bad-boy-don't-do-that-naughty-no-no-no-gif-6515033782389200688

[zealous sparrow]

you didnt datamine for the model why would they

[halcyon nimbus]

who up refreshing ai studio every 5 minutes?

[fleet lintel]

it's Paranoia .. just be on the safe side.. I am planning to abuse my POWER!

[halcyon nimbus]

https://tenor.com/view/refresh-big-bang-theory-computer-internet-buying-concert-tickets-gif-5189345

[summer laurel]

do you know a "workaround" for not being able to put files in LMArena (.pdfs, .mp3...) ? I especially have some powerpoint with 200+ slides (with images) I want AI to analyze, just like Gemini can in Google AI Studio, but with the plethora of models there is in LMArena.

[lunar kettle]

Hello

[quartz light]

what model

[queen veldt]

Ai CAN'T analyse slides

It will just read few lines or if you're lucky half of first page

[quartz light]

it can

[queen veldt]

And than hallucinate the answers

[summer laurel]

Usually it works when slides are converted to PDFS, or a Google Slide document for Gemini

I usually have satisfying answers

[zealous sparrow]

i had the idea of making more google accounts to maybe get the gemini new model

but google recently added

if you want a google account

own a phone

[empty stump]

if you have phone will you need number now still?

[summer laurel]

you can have the new models in google ai studio, no ?

but still, there's no way to insert documents in LMArena ?

[zealous sparrow]

there seems to be a google rollout on gemini website

[summer laurel]

ah ok

[cloud zinc]

get a new phone then

[tired shadow]

guys do anyone will the gpt 5.1 be out in LMArena?

[cloud zinc]

how do i use gpt 5.1 in lmarena?

[tired shadow]

not out yet

I hope it will get released soon

[fleet lintel]

i dont care much about image generation but looks like Nanobanana-2 is also coming up :

https://www.reddit.com/r/singularity/comments/1ow6j7q/i_have_access_to_nanobanana_2_send_promptsedits

[cloud zinc]

we been known this.

[fleet lintel]

ohk. I am not following image gen models. no use to me

[inner gate]

Has gpt 5.1 replaced 5?

[balmy mist]

i think

[cloud zinc]

damn, nano banana 2 is nerfed

[zealous sparrow]

are you the only person here with that thinking model in your gemini on web

[halcyon nimbus]

15 minutes till gemini 3!!! (maybe)

[cloud zinc]

why 15 minute

[halcyon nimbus]

just what a leaker said

i give it a 1% chance lol

[cloud zinc]

which leaker

[fleet lintel]

horse helmet is not air-tight... fail!

[halcyon nimbus]

strawberry man

[cloud zinc]

100% fake

[cloud zinc]

they just announced sima 2 research. they wont overshadow it with gemini 3. its coming next week

[fleet lintel]

launch without tease from OfficialLoganK.. I doubt it

[cloud zinc]

who officiallogank? another leaker?

dont trust those twitter fake leaker dude

[fleet lintel]

nah.. he is one of the top PMs working in gemini. generally he teases 1 day before any major gemini launch

[cloud zinc]

u are just falling for bait and fake things.

[fleet lintel]

(https://x.com/OfficialLoganK) : his bio says : Lead product for @GoogleAIStudio + the Gemini API.
not really a leaker

Logan Kilpatrick (@OfficialLoganK) on X

Lead product for @GoogleAIStudio + the Gemini API. My views!

X

[cloud zinc]

its gonna next week.

only two weeks left in this month.

[fleet lintel]

yup.. i am 90% confident that it's next week and 100% confident it is in 2 weeks

[tulip tree]

@fleet lintel

Is Gemini 3 gonna be free

In the Gemini app

Or a free trial or something

[cloud zinc]

yes gemini 3 flash will be free

[tulip tree]

And pro?

[cloud zinc]

limited. 5 a day.

[balmy mist]

all will be free

[tulip tree]

I have pro will it unlock Gemini 3 pro

[cloud zinc]

100%

[balmy mist]

you get gemini 3 pro for free in aistudio

why the hell would you pay for that?

[tulip tree]

Is there gonna be a difference in image quality

[halcyon nimbus]

3 pro will be free (with a limit) in ai studio

[tulip tree]

@halcyon nimbus

[cloud zinc]

yes nano banana 2 coming

[tulip tree]

What is dat

[halcyon nimbus]

we dont know if nano banana is coming at the same time as gemini 3 or not, but it will use gemini 3

nano banana is gemini 2.5 image gen

[tulip tree]

Nano banana 1?

[halcyon nimbus]

yeah

[tulip tree]

Cool

https://tenor.com/view/no-noooo-british-cop-screaming-no-despair-crying-gif-14408871079120163442

[fleet lintel]

next few weeks are going to be amazing.
I am quite sure that OpenAI will release somethign as soon as google release gemini 3 .. on the same day to steal or reduce the thunder. They always do that.

But good for us... more stuff for us 🙂

[stray aspen]

where can i use nano banana 2

[ocean vortex]

They gonna release smth for sure this year still. 5.1 was such a non-release filler...

Did it out of spite to bump the number before gemini3 drops lol

[plucky sparrow]

What does gemini 3 on lmarena say it's knowledge cutoff is?

[ocean vortex]

Like people can't even reference the performance of gpt5.1.... it's all just boring irrelevant tuning. Not their next version with actual gains

[stray aspen]

5.1 sucks

[fleet lintel]

I think 5.1 release shows that that chatgpt usage is not going as fast as they expected after 5.0 launch. And they are partially attributing it to the "tone" of the 5.0 model and it is an attempt to fix it and go back to gpt 4 tone

[ocean vortex]

yeah it does. I can't actually believe how lacking that 'release' was. It's just a number bump and personality tweaks.

[stray aspen]

new gemini canvas is a freak

[halcyon nimbus]

5.1 is probs one of those efficency upgrades and not a power upgrade

[stray aspen]

5.1 is like the the gemini 2.5 flash update to flash latest

[ocean vortex]

yeah but they haven't stopped making performance gains. For some reason they decided to release that separately

Though the naming gonna get interesting now

either 5.1 gonna be an extremely short weird stop gap

or they gonna name it smth else entirely

gpt5.5? gpt6?

[fleet lintel]

just perfoemance (gpu optimziations) gains doesn't really need external version change.

[ocean vortex]

what

performance typically refers to model accuracy

[grand cliff]

Dabbling with 4o...It somehow guessed the exact model and state of my phone.

[ocean vortex]

that absolutely needs version change when you change the weights and train it to make it more precise

[fleet lintel]

oh.. i thought you were referring to compute efficiency.. my bad

[eternal spruce]

Is there an unlimited credits website for free video generation?

[halcyon nimbus]

LOL

[stray aspen]

qwen

lol

[halcyon nimbus]

grok has the highest limit you get 50 videos a day with premium (at least with sound), and you get one or two a day free ig

[stray aspen]

grok videos suck

[halcyon nimbus]

grok videos suck IF you use their video model to generate your base image

i use hunyaun 3 as a image and make a video with that with grok and its epic

[woeful elk]

hello everyone. happy to be here

[runic scarab]

hello guys I wanted to ask what's the limit of how long of a video the bot can make?

[halcyon nimbus]

depends what model you get, longest is what, 8 seconds i guess?

[runic scarab]

Oh okay okay and we can essentially generate infinite videos..?

[halcyon nimbus]

5 a day

[runic scarab]

Alright got you

[runic scarab]

also if we ask in the prompt to specifically add sounds and what kind of sounds to add will it follow these instructions and give me a bot which can generate sound?

[halcyon nimbus]

no, its still random (i tried)

[runic scarab]

oh damn alright Thank you!

[halcyon nimbus]

np

[jovial sapphire]

Gemini 3 on Canva is not as good as the checkpoints we had on AI Studio

[jovial sapphire]

Nah

It's not that good

Fails all my game coding prompts

[thorny berry]

I’m a freelance LLM engineer (Python / LangChain / RAG / fine-tuning) taking on a few small projects.
If you need help building assistants, search + retrieval, or productionizing models, DM me with a short summary of the task and budget. Happy to share quick examples and an estimated timeline. 🙌```

[halcyon nimbus]

already got my system prompt ready for 3.0 !!

https://tenor.com/view/let-me-in-eric-andre-wanna-come-in-gif-13730108

[wheat onyx]

Unverified

[balmy mist]

lol lies

[jovial sapphire]

Riftrunner is bad

[zealous sparrow]

whats your evidence

[gleaming roost]

true

[halcyon nimbus]

5.1 just released on api, but i sleep till 3.0

[zealous sparrow]

5.1 API is out?

oh it is

[wintry tinsel]

Can you give me a prompt to test on IOS canvas to see if it is Gemini 3?

[quartz light]

funny https://storage.googleapis.com/generativeai-downloads/data/Sherlock_Jr_FullMovie.mp4

[balmy mist]

it is?

[wintry tinsel]

I’m not sure what I should ask to determine if it’s the new model or not

It feels different though

[balmy mist]

lmaoo they just removed polaris noooo

now we gotta pay for gpt-5.1

[wintry tinsel]

You made me think this was the star citizen discord for a minute

[balmy mist]

is 5.1 better than codex?

ughh i wish i used polaris more on openrouter

[neat apex]

polaris did 1700 rating at eqbench creative writting v3

[cloud zinc]

openai cooked

[balmy mist]

why is grok code fast used so much on openrouter?

[balmy mist]

hmm

[neat apex]

some coding by voice app uses that by standard, infinite token usage farm

it is only sigthly faster than grok 4 fast and way worse in anything else xd

[cloud zinc]

some kind of abuse to move up leaderboard.

[neat apex]

gpt 5.1 aint better than sonnet 4.5, it will not reach the 1500 elo

[upbeat dune]

gpt 5.1 not bad not good

[balmy mist]

tbh codex is not as good as 4.5 at times, i was using codex for a while but idk what happened the tool call fails are horrible now, so i went to 4.5

and i used polaris a bit, but ehh

[neat apex]

gpt 5.1 looks great, but sonnet 4.5 is just like sonnet 4.5

[balmy mist]

when you use pro does it use 5.1?

[flint sandal]

Wait like if on the code arena trailer there is lithiumflow example, does that mean lithiumflow is back for code arena?

[queen veldt]

Guys ts started

[balmy mist]

how yall been liking code arena? is it supppposed to replace web dev arena?

[queen veldt]

Gemini disc

[balmy mist]

whats this?

ahh

can u dm link

[halcyon nimbus]

i guess there is a chance they would release it on the stream

[upbeat dune]

release what

[flint sandal]

On discord stream? Nahh

[queen veldt]

Discord . Gg / gemini

[halcyon nimbus]

3.0

[flint sandal]

Maybe release the experimental model like it was with 2.5 pro

[upbeat dune]

small chances but we hope

[flint sandal]

Google can release this model rn and be the best. Like we seen lithiumflow, aistudio checkpoints. Its just good, better than GPT-5.1, if g3 was released same day as GPT-5.1, OpenAI will be cooked.

[wintry tinsel]

What da hell is Polaris?

[flint sandal]

Gpt-5.1 checkpoints ig

Similiar

[queen veldt]

Why they need more special occasion gpt 5.1 dropped out of nowhere

[cloud zinc]

gpt 5.1 instant

[flint sandal]

Def gpt-5 series

[queen veldt]

Nobody was hyped for it tho

[wintry tinsel]

Gemini 3 is cooking

It’s more creative than 2.5 pro

[flint sandal]

Many people was hyped for it.

[cloud zinc]

google doesnt do product release live stream

[upbeat dune]

will be it better then sonnet 4.5🤔

[flint sandal]

But i think GPT-5 uses always the same styles on frontend. Its just so uncreative and repetitive

[flint sandal]

For sure

100%

[tight meadow]

I wish there was infinte video gens on video arena and pick what model you want to use

[flint sandal]

And why the hell openai wont open source even old deprecated models like gpt-3. I would seriously be more hyped for this than for gpt-5.1. Like grok open-weighted grok and grok 2 arleady

[tight meadow]

Yeah

[flint sandal]

Because you want to use paid ai for free? Like i dont understand, ai especially video ai is really expensive, and even with llms many people treat lmarena not as a benchmark, they dont even use battle mode, they go to direct chat and use llms for free

[halcyon nimbus]

"open" ai -takes off mask -closed ai

[tight meadow]

Yeah like why not make it free,unlimted Ao people can use ai models like Sora 2 Veo 3.1 ,Veo 3 Hauilo ai and more.

[halcyon nimbus]

lol it costs like a dollar a video

[flint sandal]

BECAUSE ITS EXOENSIVE BRUHH

Please dont treat lmarena as a daily use tool

Its a benchmark

They should add more limits on direct chat

People are overusing it

[tight meadow]

Yeah

[ocean vortex]

they released gpt-OSS. That's orders of magnitude better

[upbeat dune]

how gemini 3.0 can will be strongest model?🥀

[ocean vortex]

gpt3 is just bad from all angles. Would be completely utterly useless model in 2025 lol

[flint sandal]

But it was part of the history

How chatgpt got created

And then other companies started to do this ai

[ocean vortex]

who cares... there are no use cases for it anymore and it is useless now

[flint sandal]

Yeah, so why dont they open source it if its useless?

[tight meadow]

If someone makes a ai video model thats free,unlimted and opensource ill use that ai

[frosty hill]

They should not allow people to generate anything if they don't vote on the content they generate

[flint sandal]

If someone makes a ai video model thats free unlimited and opensource they will go broke

[flint sandal]

TRUEE

[ocean vortex]

Cause that would be pointless and take their time for nothing

[tight meadow]

It's an experiment I call it

[flint sandal]

Take their time like get the weights and post them on huggingface. Like i want the smaller gpt-3 variants, just for nostalgia and research.

For example, to see how gpt-3 would do with the chain of thought.

Or finetune gpt-3 with gpt-5 genersted data.

[ocean vortex]

I mean I suppose they could do that. But it's very low on the list of things to invest their time on for obvious reasons. They would also have to provide at least the basic support to make sure it runs etc

[flint sandal]

Nah. If they just post only weights, someone will create a tutorial how to run them. Even me.

[tight meadow]

Gone wrong sora 2 Llmarena vid

[ocean vortex]

They are not gonna do that though. It;s a bad look to associate your name with some half assed repo with no proper documentation

[flint sandal]

Grok didnt say anything while open sourcing grok 2. Just weights and bye. Its like 10 minutes of work.

And grok is owned by the richest person on earth

[cedar tide]

GPT 5.1 reasoning none
Added in the arena.

Waiting for the reasoning version and chat version

[gleaming roost]

?

[proud hazel]

Hello there.

[tight meadow]

How

[flint sandal]

It is there

Not in every domain

[zealous sparrow]

5.1 is out on the arena

[empty stump]

I'm what

[ocean vortex]

yeah it is

[halcyon nimbus]

lol its going to be the best model for like a few hours till 3 hits this is so funny

[ocean vortex]

ik. But it's a different model. May not interpret it the same way

o3 had higher too

[empty stump]

Gemini 3 today?

[ocean vortex]

didn't reason for longer

just understood the number differently

[halcyon nimbus]

25% chance

[ocean vortex]

Also Mac app users are discriminated cause there's no extended thinking feature lol.

so the juice is at 96 there

it shouldn't

[flint sandal]

Polaris alpha is lowkey better than gpt-5.1

Both non-reasoning

[ocean vortex]

wait you are right. Even summary parameter. WTFFF

[quartz light]

oh yea btw that menu

[flint sandal]

Yeah cuz gpt-5.1 is like a router too

[quartz light]

u can put in any model id and it would still "work"

[quartz light]

i tried gpt 67 and it worked

it gave me a response

i just edited requests

so something is wrong

[tight meadow]

Meta AI

[half mist]

Code Arena is honestly amazing. Next feature we need is a canvas mode which allows for better editing of text or code when using ai

[ocean vortex]

yeah they remade it entirely lol

there was no 'none' option before

[quartz light]

bruh

#1397655624103493813

@tight meadow

[ocean vortex]

wtf is going on

way too many changes for not a single benchmark

🤯

[quartz light]

NICE

its on arena now

i havent been able to test

[slim gorge]

how's gpt-5.1 for y'all so far?

[tired shadow]

good

it always gives good results for me

[quartz light]

dude

[tired shadow]

I prefer him more than gpt 5

[quartz light]

its literally just called 5.1

😭

so do i choose high or chat

to compare to 5

[flint sandal]

Chat

[tired shadow]

instant (chat)

[tight meadow]

Wan 2.5 model now

[tired shadow]

comapre to gpt 5

[quartz light]

its obvious

[flint sandal]

He wants to compare it to. Gpt-5

[quartz light]

but i asked high or chat

[flint sandal]

Better than o1 o4 mini 4o mini o4 mini high, o3, o1 pro, o3 pro, o4 mini low

[tired shadow]

LOL

[ocean vortex]

to compare to Instant you choose gpt5-none

[quartz light]

what

[flint sandal]

Like chat

[ocean vortex]

they somehow keep f'ing with the names on each new release

[flint sandal]

He wants to do this on lmarena

[tired shadow]

im comparing claude 4.5 sonnet to gpt 5.1 rn

gpt 5.1

[fiery gull]

too, but the sonnet for me is falling 🙁

[tired shadow]

claude 4.5 sonnet (I asked for a simple not proffesional)

so claude 4.5 sonnet followed the prompt

but gpt 5.1 just started creating a proffesional one, and I didnt ask for that one

[proud hazel]

What's your prompt?

[cloud zinc]

is this gpt 5.1 thinking?

[quartz light]

"kinda chill" "thingy" "weird dots"🥀

uh

[tired shadow]

what is that?

what was ur prompt?

[quartz light]

doom

apparently

[tired shadow]

oh

[slim gorge]

oh.

[quartz light]

yea..

prompt:

perfect doom replica full mobile support 2 joysticks for movement and left/right camera rotation, wall collision, proper enemies, textures animations vfx and 3d assets made with code, block text selection, replicate nipplejs joysticks transparent circle style for action buttons like interact and fire

[gaunt spade]

brodie wth is that model

gemini 3 always create the same doom

[quartz light]

5.1

and this is 5 medium (this is actually pretty good)

ignore the uhh red

[gaunt spade]

that cyberpunk lol

[quartz light]

thats the projectile

[gaunt spade]

not doom

futuristic ahh game

[cloud zinc]

lame

[quartz light]

5.1

5 medium

[slim gorge]

i got a chatgpt plus subscription rn but should i switch to claude?

[gaunt spade]

idk tbh

there's not a big difference between claude and gpt now

but gemini 3 pro is a big leap

so u might hold on

[quartz light]

https://019a7ed0-569f-78d7-89e3-c6685a5cc90b.arena.site

https://019a7ed0-569f-7183-8571-a2013c79469e.arena.site

if yall want to try it

1 is 5.1
2 is 5 medium

[gaunt spade]

futuristic ahh doom

[echo aurora]

Made a pretty good version of Boba Arena - https://019a7ed4-a4ac-7186-adf5-5a42a94054c5.arena.site

[quartz light]

hardcoded ui design ahh doom

[ocean vortex]

wasn't minimal at 8?

Or do I remember it wrong?

[cloud zinc]

which model

[ocean vortex]

Ok fair. Was awhile since I checked it

[echo aurora]

5.1

[ocean vortex]

They said it's more dynamic now too. As in short reasoning is now shorter. So there are more variables at play

[quartz light]

oh my god

what is this slop

its worse in functionality than it looks btw

just look for yourself

https://019a7ee1-f5f6-781e-909b-68ca6441cf79.arena.site

this is so bad

clearly prioritising ui now because thats what people judge outputs off of

[quartz light]

@ocean vortex @deep adder what do you think 🥀

[halcyon nimbus]

5.1 is. 3 was but isnt rn

[jovial sapphire]

Evidence? I'm a dev and I tested it, it's bad

It's not even close to the checkpoints we had on AI studio

[quartz light]

riftrunner is still there

[quartz light]

ecpt, k0t, x28 💔

[ocean vortex]

looks weird lol

[quartz light]

LOL

[jovial sapphire]

Yes

x28 was crazy

[zealous sparrow]

This is the best thing we got riftrunner to make , so i would say its decent

[quartz light]

gpt 5 medium (same prompt)

https://019a7ee1-f5f6-74fb-aacd-2c54242a3c6b.arena.site

[jovial sapphire]

Just to show you the difference

[quartz light]

dude look

this is good and its 5 medium

[jovial sapphire]

with one I had on AI studio

[halcyon nimbus]

riftrunner is like flash or something?

[zealous sparrow]

AI studio was overbuffed and is prob gemini ultra bro

[flint sandal]

Hopefully flash lite

[jovial sapphire]

this is AI studio gemini 3 pro: https://x.com/i/status/1978556493625450884

ً (@phth0nus)

this is gemini 3
and it is AMAZING!
i asked it to create a geometry dash game if it was made in 2000s and the result is just mind blowing.
the music you hear is 100% produced by gemini 3.
#ai #gemini3 #agi

X · Oct 15, 2025, 8:20 PM

Now try the same prompt with Riftrunner

"Generate full HTML file of a clone of Geometry Dash, but if it was made in the 2000s, add music to levels (make the music using JS varied music that reflect levels) same physics as Geometry Dash game we want a full playable game. All in one HTML file, minimum 1k lines"

You'll see how bad is the result lol

[jovial sapphire]

this was an amazing model

truly surpassing every other model

[zealous sparrow]

either x23 or ecpt

[jovial sapphire]

there is no ultra

[jovial sapphire]

it was x something

not ecpt

[quartz light]

x28

[zealous sparrow]

that is deada- the best checkpoint like

[quartz light]

?

[jovial sapphire]

x28 yes

[zealous sparrow]

what do you want

[jovial sapphire]

yes it is the best one

i want them to release this one

not riftshitty

wait i'll try right now on canvas

the prompt

"Generate full HTML file of a clone of Geometry Dash, but if it was made in the 2000s, add music to levels (make the music using JS varied music that reflect levels) same physics as Geometry Dash game we want a full playable game. All in one HTML file, minimum 1k lines" and show you how bad it is

[zealous sparrow]

x28 wont be released imo

[noble jewel]

weirdly it worked on canvas for 1 time only, tried a couple more times and no 3.0

[zealous sparrow]

too much processing power goes for one gen

[zealous sparrow]

canvas is not x28

[jovial sapphire]

Yes I know

but apparently it is g3

[zealous sparrow]

also music can be made in canvas

[jovial sapphire]

and I have confirmation that

it's not G 2.5

sometimes

on canvas

[zealous sparrow]

imo i just think people have greed

I think google went by this

x28 - too expensive to run lets make a new checkpoint

then they drop other checkpoints

and when they finally got one they can keep up

they release it

[jovial sapphire]

nah

i think they can release

x28

for google

it's nothing lol

[noble jewel]

well they have to stay on top of their game

[jovial sapphire]

they will release genie 3 too

they did veo3

yes they have more to win

than lose

[zealous sparrow]

if that is x28s processing power i want to see them keep their ai business up if this happens

[noble jewel]

there are others also working to bring out the best ai and they dont wanna be left behind

[fleet lintel]

one shot? that's so good

[jovial sapphire]

yes

1 shot

no other model can do it

[zealous sparrow]

imo justice for riftrunner

first thing people did with they greed is hate on the checkpoint

it was apparently kind of seen that riftrunner outperforms the Gemini 3 canvas CHKPT

[jovial sapphire]

it's not greed bro

i have tried multiple checkpoints

it is simply not as good as the others

that's it

we're not here for emotions

[zealous sparrow]

yeah but good results are costs

[jovial sapphire]

theyre ai models, we want performance

yes and?

we will pay

for subscriptions

[quartz light]

that just means canvas is either more lobotomized or its just the format canvas asks gemini to use

[jovial sapphire]

we give google our data

everyday

for free

theyre a giga company

who cares

[zealous sparrow]

Sure is but think about it

Millions of prompts each with 1k lines of code.

That's costs in the billions for keep up.

[jovial sapphire]

it's nothing bro

money isnt the issue

they have waaay enough

[zealous sparrow]

I mean they bought a nuclear power plant so what do we talk

[jovial sapphire]

were 100% legitimate

to not be happy

when the models are bad

theyre trained on our data

customer is king

[empty stump]

company want more money

[zealous sparrow]

Think i got riftrunner not sure

tailwind css sounds like its type

Also, things dont need to be above 1k lines to be perfect

nope glm 4.6 tricked me

[quartz light]

tru, i dont think any major ai company is profiting off of selling subscriptions/api but rather the investors and it helps the company stay known in the future for being a good service early on or smth idk

https://tenor.com/view/yeojinvevo-gif-23718088

askdkdkslfkfjfjkddknfn

[zealous sparrow]

the investors are mostly their income

[quartz light]

ya

[zealous sparrow]

A company's life is investors

lose em all, you are nothing.

I want to believe one thing

riftrunners api was just pulled

Got it and it only errors

@echo aurora Is riftrunner still on arena? Any prompt with it errors.

[jovial sapphire]

lmao

this is not how it works

you have to go on the website

[zealous sparrow]

i wanted to show you but cant because either google pulled riftrunner or just lmarena issues

[ashen mauve]

ok but is lil bro going to be ok with his full brain and hair transplant?

[zealous sparrow]

no i dont think riftrunner was pulled..

It just errors mid gen.

we are back some AI generated me 1432 lines of code but dont know if riftrunner

we will see...

[echo aurora]

Will look into and try to repro.

[zealous sparrow]

Alright.

[echo aurora]

Are you able to send me a screenshot of the model + error?

[zealous sparrow]

One second. Searching for it through a lot of prompts ratelimited me from the site.

[jagged crown]

Same thing is happening for me

[zealous sparrow]

Yeah, will be a while till i find it, doesn't help that i buried the gen deep down and cloudflare thinks im goin too fast.

[echo aurora]

Are you able to share a screenshot by chance?

[jagged crown]

i closed the tab but will try and reproduce error and will send

[echo aurora]

Ah yeah okay we're now seeing the error.

Team is investigating

No need to get that screenshot @zealous sparrow

[zealous sparrow]

I am glad.

[jagged crown]

thanks!

[zealous sparrow]

Because unfortuanently the one that i had where that error was.

Was deleted..

Also speak of the devil..

Don't want to reveal it but 110% prob riftrunner.

[jagged crown]

yup

[balmy mist]

man can google just release the g3 api already

all this teasing is leaving a bad taste in my mouth

[queen veldt]

Fr

Yesterday was wednesday they didn't released it so ig next Wednesday?

[zealous sparrow]

@jovial sapphire this is what your prompt became.

[whole swallow]

What yall think of gpt 5.1?

[quartz light]

trash

[zealous sparrow]

cant provide arena link because model uh

you kno

broke

but it gave me the file

[whole swallow]

First model uncensored that doesnt treat you like a kid tho

[whole swallow]

Why you saying so

[quartz light]

hardcoded ui instructions, applies to games even

bad in general

[whole swallow]

• follow instructions

[whole swallow]

Oh so coding

It wasn’t bad with detailed writing

[zealous sparrow]

riftrunner

no where close to x28 but nerfed or not it still did nicely

[fiery gull]

sonnet 4.5 still humiliates

[fiery gull]

but, very better that gpt 5, the worses issues fixed

[pastel mirage]

I joined cuz i wanted to say
the flagging is kind of ridiculous, isnt it? Anything involving potentially mild violence or something gets flagged
Even shooting finger guns gets flagged, the word slice is flagged..

Very funny lol

[gusty helm]

how's the vibe on 5.1 ? any good?

nowhere near gemini no?

I didnt get to play with it, just saw the announcement. I tried out quite a bit lithium back when it was here so I was curious

[whole swallow]

Good

[jagged crown]

5.1 makes all UI look exactly the same. It's clean but busy bc of all the text it adds

Gemini makes unreal UIs

[whole swallow]

What’s the best model for creativity? A model that really pushes hard into creating new stories, uses well the inputs etc..

[burnt sinew]

Any gemini 3 news guys.

[obsidian cargo]

reportedly people have access to it through gemini enterprise

[whole swallow]

Wonderful I will try gpt thank you

[sharp mirage]

no

still waitting

:/

gpt 5.1 is good 😄

[sharp mirage]

gemini is good right ?

[queen veldt]

https://youtu.be/FDeMq8W4-sU

Any app with 0 permissions can steal your data and you can't do anything about it 😭

And fix isn't until the end of year

[whole swallow]

Only android right?

[queen veldt]

Yeah

[sharp mirage]

hat is htat ?>

that ?

what is that ?

[whole swallow]

W iOS ❤️‍🩹

[sharp mirage]

what that do ?

[ashen mauve]

thats fine they cant steal what i dont have

[cloud zinc]

is gemini 3 still on mobile canva?

[jagged crown]

Is riftrunner gone? I can't seem to find it anymore

[burnt sinew]

does it say gemini 3? or does it route to gemini 3?

[quartz light]

just look at this (its supposed to be doom)

lolololol

[echo aurora]

Should still be there.

[quartz light]

https://fixupx.com/ammaar/status/1989071809438707731

Ammaar Reshi (@ammaar)

@stefantheard lmaoooooo just g3aring up!

**💬 7 🔁 1 ❤️ 33 👁️ 3.8K **

FixupX · Nov 13, 2025, 8:44 PM

🤑

l3ts.g0

[cloud zinc]

who that

[cloud zinc]

dont trust random fake leaker.

that guy is fake always posting that i see

[quartz light]

its not a random wdym

[civic spindle]

did they remove the image edit thing, cus all i can do is just generate text to image

i dont see an image text to image

[quartz light]

he works for ai studio

[quartz light]

try pasting an image

btw only supported models have image edit

[civic spindle]

ok well it works with battle mode, i dont see it work with side by side mode

[cloud zinc]

he constantly does this fake bait tho every week. he is no better than any random leaker.

[keen beacon]

Gemini 3 delayed

?

[quartz light]

what

no

[keen beacon]

Oh ok

[quartz light]

nobody knows when its releasing

but prob this or next week

[keen beacon]

Yeah, I heard that I got delayed because of Kmi2

[quartz light]

lol no

[barren ore]

When will gpt 5.1 be available on LMArena?

[quartz light]

kimi k2 is trash

[quartz light]

it already is

[cloud zinc]

we already know this

[cloud zinc]

this nothing new

[keen beacon]

Well, why they change it all from November 18 and now they updated it all the way up to December 9.?

And spaced it out more

[cloud zinc]

cuz first they release preview and general release in dec

[keen beacon]

What? This is about the models they deprecating.

Originally all the models were supposed to be deprecated on 18 November

[balmy mist]

chill

[keen beacon]

But now they changed it and paste it out all the way through December 9

[cloud zinc]

kimi k2 is trash

[keen beacon]

No it’s bro

[balmy mist]

eww dont use it, extremely mid, use gpt-5.1 codex

[cloud zinc]

gemini 3 is delayed by 1 week.

[balmy mist]

its out bro

what you talking about

[keen beacon]

Is it?

[balmy mist]

you mean for general ppl?

cause on android its out

its prob dropping tmw tbh

[keen beacon]

https://www.androidsage.com/2025/11/13/google-gemini-3-rollout/

Looks like

Platform Availability: Mobile Gets Priority

[balmy mist]

yeah it makes sense for it to drop tmw

well fully

or in a week(or during shipmas lmaoooo)

is OpenAI even doing that this year?

[keen beacon]

I think so

12 days thing

[balmy mist]

lol

[cloud zinc]

i dont think they doing 12 days thing

[balmy mist]

they dont anything to ship lmaoo

imagine googlr does shipmas with nano 2 abd g3?

[keen beacon]

https://youtu.be/pRa68PZlvzI?si=zXkg9Y42SWE7JTjD

[cloud zinc]

they dont release anything on friday

[keen beacon]

Good point

[cloud zinc]

its either tuesday, wednesday, or thursday

[keen beacon]

Valid

[cloud zinc]

today google announced this

[balmy mist]

damn, y is that?

[keen beacon]

So they can fix all the issues probably that’s gonna rise from day one

lol

[cloud zinc]

weekend, no time to fix bug

[balmy mist]

makes sense

[keen beacon]

I heard it got delayed all the way till December I’m not sure

[balmy mist]

bruhh

[keen beacon]

Everything is rumors these days

[balmy mist]

fr

i just wish never git any leaks for g3

cause i didnt care until we saw the leaks

[keen beacon]

Some people are thinking it’s because of kimi2

[fresh mirage]

I know they released gemini 3.0 on enterprise

[keen beacon]

Because they can’t drop a model that under performs

In coding

[fresh mirage]

gemini 3.0 as far as I've seen was performing pretty goddamn well

but hey, they wanna take precautions, that's a pretty good mindset

except

when they do their daily hype-posting shenanigans

[keen beacon]

https://finance.yahoo.com/news/airbnb-picks-alibabas-qwen-over-093000045.html

Interesting to see American companies, choosing open source Chinese model models

[stray aspen]

the automatic switching from text model to image model when you paste an image is extremely annoying

[queen veldt]

Admins are aware of it they'll fix it someday

[gleaming roost]

someday

[echo aurora]

Sorry if there was incorrect information given previously, but this is actually intentional.

We saw more user expecting to use image-edit instead of vision after image upload.

[mighty ocean]

Wait is the new gpt 5.1 model thats dropped not thinking? Like on lmarena

[echo aurora]

Correct

[queen veldt]

Well maybe some state like if user already typed 2 prompts to regular model it shouldn't switch to image in middle of conversation

[mighty ocean]

Are you not able to drop thinking?

[echo aurora]

We're looking into other variants and will be sure to put out a new announcement when/if added.

[mighty ocean]

Interesting

Thanks for the responses

lol pineapple

[echo aurora]

Yeah this is a good callout. I was thinking about this earlier too where maybe it'd be a better experience if the behavior was different in Battle vs Direct/Side. Since in Direct/Side users are picking specific models that intent for what they're looking for is a bit higher.

Downside being if those worked differently from each other I can see how that'd be an odd experience.

[mighty ocean]

I have to say I understand the point of the battle arena

Allowing stealth models to be direct

Doesn't make a lot of sense in my opinion

It's meant to be hidden so if you have direct access how is it a stealth model

People just want access to unreleased models right at their fingers instantly

[echo aurora]

YEah this is something we've put some thought into, and will continue to debate internally about.

[mighty ocean]

I am one for balance and anticipation

Having little bits and trials of the unreleased models is cool, but it breaks balance when you allow that complete freedom

What model would you guys say would be by populous the best for factual information

Like less hallucinations

[safe sleet]

Gemini 2.5 Pro (if given RAG info to work with)

[sullen quest]

no

[safe sleet]

Its training data/internal knowledge is dated, though

[sullen quest]

Not if you are using tools like web search

[sullen quest]

depends on the type of factual information you are looking for

[jade egret]

Gemini app cavans looking fire rn

[jade egret]

I think it gemini 3.0?

cuz it really good

[unreal dagger]

Would you guys ever make a leaderboard for users to see who has helped the most?

[sullen quest]

the argument for is that there's a ton of people who when they hear about a new exciting new model, immediately rush lmarena with prompts and stuff, without caring about accurately voting.

[dry folio]

hello

[sullen quest]

hi

[hasty thorn]

Hey, why did you remove the "Retry" button?

[astral blaze]

Why is the retry button removed?

[hasty thorn]

It just disappeared, I don't know why, I refreshed and it's still not there

[astral blaze]

They removed it for battle mode

[hasty thorn]

Because?

[astral blaze]

They'd do that but not fix the absolutely atrocious rate limit that bans you for scrolling down your chat history

Lol

[astral blaze]

Do I look like I know

[unreal dagger]

Any reason after 20 lines of talking it just dies every time

[hasty thorn]

Can they fix it?

[stiff glacier]

🫠

[hasty thorn]

They haven't answered my question

[cloud zinc]

canva is gemini 3?

[rocky mauve]

When is Gemini 3 coming to lmarena

Or is it somewhere else already?

Oh it’s not even released yet, my bad

[astral blaze]

People think that the Gemini app is routing people to Gemini 3

Which I think is probably not true

[keen beacon]

I’m very skeptical

Curious how they’re gonna do pricewise

[leaden egret]

Wait what are the different models

[stiff glacier]

I don't know

[magic stag]

There was an exploit regarding retry mode and stealth models

[magic stag]

Doesn't matter what you think its undeniable

Generate an SVG of an xbox controller

[jade linden]

hey

[magic stag]

Give me a min ill do more

[jade linden]

how can i upload an image and generate a rendered image of the same?

[astral blaze]

Placebo effect in full swing

SVG tests are subjective to begin with

It's clearly not the same model as lithium

[magic stag]

well, it's not

that's very obvious to anyone with a functioning eyeball (dont even need two) and a functioning brain cell

[magic stag]

same for the difference between these

this is not "subjective", lol

[cursive shoal]

How do you guys get the Gemini 3 in the mobile app?

I'm on USA VPN, pro subscription, canvas mode and it still looks like 2.5 pro

[sullen quest]

supposedly its random

like a really low shot

[dull jay]

why removed the retry button of LMA?

is retry not ok or what?

[astral blaze]

Yeah but the output is worse than lithium. It's not Gemini 3

Not that you have a way to know for sure unless you think Reddit and x are reliable sources

[sullen quest]

@echo aurora

[sullen quest]

its a google model

I'm not sure what that logic is

[simple copper]

@echo aurora

[astral blaze]

Apparently if I do not hype I don't have a functioning brain cell

[dull jay]

im sure that gemini3 in app is already dead, cant even answer 9.9-9.11 correctly now

[astral blaze]

It never was confirmed and why would google even roll it out like that

Its literally mass delusion

[sullen quest]

it wasn't a guaranteed chance to get

[echo aurora]

Hello - yes this is an intentional move we made today. We're very open to feedback so don't hesitate to share.

[sullen quest]

idk, previous gemini checkpoints have been rolled out in google ai studio, gemini.com and lmarena at different times, there's no obvious pattern

[sullen quest]

why?

[empty stump]

to ruin the user experience?

[dull jay]

I would say it's really a bad update and ruin all users' experience.

[echo aurora]

We were seeing a lot of what looks like abuse

[sullen quest]

wdym, how would you abuse that.

[exotic crest]

i'm sorry what

[sullen quest]

you don't know what the model is. and you can't change the prompt

[dull jay]

yep

I'm not even dreaming LMA add prompt edit feature now, since you even removed the retry button.

[marsh sundial]

did the canvas work for writing test? I'm tring to get 3.0 but the quality is obvious 2.5pro

[sullen quest]

how about the retry button exists, but only if the llm errors out?

[sullen quest]

idk, most people testing it were comparing svg's and stuff

[astral blaze]

No no if you don't see it you must not have eyes as per @magic stag here

[unborn lantern]

Is lmarena or yupp ai have alternative?

[half mist]

I think the next big step for LMArena is to either 1: Make a canvas mode similar to how Gemini handles it or 2: Make LMArena an iOS and Android App. Both of those can significantly make LMArena better

[astral blaze]

So they killed the retry button?

[exotic crest]

it appears so. Asinine

[half mist]

It’s still here for me

[echo aurora]

bad update and ruin all users' experience
Can you help me understand this more? Like I said we're really interested in hearing feedback on this.

[exotic crest]

on one of my chats it no longer works

[echo aurora]

This should still be there for if the model errors out, it just won't on successful.

Is still there in direct/side too.

[sullen quest]

what

[dull jay]

I'm sure you know models got error and returns nothing very often, and without a retry button there is no chance to see the response of the model who errors out

[grim granite]

Same here, the retry button go away!! them go put again? this is temporal or nover go back? D:

[sullen quest]

what? also isn't douyin the chinese version of tictok?

[half mist]

If anything, it’s likely a visual bug that will be patched soon

[fiery gull]

Has the webdev arena, but why? 🤔

[astral blaze]

I see the retries for direct chat is also limited. Coupled with the insane cloudflare timeouts... Lol

[minor storm]

I just need a douyin account please help me bro

[astral blaze]

It's not in battle anymore and the direct chat now has a retry cap

[exotic crest]

This is beyond stupid

[minor storm]

Bhai check the dum

[astral blaze]

Just go back to aistudio I guess

[minor storm]

Brother check the DM

[ashen plaza]

I think you might be in the wrong place here

[fiery gull]

I think nobody here is chinese, I don't remember a chinese online here

[astral blaze]

If they don't want people abusing the battle system maybe they shouldn't put cloaked models (which shouldn't exist in the first place) exclusively in battle

[fiery gull]

Has another country with this app?

[hasty thorn]

Fix it, put the Retry button back in.

[sullen quest]

why should they not exist?

[fiery gull]

No, I'm from Brazil, here don't have the douyin (I think)

[grim granite]

PLz!! put again, without the retry button is hard work in the website! D:

[astral blaze]

Its just for companies to create hype now and just makes everyone ask "who are you" for 100 times instead of actually AB testing

[dull jay]

And I cant even imagine how a retry button could be used for abusing

[astral blaze]

Don't wanna be mean but this sucks

[sullen quest]

lol, I disagree with the first bit though, around something like 2-5% of the cloaked models actually cause hype

But the who are you thing is spot on

[exotic crest]

Even if that was somehow true, what people generate for themselves is their business

[dull jay]

agree

[astral blaze]

It's clearly what some companies are using it for, despite the outcome
Microsoft tried to name one of their models similar to Orion a while ago to ride the coattails I guess

[sullen quest]

not if its a org botting.

[astral blaze]

The real reason was some people were using API bridges for infinite generations
But punishing all users for the actions of a few individuals... Lol

[sullen quest]

I have no idea what the point of orion was, that could be a coincidence.

But just to be clear, a vast majority of cloaked models are like Ernie, or raptor

[astral blaze]

I know but it's annoying now that it's the norm

[sullen quest]

And most cloaked models come and go with barely anyone realizing

[astral blaze]

Just test it internally if they don't want people to find out it sucks

Just my $0.02

[sullen quest]

all times cloaked models have created hype, its because they were significantly better than the opposition, that or they were a google model

[astral blaze]

The problem isn't the 5%, it's the 95% of other companies trying to emulate it

[sullen quest]

meh, its improving it with every test, and not all of them suck that much, they clearly are fine with people trashing the models, its genuinely becasue they are working to improve them

[astral blaze]

These models don't appear on the leaderboards and just dilute the pool of models for actual evaluation and comparison

[sullen quest]

Plus, those models subsidze lmarena, since lmarena usually gets to use them for free

[sullen quest]

the goal of lmarena isn't for you to do eval, its for lmarena to do eval

[astral blaze]

Then they should put them in their real names and allow them to get on the leaderboards

[sullen quest]

and in reality theres only a handfull of models that lmarena uses at any given time, so cloaked model increases the pool

[astral blaze]

Then what are we doing here? We're testing for things that we don't see the results of

[sullen quest]

I mean, theres been like 20+ different versions of erine, I don't think I need 20 different versions of it cluttering up the leaderboard

[sullen quest]

the idea is that you get free, basically unlimited ai use

[astral blaze]

I think they need to make a good Ernie and put that on Lma instead

[astral blaze]

Well... Not anymore

[sullen quest]

wdym

[astral blaze]

Obviously removing the retry button is to stop you from using it for free and with no limits