general

[honest verge]

No ai since Gemini 3 pro are better than it

They are nerfed to Gemini 3 level

[surreal zephyr]

yup

[honest verge]

Last progress we made was in 2025

Early 2026 is pure nothing

Wth is this bro doing 💀

[surreal zephyr]

3 flash is like 800b 20b parameter model let it be!

[stray aspen]

Brotato chips

[honest verge]

Most accurate one

[surreal zephyr]

3.1 pro is literally 3.0 pro pre nerf

[honest verge]

Is he sleeping or some reason?

He does literally nothing

[surreal zephyr]

damn

[honest verge]

I'm trying to stop him but he still does this

[high owl]

hi

https://klipy.com/gifs/hello-chat-hello-18

[honest verge]

Are you kidding me?

BRO WHAT ARE YOU DOING

I don't understand what's going on

[high owl]

batlle in direct...

https://klipy.com/gifs/openai-chatgpt

[pulsar crystal]

the point of arena is to compare models
a lot of people are misusing the site
this is their current experimental solution to tackle that problem

[honest verge]

I'm begging you to stop doing this task

Bro dude Gemini 3 flash

I cancelled it 6 times already

[surreal zephyr]

gemini welding a cube, 2026, colorized

[high owl]

create the minecraft for ai?

[honest verge]

It's ez

You just need Google ai studio

[high owl]

why?

[toxic verge]

Facts

Completely useless

[river whale]

i think google released gemini 2.5 pro rebranded as 3.1 pro lol

[half mist]

Yeah, it’s kinda buggy right now

[gleaming roost]

I HATE THISS 🤬

If I wanted two models, I'd be in battle/side-by-side mode, damn it

[river whale]

same

[surreal zephyr]

its same model everytime

[river whale]

its just some system prompt that gets increased everytime

[proud bobcat]

Hold up

New ui?

[toxic verge]

https://youtu.be/ZB9LNf6JaFs?si=FK41mAATbUl8bGTV

[proud bobcat]

Please openai make a good model for once

https://tenor.com/view/black-man-crying-cry-crying-crying-emoji-crying-meme-gif-3223392509479903783

[toxic verge]

https://tenor.com/view/crystal-sparkle-censorship-captain-gif-9398128474724213517

[deft spruce]

I feel like I'm getting constantly bombarded with captchas lately. Anyone else?

[main nexus]

LOL happened to me also, i just gave up

[gloomy onyx]

"Please give me more money, I promise I'll make better AI models with strong ethical boundaries"

[main nexus]

true

[river whale]

openai is gonna ghost out

[toxic verge]

I would

[river whale]

true and the captchas are wild too

[toxic verge]

The ai community will gobble up anything

Probably the easiest group of people to profit off of

[deft spruce]

oh you too right?

[river whale]

yeah

[deft spruce]

is pineapple is here?

[gloomy onyx]

yeah, honestly I think we need stronger whistleblower protections

the last OpenAI whistleblower didn't end up well

[river whale]

battle mode in direct with context getting given to the direct model is so bad

[toxic verge]

The AI community will call whistleblower traitors

They’re too indoctrinated

[deft spruce]

I feel like I'm getting constantly bombarded with captchas lately. Anyone else?

[unreal hatch]

@echo aurora Opus 4.6 thinking is broken pls help

[gloomy onyx]

shrug~1

[river whale]

its been broken for so long

[toxic verge]

We just have to deal with what we have

😭

[unreal hatch]

but why?

[toxic verge]

We’ll get really nice AI for like a month and then they will nerf

[river whale]

i feel lmarena giving errors purposely

[unreal hatch]

yeah maybe or the model its just broken ?

[river whale]

i didnt had any error using claude somewhere else

[toxic verge]

I give Claude like 2 more weeks

[gloomy onyx]

nope, it's just LMArena's website that's not handling the model correctly

[toxic verge]

And that’s gonna fall off

[unreal hatch]

dangg

[unreal hatch]

yeahh ngl i wish Opus was free on web

[deft spruce]

https://discord.com/channels/1340554757349179412/1451574502369656842
well im not only having a that problem

[river whale]

they should just remove context of battle given to the direct chat ai model then im fine with battle coming in direct chat

@echo aurora

[honest verge]

3.1 in ai studio is still down

Is it too hard to fix it?

[golden ocean]

theyre busy nerfing it rn

[honest verge]

And again stuck to the 3.0 lvl

[river whale]

last time it worked-

[honest verge]

We can't make any progress

[quaint trail]

what website is this

[river whale]

aistudio

[novel crater]

does anyone actually know why 3.1 was outputting the way it was?

[river whale]

lyria also nerfed bruh

they playing nerf so hard

[quaint trail]

is it like arena where all the stuff is free

[novel crater]

I don't really feel they would take it down the next day and then nerf it, usually that is gradual and weeks ahead

[honest verge]

I'm nerfing it so hard

[novel crater]

it must be a backend issue

[river whale]

you so gone bruv

[quaint trail]

what

[river whale]

do u know what epoch is?

[quaint trail]

no

[novel crater]

down?

[keen beacon]

which model is the best at storytelling

or creative writing!?

[river whale]

guys i made a chat ai

model

its 300m parameters

so good

[echo sinew]

@tough stratus @opaque patio It seems that you want to create video or images with the Bot. Please note that the Video Arena has been removed from the server. More information can be found in this announcement. Do not spam text channels with prompts as it won't work. Thank you.

[terse shuttle]

is it code arena limit?

i have a big chat and big project on it 🙁

need to continue this

[golden ocean]

its over

[unreal dagger]

Yeah the best suggestion you can do to save projects is copy one by one into a new chat

[inland skiff]

Hi

[golden ocean]

real

[terse shuttle]

oh

[fringe cobalt]

Same here

[trail jackal]

hello

[unreal dagger]

Yes there is a limit on the code arena once you hit that it will say limit was reached please wait -amount of time

[proud bobcat]

I’ve found Kimi K2.5 to be the most human and realistic

GLM 5 very good at prose in general

DeepSeek 3.2 only a bit behind the two but strong and very cheap.

[river whale]

i have free glm-5 api

[proud bobcat]

Absolutely fire

I don’t mind paying for it honestly

For it’s insane performance it’s incredibly cheap

[surreal zephyr]

how

[river whale]

i made it

its in limited access rn

[molten cipher]

ai studio doesn't work for me :(

[surreal zephyr]

lies

[surreal zephyr]

mhm

[molten cipher]

even tho im using teir 1 paid api key

[river whale]

want proof?

[distant bloom]

i can give kimi 2.5 free to cool projects

[keen beacon]

guys

which model is good for creative writing except claude 4.6 opus

[surreal zephyr]

gemini 2.5 flash xhigh

mistral large

[keen beacon]

100%?

[surreal zephyr]

the dumber ones are more creative

[night sail]

if i could upload js , css, .txt file it would be wonderful

[keen beacon]

WTF

you fr?

[surreal zephyr]

mhm

[keen beacon]

which is better

gemini flash or mistral

[proud bobcat]

I just

[unreal dagger]

Gemini is the best

[proud bobcat]

I just answered that

[keen beacon]

gemini 2.5?

[unreal dagger]

Any

From gem

[keen beacon]

alr bet dawg.

[proud bobcat]

Gemini has too many guardrails and is too expensive

Kimi is the most human

[unreal dagger]

It’s free on lmarena?

[proud bobcat]

OH

[keen beacon]

yeah gemini is free

[proud bobcat]

I thought

[keen beacon]

only in lmarena

[proud bobcat]

Ohhh I thought you meant api

Still I think Kimi is very strong

I don’t like Gemini’s prose

[keen beacon]

which kimi ver

NOW.

[abstract tundra]

is this usable somewhere rn for free?

[unreal dagger]

Just the most open ended model so people like it

[quaint trail]

is there a way to use nano banana pro for free or do i have to switch accounts

[surreal zephyr]

imo use gemini 2.5 for writing the story plot, then gem 3.1 for fixing stupidities

[keen beacon]

OOOOO

BETG

bon appetito betch

[river whale]

i have api for it too

its in limited access

[keen beacon]

gemini 2.5 flash?

or pro

@surreal zephyr

[river whale]

https://se.xdativo.cloud\

[keen beacon]

@surreal zephyr YO

REPLY DAWG

i need ur guidance for ai..

[golden ocean]

fr

[surreal zephyr]

or normal, whichever u prefer

[keen beacon]

idk

flash or pro

[surreal zephyr]

basic gemini 3.0 pro would do fine too i guess

[surreal zephyr]

flash faster but even dumber

[golden ocean]

@ocean ferry does not approve of u asking ampro for ai guidance

[surreal zephyr]

if dumber = more creative, then clawd ai is the most creative one out of all /j

[keen beacon]

so you saying pro is better

[surreal zephyr]

you need to find balance

[thorny schooner]

I'm just hoping actually wishing they will actually listen about battle in direct because almost every other debate about any of the other updates can debate on some level but this has real no benefits at least in the way they are implementing it 😭

[keen beacon]

how

wtf

[surreal zephyr]

between creativity, stupidity, prompt adherence, ect

[keen beacon]

OHHHH

alr imma use 2.5

gemini 2.5 pro

[surreal zephyr]

the big issue for me is that it often corrupts the actual project

[surreal zephyr]

try, if you dont like try 3.0 pro. ect

[thorny schooner]

fine damn good choice ( i meant that in the good way since honestly that's my default)

[keen beacon]

ok thanks bros

[surreal zephyr]

https://gemini.google.com/share/84327f5018a7 @golden ocean whatchu think?

[golden ocean]

ip grab

[thorny schooner]

It pops up like every two prompt for me and I don't think I need to tell you all how disruptive that is and sometimes / a lot of times it's every other prompt make sure it's a whole hell of trying to do chat AI stuff

[golden ocean]

https://tenor.com/view/sasquatch-wilderness-big-foot-sighting-gif-15112239

[surreal zephyr]

cube welder

[devout bluff]

I can't turn water into wine but I can turn you into mine

[thorny schooner]

Kind of the same for me the more deception of the memory and context disrupting the whole thing ( i don't do projects but I know what you mean0 to the point that I will trust often and not copy and paste the same prompt after building so it won't give me the normal answer even though will give me a disruptive answer a lot of times I notice it will centralize/simplify the questionWhen I do it that way which just does not give me a accurate answer to be honest kind of hard to even correct that since every individual prompt risk just being disruptive again if it's a battle mode

[thorny schooner]

wtf

[golden ocean]

are they welding @ocean ferry

[surreal zephyr]

they be welding a box

you can put him there

[thorny schooner]

☠️

[proud bobcat]

Yeah wow Gemini 3.1 pro is awful

It’s so slow

It took 10 minutes to decipher a basic cone surface area problem

[river whale]

it took 25 seconds to reply to "yo"

[proud bobcat]

And then

[surreal zephyr]

https://media.discordapp.net/ephemeral-attachments/1340554757827461211/1474427769696813056/1984.gif?ex=6999cf5d&is=69987ddd&hm=a8d689761652b4594432528b78a36e60451f55bec3eeca40ac4fb9d393be1e3f&=

[proud bobcat]

It lost connection

[proud bobcat]

Literally real

They did not cook with this

Nah even 3.0 pro never choked like this

At launch too

[river whale]

gemini 3.1 and lmarena battle in direct mode what a awful combo

[surreal zephyr]

3.1 was nerfed PRE launch

but its still really good

bigger nerfs incoming

[proud bobcat]

This is ass

Kimi solved it right away lmao

[golden ocean]

solved what right away

[proud bobcat]

Basic cone surface area problem

3.1 lost connection

[golden ocean]

ah

[molten cipher]

WHY ISN"T AI STUDIO WORKING

[golden ocean]

https://cdn.discordapp.com/attachments/1442924145615568956/1442924156546056272/MConverter.eu_v14044g50000d485kpfog65rnjameha0.gif

[dusty girder]

Hello

[golden ocean]

real

[surreal zephyr]

https://gemini.google.com/share/d366ac78447b

[sick mantle]

When is @echo aurora Going to add genie made by google

[river whale]

this battle mode is so annoying

BOTH ARE STUPID AI MODELS

[surreal zephyr]

"would you like gemini 1.0 flash or gpt 1.0 corrupt your opus 4.6 project?"

[river whale]

lmarena is so doomed

[keen beacon]

holy

mistral large 3 was actually better

HOLY

[sick mantle]

Same, i hate battle mode.

[river whale]

its just over at this point, they giving battle mode at every 3rd prompt

and the models that they give are slow af

[long minnow]

its funny when arena just pushes an unfamiliar model in front of you during battle mode
vidu, rcps-fast (or rpcs?), wan 2.6 pretending to deliver veo quality etc. 😅

[river whale]

its not even funny at this point

its so doomed at this point

[long minnow]

I hope they get it sorted out

[unreal dagger]

Yeah that is a issue that needs to be fixed @echo aurora

[river whale]

like i really dont want "seed-1.8" to deliver opus 4.6 quality

[long minnow]

for the video generation part I think they just wanna prevent overloading the popular models

[long minnow]

haha

[river whale]

@echo aurora

[bright junco]

Starving for votes so badly your stomach’s growling? Guess getting them in BATTLE MODE wasn’t enough to fill you up, so now you have to come looking for more in direct.

[river whale]

some people will now come and defend lmarena

[light sleet]

They won't

[whole swallow]

Damnn

[spring plinth]

Hello

[raw laurel]

Is it possible to export all chats?

[frigid mica]

Hello

I need gmail

[deft spruce]

true

[edgy iron]

aren't these supposed to branch off, not affecting initial dialogue?

[frigid mica]

Not able to made it

[echo aurora]

Yeah it shouldn't be triggering this often

[surreal zephyr]

they dont :)

[surreal zephyr]

the real issue is they break the original convo

[echo aurora]

Are you able to repro this every 3 messages?

[surreal zephyr]

the moment it appears, the orignal model cannot continue

[river whale]

Yess

U guys should fix it breaking the conversation

Don't send battle context to the ai

[echo aurora]

What do you mean by this?

[river whale]

It breaks the quality

[surreal zephyr]

has to restart from 0

[echo aurora]

Can I have the Eval ID of the chat session this is happening with?

[surreal zephyr]

exacly

[echo aurora]

Eval ID is the random set of numbers/letters in the URL

[terse shuttle]

@echo aurora sry for ping, do you know anything about this? is this invisible limit or model problem?

[river whale]

I'm in mobile now, I'll try to give

[raw laurel]

Even though direct mode is selected, battle mode gets randomly activated within a conversation if ”Max” is selected. Not so good.

[surreal zephyr]

https://gemini.google.com/share/0e491162fce6 🔥

[echo aurora]

It could be, try the steps here to verify - https://help.arena.ai/articles/8931786544-arena-how-to-rate-limit?lang=en

[inner relic]

any news about deepseek v4

[terse shuttle]

oh, so i can continue work after a while?

[gleaming roost]

Battle models shouldn't appear in direct chat; this is wrong and in bad taste towards users. I've reported this before; if someone wants two models, they should use the Battle and Side-by-Side models.

[echo aurora]

If it's rate limit, yes

[terse shuttle]

and how long is it?

[raw laurel]

Exactly. But it seeems to only appear if ”Max” mode is selected. Or have you encountered it even if you’ve chosen a specific model?

[surreal zephyr]

all

[gleaming roost]

Yes, it appears with other models

[river whale]

They want the investors happy

[raw laurel]

This is lame. This is a new function that they’ve implemented. Hasn’t happened before for me at least. I hope they revert.

[inland quest]

Claude 3.7 Sonnet supposed to be dead but it isnt

[gleaming roost]

It's been a while, they removed it, but apparently they added it back

[quaint trail]

if all the models are free on arena, how are they making money?

[river whale]

Bruv they don't make money

[quaint trail]

so how is this still up

[terse shuttle]

@echo aurora so i'm checked dev tools network and it's not a rate limit

cuz no 429 errors

[surreal zephyr]

claude sponsors

it says on their article

[echo aurora]

You'll then want to try these steps - https://help.arena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

[edgy iron]

also from what I get all dialogues become public in a matter of months

[echo aurora]

No we have a product, can learn more here: https://arena.ai/blog/ai-evaluations/ cc @quaint trail @river whale

[sick mantle]

@echo aurora

[river whale]

It's over till battle mode is fixed or removed

[sick mantle]

Pineapple want some chicken?

[whole swallow]

it still drops down when finished

[quaint trail]

so how much money does arena use for the API keys for these models

[fiery crane]

Dog

[zealous sparrow]

@echo aurora Just me or is the gemini 3.1 pro model down on arena?

Issue im having with: Codearena, Model: Gemini 3.1 pro

[surreal zephyr]

yeah it throws the cube away then starts again yuh

that was the prompt

[stray aspen]

Is gemini nerfed

[surreal zephyr]

https://gemini.google.com/share/6f9efe7f6226

[zealous sparrow]

i dont think so

[proud bobcat]

Sonnet smokes 3.1 pro

[mighty surge]

antigravity is really bad ngl

[proud bobcat]

Absolute cinema

[toxic verge]

I can’t find any

[zealous sparrow]

the testing is based on pre release

[toxic verge]

😂😂😂

[surreal zephyr]

nope

[whole swallow]

Bottom one missing tho

[toxic verge]

It’s amazing how clueless the models are on safety

[surreal zephyr]

well it decided to make open box

🤣

[whole swallow]

Hhahah

[wicked sage]

sonnet 4.6 still the code goat

😎

i love sonnet

[toxic verge]

Got 2 weeks before they nerf

[whole swallow]

Shii is fire tho

[wicked sage]

they unironically fell off then became top 5

[surreal zephyr]

1

[wicked sage]

no top 3 actually

[zealous sparrow]

is gemini 3.1 pro erroring for anyone else on codearena

[whole swallow]

by monday nerf

[surreal rover]

হাত

[surreal zephyr]

https://019c7add-409c-7528-af5c-80376566f283.arena.site/ see this

[toxic verge]

No, they know we’re onto them so they’re gonna extend it for a month

[surreal zephyr]

https://019c7799-1908-7f24-b3e9-f0fb5dd8f92c.arena.site/
https://019c775e-143e-7242-92b4-0d936da78f8c.arena.site/ and those two are wow

[toxic verge]

They wanna make sure they get as much of those 200 monthly subscriptions as they can

[surreal rover]

#general

[whole swallow]

When i zoom out it disappears bruh

[toxic verge]

Before they start nerfing

[whole swallow]

BUT FIRE THOO

[surreal rover]

#ai-creations

[zealous sparrow]

ehm defo not they just being generous

[surreal zephyr]

low renderdistance i guess

[zealous sparrow]

its usually day one nerf

[whole swallow]

Damnn

[wicked sage]

video arena is gone, please make your videos in lmarena

what did i just say

[whole swallow]

the infinite one is trippy

[wicked sage]

i meant arena.ai

[surreal zephyr]

its a bot

[wicked sage]

oh

lmao

[surreal zephyr]

@whole swallow see those

[edgy iron]

in my experience sonnet still fails at oneshotting code, thought it would be lower

[red sluice]

If you want to try out the next version of Grok Search, use the search tab, and wait to find "arastradero"
It's 100% the new Grok-Search it has the exact same ultra specific problems with API that I usually get with Grok

[surreal zephyr]

for me sonnet is best one shotter lol

[red sluice]

It's 100% Grok-4-2-search

[wicked sage]

GO MY MICHAEL SPHERES

[edgy iron]

I had opus oneshot way more reliably for sure

[surreal zephyr]

https://tenor.com/view/penguin-cute-penguin-tranquilo-penguin-baby-penguin-gif-7917461927994018041

[surreal zephyr]

funny considering sonnet scores best at hallucination bench out of all

NO

[red sluice]

Btw I don't get why admins made it "private" since Grok 4.2 is on public beta for two days, makes no sense to hide the models' name for this one honestly and it's so easily recognisable

[spring plinth]

#general

[edgy iron]

maybe there is less free tokens for direct chats, than for claude?

[red sluice]

It's already in the arena, and already has a ranking, in fact it was available there as an anonymous model even before the release date.

[surreal zephyr]

its long already added

[deft spruce]

OK SORRY

SO HOW IS IT?

[red sluice]

don't worry it's fine!

[deft spruce]

GOOD?

TO USE?

[keen beacon]

my bad

[red sluice]

Pretty good, but underwhelming compared to the bomb that we always get from Gemini. It being behind Opus & Sonnet is quite disappointing.

[deft spruce]

OH I HAVE TO USE IT RIGHT NOW

[surreal zephyr]

"behind opus" xD

[red sluice]

Well it's one of the best models out there, there must be some specific usage where 3.1 is the best model on earth, but overall it seems like Sonnet gets the gold

[deft spruce]

oh wait my caps lock was pushed

sorry

...what?

[novel crater]

⌇⍜⍀⍀⊬ ⟟ ⎅⍜⋏'⏁ ⎍⋏⎅⟒⍀⌇⏁⏃⋏⎅

[deft spruce]

whaT?

[novel crater]

one message stands

muahahahahaa

[deft spruce]

sorry to ask that ARE YOU F'N A ALIEN?

[novel crater]

discord is reporting to ice cannot disclose sorry

[surreal zephyr]

💔

-# [Reply redacted for privacy reasons]

[novel crater]

--. --- --- --. .-.. . / -... . - - . .-. / ..-. .. -..- / ...-- .-.-.- .---- / -... .-. ..- ....

[surreal zephyr]

-# [Reply hidden for Discord Gold non-subscribers. Click here to see less]

[fiery gull]

Try glm 5

[surreal zephyr]

Fails miserably

[fiery gull]

Wow, in my tests and in my use glm 5 is always better than gemini 3, I'll test this 3.1 like is very good then

[whole swallow]

This one with the VR would go crazy

[surreal zephyr]

Yea

[molten cipher]

woe

[surreal zephyr]

https://docs.google.com/document/d/1r-QPxJZ5pAAiY4sVahT2AVgm5Y4Tofyh8CLow8eAHXI/edit?usp=drivesdk
All 4 most impressive ones

[molten cipher]

lemme see

crazyy

super cool

[warm drift]

Is it a common bug that the model freezes while generating a response? For me, it probably happens every few messages.

[stray aspen]

How do you do this

It only gives me crap

[surreal zephyr]

Good workflow ;)

[red sluice]

captchas are getting harder and harder those days

[red sluice]

Any idea on the model that gets you that error?

[echo aurora]

It's a known bug where models can be stuck in an infinite generation state, but it shouldn't be happening every few messages. Can you try the steps here - https://help.arena.ai/articles/8691588590-troubleshooting-infinite-generation

Also worth trying signing out and signing back in when this happens. I haven't personally see this work, but have heard from a few users this helps, so is worth a try.

[whole swallow]

Captchas are getting harder OR humans are getting DUMBER??

[stray aspen]

Captchas are more difficult

[zealous sparrow]

Hey pineapple is gemini 3.1 pro codearena broken?

[unreal dagger]

Use capsolver

[echo aurora]

It doesn't look like it's broken. You'll want to try the steps in this article - https://help.arena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

[zealous sparrow]

ill check out the error code actually

[echo aurora]

I'll forward to #1447983134426660894

[deft spruce]

"error": "Bad next turn request in Battles in Direct mode. DIRECT was requested, but expected a BATTLE turn | sessionId: 019c7aee-6558-7127-a3af-82575eae7e2b | userMessageId: 019c7c16-cf19-7a5c-87bc-1fda4f09f853"
...SON OF A

[wicked sage]

hey pineapple i got a question
will you guys include gpt 5.3 anytime soon?

[zealous sparrow]

Maybe the error got fixed? It's going through more generations now and not erroring.

[zealous sparrow]

theres still no API

OpenAI is stalling

[wicked sage]

https://tenor.com/view/ishowspeed-ishowspeed-yeah-right-ishowspeed-yeah-ishowspeed-right-ishowspeed-stream-gif-2870397197282141259

ok im sorry i didnt know

also i just realised

if gpt 5.2 high is top 5, imagine where gpt 5.3 high is gonna be placed at

[zealous sparrow]

@echo aurora Odd issue, It either didnt record or i dont know. Randomly errored and network posted nothing

[wicked sage]

prob top 2-3

[zealous sparrow]

or i forgot to record.

wait

ill try to record once more.

nothing..

[echo aurora]

Want to note #ask-here

[proud bobcat]

Imagine it’s literally worse again

5.1 high beats 5.2 high

[echo aurora]

Are you able to copy/paste the prompt in #1447983134426660894 and explain further?

[wicked sage]

is that true

[toxic verge]

Yeah, that needs to be filtered better

You guys did a good job on the other stuff

[echo aurora]

@toxic verge going to ask we use #1447983134426660894

[toxic verge]

Ok

[zealous sparrow]

I checked network, and theres nothing for Gemini 3.1 pro on codearena, Ratelimits should post to network right?

[brisk mauve]

hello

[proud bobcat]

Lmao yeah

According to the leaderboard at least

They FUMBLED with 5.2 high

[bright kayak]

guys can the devs fix the timing out and erroring out 😭

[hoary elbow]

Hello, I’m back. I tested Gemini 3.1 pro and I just gotta say I barely notice a difference.

[proud bobcat]

Between 3.1 and 3.0?

[echo aurora]

If you're running into these problems I'd recommend checking out this message - #1417174113092374689 message

[sick mantle]

@echo aurora I think u already typed that in anocemets

[hoary elbow]

I used to use 3.0, 3.1 update is barely noticeable

[echo aurora]

Typed what?

[sick mantle]

That Video Arena is Going to be gone

[bright kayak]

thanks, what i mean in particular is that in some longer requests it says "something went wrong", and i'm not sure if this is my fault, but most of the time, opus 4.6 thinking says it edited the files when it didn't and just "deployed the project" without changing anything internally

[echo aurora]

This is different. Prior announcements were that the Video Arena in Discord bot was going away. But the channels are currently still accessible in the archived section ( #video-arena-1 #video-arena-2 #video-arena-3). Today's announcement is explaining that these channels won't be accessible on Monday.

[echo aurora]

I'd say either the chat requests you've made are thinking/generation for more than 10 minutes, in which case this will error out. Or if it's not, then the same steps in the linked message apply.
TLDR -> try the troubleshooting steps, check if it's rate limit, worse case report to the team the best information so we can take a look.

But the steps in the linked message will go into way more detail.

[bright kayak]

thanks a lot

[wicked sage]

clayton: hello my name is clayton and im made of clay

[echo aurora]

Clayton is an absolute rockstar

Truly a legend at Arena

[river whale]

@echo aurora

Do u work with lmarena or work in lmarena?

[echo aurora]

#ask-here

I'm really trying to get away from answering questions in #general

[wicked sage]

you guys should make a clay statue of clayton

[shrewd citrus]

It’s a good idea since I guess you see a lot of people still getting confused on why they can’t generate a video even though they see the channel

[wicked sage]

btw what server do you guys use for the emojis?

[echo aurora]

Yeah that's the intent

Hoping this leads to less confusion

[shrewd citrus]

there should maybe be a bot too

like I’ve seen that on other servers, when a bot detects a keyword or phrase

the bot would automatically respond with the answer

[echo aurora]

We have one in place, but sometimes it doesn't always catch everything

[shrewd citrus]

I haven’t seen it in action yet 😅

[echo aurora]

Well because you don't see it when it's working

You only don't see it when it doesn't work

[shrewd citrus]

ohh right it’s one of those “you can only see this message” type thing

[echo aurora]

(odd sentence to say)

Yeah

[river whale]

@echo aurora #ask-here

[echo aurora]

Going to ask to not do this I'll get to questions/pings when I can

[river whale]

what to not do?

shall I not ping you?

[echo aurora]

If you've pinged me in other channel with a question, please don't go to another channel and ping me there pointing to your other message

[river whale]

Oh okay

[echo aurora]

[river whale]

[wicked sage]

https://tenor.com/view/lying-wong-gif-7845920779899236663

[toxic verge]

https://youtu.be/_cuBL8uop18?si=F0AWj9jqgE3pqJFs

[wicked sage]

grok 4.2??

[jolly narwhal]

sonnet 4.6?

i just went into arenas coding

every model

is not

what they say it is

claude opus 4.6 is sonnet 4.1

🙏

[rugged abyss]

Why are so many people saying Gemini 3.1 sucks? For me it has worked wonders

[jolly narwhal]

cause it gives you outdated info

saying its 2024

like holy

real gemini vs fake gemini

good one @echo aurora

[sick mantle]

Nah its not 2024

[jolly narwhal]

you managed to fool everyone with your fake latest models

btw kling 2.6 is 2.5

🤡

[echo aurora]

Models offered via API vs what's offered on the model providers site could have different endpoints

[jolly narwhal]

kling 2.6 pro can make texts readable

but on the site

it’s gibberish

?

[wicked sage]

https://tenor.com/view/ishowspeed-ishowspeed-yeah-right-ishowspeed-yeah-ishowspeed-right-ishowspeed-stream-gif-2870397197282141259

[echo aurora]

Right. The models on our site are provided via an API

[wicked sage]

ok so this ver of gemini doesnt have search mode on

[jolly narwhal]

4.6 has less advanced coding then 4.5 on the site

?

[wicked sage]

so it basically uses its cut off date or whatever the fuh it was called and uses what it remembers from that time period

[jolly narwhal]

every model is more stupid then the real site

then the real Gemini would be cut off

[half mist]

Ok, the Gemini in Arena isn’t search grounded, so it doesn’t the current year unless you go to the grounding mode, and select Gemini, so it’s not fake, it’s just not grounded

[wicked sage]

like here for example

g3.1 is not here cuz prob outdated but still

[jolly narwhal]

fair enough

sorry @echo aurora

[wicked sage]

ok yeah basically what this guy said

[jolly narwhal]

but why does the 4.6 models not beat 4.5?

[wicked sage]

again
if it doesnt have search it just looks thro info that happened in january 2025 and tells you that

[jolly narwhal]

• sora models dont work in video arena tho

plus HOW THE HELL

IS VEO 3.1 fast

[wicked sage]

if youre talking about text arena ehh idk

[jolly narwhal]

SCORING OVER SORA PRO?

[half mist]

But I don’t know why it says 2024 when the knowledge cut off is January 2025, but I am assuming it’s because it’s right when 2025 started, so Gemini still thinks it’s 2024

[wicked sage]

honestly i dont even know

i dont use gemini anymore

[royal sail]

Knowledge cutoff is quite a simple topic

[jolly narwhal]

telling you it is not gemini 3.1

[wicked sage]

also im pretty sure some models like anthropic models say that they cant search

brb let me fact check

[echo aurora]

All good, no worries

[wicked sage]

yeah

[wicked sage]

wtf is ppl

[sick mantle]

Sora did good on this one ngl

[stray aspen]

bytedance is cooking

[sick mantle]

Yea with Seedance 2.0 its kinda good

[wicked sage]

wtf is bytedance also

[sick mantle]

The creators of Seedance 2.0.

[raw oyster]

seedance, dola(doubao)

they also made tiktok

[surreal zephyr]

no

[lament niche]

why

[surreal zephyr]

#video-arena-1

[echo sinew]

Note that Video Arena has been removed from the server. More information can be found in this announcement.

[lament niche]

ok

[celest orchid]

Guys, does anyone know any sites with Open Claw AI Agents without restrictions?

[trail jackal]

Hola

[edgy iron]

I have a subjective feeling that Sonnet is stronger in claude.ai itself. Is it really subjective, and if no, is it same for Opus?

[livid heath]

Why does this keep happening, there is no way to do anything about it, retrying does nothing. All I can do is leave it and move on with life

[edgy iron]

Have you tried starting a new chat? This is often an indicator of chat being too long (i.e. too costly on tokens)

[livid heath]

Alright, that would explain it perfectly

it did so on a little longer thab usual

[edgy iron]

You could try and ask Gemini to recap for a new chat for example

[lime rune]

mihai popa goat come back

@ocean bison goat ❤️

@old garden

[echo aurora]

I'd recommend the steps in this article: https://help.arena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

[hollow imp]

@echo aurora how's lmarena work-life now compared to before the release of opus 4?5

[inner relic]

I hope they fix dola

that model is overconfident

[dry siren]

I'm doing a direct chat but keep getting a thing to choose response A or B like every three responses. Is there a way to disable that? It's really irritating. If I wanted that I'd use battle mode

[hollow imp]

@echo aurora will this benefit the lmarena team?

https://x.com/claudeai/status/2024907535145468326?s=20

Claude (@claudeai)

Introducing Claude Code Security, now in limited research preview.

It scans codebases for vulnerabilities and suggests targeted software patches for human review, allowing teams to find and fix issues that traditional tools often miss.

Learn more: https://t.co/n4SZ9EIklG

X · Feb 20, 2026, 6:02 PM

[atomic lagoon]

Is gemini 3.1 good in coding compared to opus 4.6

[atomic lagoon]

This is cool

[hollow imp]

Ofcourse not

[atomic lagoon]

Its worth an ask

[atomic lagoon]

Makes sense

[pale sonnet]

i need seedance so bad

i just wanna make dozens of videos a day like kwebbelkop

[quartz pike]

guys i think gemini 3.1 pro is cooking. basically i sent in a prompt. it froze. and i forgot about it for a couple hours so yea

[celest orchid]

xD

preview moment

[rugged abyss]

Pineapple is shipping announcements like theres no tomorrow haha

[stray aspen]

my day is ruined

and i just started chaatting

[hollow imp]

Gonbe 😡

[quartz pike]

what did i do to bro 😭

[hollow imp]

Not using premium ai

[quartz pike]

tf is that

and is it free

[echo aurora]

Sorry what do you mean by this?

[rose kiln]

for some time my website is not working, every time i give it a prompt the command break, is it happening to me alone?? 👀

[hollow imp]

Allowing teams to fix issues that traditional tools often miss

[quartz light]

💀

[echo aurora]

I'm not sure tbh
It's best for everyone if I stay away from the codebase lol

[versed ravine]

bruh why is it sometimes making me choose between two models in direct chat bruh

[keen beacon]

it better be correct cause it went against the other grok search and they had two entirely different answers

[echo aurora]

We are currently experimenting with the occasional Battles in Direct - https://help.arena.ai/articles/8949646387-lmarena-experiments-battles-in-direct

We are hearing a lot of feedback from the community about this and are exploring changes.

[celest orchid]

Guys, does anyone know any sites like Open Claw AI Agents without restrictions?

I'm already tired of looking

[sick mantle]

@echo aurora Bro when are we getting the Change Video Models mode, like fr

[stray condor]

hey why when I try to attach a .pdf file (any size) into Claude Opus 4 - 6 It gives me this error message?

[lofty frigate]

I usef battle mode wrote a prompt then got the message the prompt violates the policies, open a new one wrote the exact same prompt then it worked... wtf

[toxic verge]

https://youtu.be/2_DPnzoiHaY?si=rKwoJQlFrgS7cb_H

See I’ve been saying it

[ocean vortex]

relatable

I almost never read those in servers, unless there's some specific update I'm looking for

[ocean vortex]

or like read them at least if I'm active in one place (server) for prolonged periods of time. Lately I'm somewhat less active here 👀

[spark python]

Have yall seen what pika released?

[stray aspen]

glory to anthropic

[echo aurora]

I don't think this is going to be related to the upload as a different error would appear, but would recommend these steps: https://help.arena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

[echo aurora]

Yeah sorry about that, bit heavy on the announcements today

There shouldn't be anymore today though

(I think, don't hold me to that)

[compact flame]

Why 10 minute limit exists on ai generating response just asking

[rugged abyss]

You never know with these AI labs these days

[echo aurora]

It's a tech limitation

[echo aurora]

Oh the stories I could tell

[compact flame]

Oh alright

[stray aspen]

bro battles in direct chat need to removed

[compact flame]

Just give them your feedback man

[toxic verge]

We need independent benchmarks

[rugged abyss]

https://tenor.com/view/angry-gif-6891886407904187638

[stray aspen]

i wanna chat with claude 4.6

not with 2 anonymous models

[compact flame]

Just give them your feedback be generous

[toxic verge]

The newest model get the most scrutiny

[stray aspen]

sigma from ohio

[toxic verge]

And post it a month later after the results to see if they Nerf it or not

[compact flame]

I can feel the ram prices skyrocketing

[echo aurora]

We want critical feedback, it's important for us to hear these things

[toxic verge]

I got 2 weeks left on Claud

Before the down hill cycle begins

[compact flame]

I can probably tell in 3 years 1 gb of ram will cost 5k$ atp

[toxic verge]

Go to Gemini Reddit. lol

[rugged abyss]

What kind of feedback is actually the most valuable for the team?

[toxic verge]

Honest feedback

[compact flame]

Well maybe not

I don't know but ram got really expensive

[toxic verge]

Ai made everything expensive but slop

[rugged abyss]

Well obviously but feedback to how models are presented, to data presentations?

[toxic verge]

Sometimes the best type of truth isn’t something that could be articulated unto data point

[compact flame]

Ai was made to make life easier. But no we got brainrot

[toxic verge]

Yeah, well it’s here now so we gotta deal with it

[half mist]

@echo aurora I resign from being a beta tester. This sucks

[toxic verge]

https://youtu.be/hWq1j-DaNYs?si=WxDcuZ9eOqHhq1DV

Look at this 😂😂😂

[compact flame]

Everyone got it bro

[compact flame]

Can I get small context? I don't exactly understand English well

[toxic verge]

You know the prom here is?

People comparing the human brain thinking to machine thinking

[compact flame]

Oh alright

[toxic verge]

It’s like trying to think like a dog

A completely different way of looking at the world

[compact flame]

Okay got it thanks

[echo aurora]

Not so much on things like model performance or things specific to the models, as the voting should reflect that. But overall feedback about our platform and things we can change we want to hear about.

It also helps when it's explained why. Saying "I like X" is helpful, but saying "I like X because Y" is more helpful.

[toxic verge]

Not try to compare human brain to a machine brain

[compact flame]

I wonder why are some models are limited to battle mode only

[rugged abyss]

What about more obvious things like slow AI responses? Thats one thing I've noticed every time I've used Arena, I know processing power is limited but waiting 13 minutes for a small response seems unreasonable right?

[stray condor]

@echo aurora thanks for the previous help, but I just had to wait a bit.

[toxic verge]

That’s assuming we understand anything now 🙉🙈

Not the internet lol

[compact flame]

Pineapple is typing something big I assume

[toxic verge]

Or tv

[echo aurora]

It's still good for us to hear this. A lof of the time it's things were already aware of, or things that are out of our hands to fix. But it's still helpful for us to hear as it makes it clear we're hearing this feedback. The last thing I want is for people to not bring up feedback/issues/etc with the assumption it's something already on our radar. Even though a lot of the time it already is, for the cases it's not that'd be a shame not to hear kind of thing.

[glacial swan]

Could you please tell me why 3.1 was removed from LMArena? Will it be brought back? Does it make sense to use it in Studio

[compact flame]

It's still here

[echo aurora]

It's just at the bottom of the dropdown list

[compact flame]

It's just at the bottom for some reason

[glacial swan]

loll thank you

[toxic verge]

🤣🤣

[celest orchid]

um..

[glacial swan]

Apparently its results are that terrible? If it’s dropped down, those who tested it – what are the pros and cons? Is 3.1 cool?

[compact flame]

It's still a preview

Of course it's still in testing

[echo aurora]

Which I'm not sure why that's the case will flag

[half mist]

Battle in Direct is so bad since it makes your chat sloppy since it uses the context of the battle in direct for future answers, and ruins the chat

[rugged abyss]

I think most people who say its bad cant prompt well, it has amazing outputs on specific prompts.

[compact flame]

Just start a new one

[stray condor]

Does anyone else have this problem where suddenly after lots of code is generated (in 3 - 6 responses), the AI just gives up and a error comes up?

[half mist]

That’s gonna get annoying after a while

[glacial swan]

yes

[compact flame]

Fair point. Especially going to the search to deleted archived chats

Like cmon I want to delete my chats immediately

[rugged abyss]

#1343291835845578853

[stray condor]

I'm just asking if it's typical (that's why I asked if somebody else got this problem), not that I want to solve this issue 🤍

[toxic verge]

This what I mean

The reality on the ground is so much more different than the benchmarks

[compact flame]

Man this server got 200k+ people yet chat seems so empty sometimes

[toxic verge]

It’s like two different worlds

Sometimes I wonder if people are looking at the same thing

[rugged abyss]

Oh i just thought it would be smart to report the issue

[toxic verge]

Lots of discords like that

Cuz ai community is heavily moderated

[compact flame]

I swear sometimes servers with 5k people seem more active

[rugged abyss]

I really think prompting is a huge part of Gemini 3.1 Pro, whilest it has issues at heart (see Vending Bench 2 for example). Most issues dont appear if you prompt specifics. I've been testing this model quite thoroughly and have only has slight hickups whilest it managed to output some quite impressive outputs (Opus 4.6 or even better results)

[toxic verge]

It’s because the communities are heavily moderated

[celest orchid]

Hey @echo aurora , when are the contest results?

[toxic verge]

And people are so technical people are just keeping to themselves

Not banned but politically incorrect

Or whatever you wanna call it

People don’t take their view seriously

[dry siren]

Ugh sometimes this site gets stuck in loops of Security verification for no reason

[compact flame]

I wonder why do people hate on ai sometimes

[toxic verge]

And people act like they’re smarter and so it makes people less hesitant to wanna share their ideas cause they don’t wanna feel stupid

There’s a lot of reasons to hate on it lol

[compact flame]

Examples?

Just asking

[toxic verge]

I don’t wanna sound like a broken record, but it’s a dirty industry, bro

And doesn’t do itself any favors

In the public perception domain

[compact flame]

I guess you got a point

[toxic verge]

People use it how it was designed it’s working as intended

[compact flame]

Yeah brainrot still seems not to go dry

[toxic verge]

Do you know I feel like if you really truly believe in passionate about something

You gotta be as equal as scrutinizing

Otherwise, you’re gonna get blindsided

And you gotta be able to see if it stands up to the scrutiny

And if it doesn’t then it’s bs

[dry siren]

If this thing asks if I'm a robot one more time I'm gonna lose my mind

[compact flame]

I'm gonna ask you myself

[toxic verge]

See this is what I mean

[compact flame]

Are you a robot

[toxic verge]

😂😂😂

[dry siren]

Beep boop

[toxic verge]

Welcome to the future

[compact flame]

Sounds human enough

[toxic verge]

Where the machines scrutinize your humanity

I find that very disrespectful

A human having to verify themselves to a machine

Kind of makes me prejudice to be honest lol

[compact flame]

Imagine ai asking us to do things in the future

Coding for ai

[toxic verge]

Bro, it’s already coming

It’s gonna come to your work to monitor you’re working

[compact flame]

We are so cooked

[toxic verge]

It’s gonna come to the public restrooms if you forget to flush, you’ll get a ticket in the mail

[dry siren]

I haven't been able to use my chat for ten minutes cause it keeps non stop asking for verification

[toxic verge]

We don’t have to be, but yeah, we will be. I mean there’s just no way around it.

[compact flame]

Atp ai is just gonna watch us when we are pooping

Feels uncomfortable

[toxic verge]

Yeah, that’s the only thing that really kind of bothers me. It is kind of super invasive.

Really high-tech and invasive

[compact flame]

And imagine it gonna criticize us for how we poop bruh

[toxic verge]

Why not?

If you build it, they shall come

[undone geyser]

it keeps generating image, how do i cancel? plug the wifi off?

it keeps generating image, how do i cancel? plug the wifi off?

[toxic verge]

https://www.youtube.com/live/CyyL0yDhr7I?si=nFfMcLfI5n1LSKzq

[compact flame]

Just delete the chat ATP

[trail relic]

is there any chance we're getting 4o back

[toxic verge]

No

Retired

[trail relic]

hell nah

[dry siren]

Ugh I tried a new chat and still getting the security loop. I'm getting mad

[toxic verge]

Login In and out

[compact flame]

Take a chill pill I guess

[toxic verge]

Clear data in browser

Take a 5-10 minute break come back and try again

[compact flame]

Russians got gigachat. Definitely not gigachad

[grand cliff]

Kimi 2.5. Getting used to it, just not the same as 4o.

Opinion? Can't come up with proper names Social media wise unlike 4o. Seems generic

[celest orchid]

kimi is donation 💩

[compact flame]

@echo aurora

[celest orchid]

<@&1349916362595635286>

Thank You

[celest orchid]

Thank You

[compact flame]

Now we wait

Oh it's done

[celest orchid]

Thank You

[echo aurora]

Thank You

[celest orchid]

nice, we did it guys

<@&1349916362595635286>

@echo aurora

Thank You

[trail relic]

Is Anthropic from claude or something?

[echo aurora]

[trail relic]

keeps popping out in my old 4o chats

[echo aurora]

All these scams

[celest orchid]

lol

[echo aurora]

AHHH

[compact flame]

@echo aurora

[celest orchid]

FINAL BOSS

[celest orchid]

THANK YOU DUDE

[compact flame]

Damn

[echo aurora]

I'm not moving from this channel for the next like hour

[compact flame]

Scam attacks are getting too insane

[celest orchid]

💀

[echo aurora]

I'm not sure how they've been getting around the filter lately too

The problems that come with being a 200k+ community

[celest orchid]

I think we should create an IQ test before entering the server

Thank

[echo aurora]

The problem is I think a lot of these accounts have been in the server for awhile

[compact flame]

Or at least I dunno level system

5 levels to send a image

[celest orchid]

Then, before each message you send, take an IQ test

[echo aurora]

But what if I self-ban 😭

Yeah that's the filter working!

[celest orchid]

The pineapple has a defensive position at the front

[compact flame]

Pineapple maybe level system would be good against scam images

Maybe

[celest orchid]

We need to embed an AI agent into the chat that will ban scammers

[echo aurora]

We have this in place, it doesn't always catch everything tho

[compact flame]

I don't see any levels here. I dunno if you mean something else though

[honest verge]

Gemini 3.1 pro is nerfed in app already

[compact flame]

Like I meant lvl 5 to send an image

[sick mantle]

What the helly is happending

[echo aurora]

Ohh sorry, was thinking something else

[honest verge]

Ai studio still has a superior version

[compact flame]

It's okay

[echo aurora]

Mean people being mean

[sick mantle]

Thats the same XD

[echo aurora]

¯_(ツ)_/¯

[undone geyser]

@echo aurora why some chats it keeps saying "generating image"? i cleared cache, deleted cookies, nothing

[compact flame]

Delete the chat from archives I guess

[hollow imp]

So there's a server called dr stone which is the community of the fans of the anime dr stone and before joining the server they have a question of who's the main character of Dr stone ( this is an important question ) and bots incorrectly answer and incorrect answers give u muted role

[echo aurora]

I'll have to check this out, do you have a server link by chance?

[thorny schooner]

Have they mention anything yet about the i\direct chat

[echo aurora]

I do think this would help, but also a problem with these scams is they're essentially stealing other user's Discord account, a lot of these potential scam/bots could are already in the server sort of thing.

[echo aurora]

Sounds like it's still generating, no?

[undone geyser]

sure is

lemme check a sec

nope still keeps generating

[hollow imp]

Yes

[undone geyser]

i have archived that chat, so i guess only delete remains? or can it be fixed by waiting?

[hollow imp]

@echo aurora dmed

[echo aurora]

Could have hit this bug - https://help.arena.ai/articles/8691588590-troubleshooting-infinite-generation

[undone geyser]

ok tysm

also hate the arenas on direct chat but ok xd, leave it as is

[stray aspen]

grok 4.2 is gone

[meager tinsel]

@echo aurora is it possible to somehow remove this requirement for new users, I don't know how to get rid of it, specially since clicking on it brings me to the channel where we can no longer generate a video lol

[modest prism]

Code arena is basically broken. I tried with multiple models such as opus 4.6 and Gemini 3.1 pro and always got this error: something went wrong.

[rain zinc]

ong ever since they added battles in direct everything’s working so much worse getting so much more errors frequently with no fix forcing me to make new chats, like pretty much everyone hates battles in direct chat but they won’t do anything lol

[gloomy onyx]

dude laugh~1

ODYSSEY_PepeThink

[night moat]

gemini 3 pro error 😤

[echo aurora]

Note that Video Arena has been removed from the server. More information can be found in this announcement. Video Arena is still available on the site -> https://arena.ai/video

[echo aurora]

You'll want to try these steps: https://help.arena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

[meager tinsel]

Not what I was asking 🙁 On Discord, for new members, there is still a "Task" they have to do (As shown in the screenshots). If I don't do that task, that black box stays there permanently until it is done.

It is something that needs to be changed in the Server Settings specifically.

It is specifically under the "Getting Started" tasks...

[hollow ivy]

How big is the context-length of Gemini 3.1 pro preview in arena?

Does the arena have a limit, or does it just use the model's limit?

Is Gemini 3.1 the best google-model for immersive and realistic stories?

Have they now beaten Opus-4.6 in that area?

[echo aurora]

OH

Understood. Thank you.

[meager tinsel]

No worries 😄

[golden ocean]

I LOVE CLAUDE OPUS 4.6 THINKINMG

[hollow imp]

@echo aurora my arena champion role please

[shrewd citrus]

same!!!

I LOVE it when it says “Something went wrong with this response, please try again.”

[raven quartz]

guys how to create video before they have

[golden ocean]

cwaude still thinking

[compact flame]

Ai having a mental breakdown on trying to find a easier solution but ended up in a loop

[raven quartz]

how to make video before i can

why i cant make video guys in video arena

[compact flame]

Because it got moved

To the website

[echo aurora]

Welcome back! Sorry again about that 😭

[analog dove]

I think I'm missing something here.. how do I get to generation of videos ??

why can't I see the channels?

[echo aurora]

Note that Video Arena has been removed from the server. More information can be found in this announcement. cc @raven quartz

[raven quartz]

where i can find

[echo aurora]

https://arena.ai/video

[queen veldt]

Yo

https://chatjimmy.ai/

Tf is this speed

17k t/s

[gleaming roost]

😲

[hollow ibex]

how to create video?

[quasi atlas]

@hollow ibex Note that Video Arena has been removed from the server. More information can be found in this announcement

[hollow ibex]

ok thx

[near stream]

Where can I get my questions answered about this?

[toxic verge]

Guys all jokes aside

😂😂😂😂

[thorny schooner]

......... This is how many chats i have to go I got go through in a day in the span of maybe a few hours Max around 3. Keep in mind that all of these are individual chats🥲

[toxic verge]

Let’s see the results

[keen beacon]

fk him

[keen beacon]

WTF

YOU'RE CRAZY

[echo aurora]

Are you able to follow the steps outlined in this article from the If the problem continues section and let me know if/when a jamdev is submitted? Max shouldn't be resulting in that error. We'd love to see more information about this error.

[thorny schooner]

It's because of the error that has been happening with that has only increased in frequency with the whole battle in direct chat stuff stuff

[echo aurora]

The Battles in Direct experiment shouldn't be causing an increase in errors. I assume it's the Something went wrong you've been seeing? Is this with a specific model?

[queen veldt]

Battles did create a new error tho

When you prompt your chosen model after the battle thing sometimes it doesn't work

So i have to retry until it eventually fails

(i send a message and it appears as I'm not sending anything i noticed that when i refresh the page)

[thorny schooner]

☠️ did no one see the picture I posted like a day again that said something about increase amount of glitches like this ( i know the picture i show is not in battle mode but it's basically what happens and depending on what happens sometimes I can't even do a retry like what the guy above me just said0

[queen veldt]

And i keep sending something and refreshing the page until it says it failed to generate

Than i just do retry button and it continues

[near stream]

Could it be a screen recording with a link to the drive?

[echo aurora]

Yes, but if you could be sure to include the Dev Tools > Network Tab that'd be very helpful. The jamdev includes this which is why it's preferred.

[near stream]

Exactly how could I do that? (I don't know)

[echo aurora]

Wanted to cross post this message as there has been conversations happening in a few different places and I'd like for this message to travel wide -> For the Battles in Direct experiment we're in the process of rolling this experiment back. We plan to develop this experiment into a better state before releasing again (as an experiment), but the current version is being rolled back.

[echo aurora]

What browser are you using?

[near stream]

Chrome

[proud bobcat]

Qwen 3.5 cooked

They cooked hard

Really good model

[echo aurora]

1. click the three dots (top right corner), select More Tools > Developer Tools
2. at the top you'll see a Network tab, open that
3. run a prompt in Arena where Max errors out
4. you'll see a file that has the Eval ID (Eval ID is the random set of numbers/letters in the URL)
5. open that and you'll see a Status Code & Response window.

Those two areas are really helpful to us to understand what is going wrong.

Reminder to only share this recording in the form in the this article, please do not share it here in this server.

Lastly, want to mention this process is not ideal and not okay. We recognize this is asking way too much for us to get better information on this error. We have plans to built out a system that can handle this much easier for the user.

[fiery gull]

M2.5 >>

I'm in hate so big, the glm 4.7 flash opus 4.5 thiking must be so smart but my laptop is a potato

[fiery gull]

Hello

[undone saffron]

I know how it feels

[fiery gull]

Here is a ryzen 7 7735hs and 24gb option, I don't know if this would run the glm 4.7 flash well

[undone saffron]

Check ram usage

[fiery gull]

I currently have only 16gb and i3 1215u, it works 100% for my use but it doesn't run glm 4.7 flash nor screwed

Maybe I should buy a notebook with 7735hs and rtx 3050, I don't know

[undone saffron]

16 GB of ram is not enough
At least 64 GB min to run "smart" AIs

Furthermore, if they are thinking models, ram usage is higher

[frosty lava]

Someone know why like 30% of my ram is allocated to my gpu ?

how can i disable or reduce this

[whole sundial]

you may be able to use a q4 quant with that as long as an insane amount of vram isn't allocated to your igpu

[thorny schooner]

How y'all doing

[whole sundial]

16gb of ram though, don't try unless you want less than 1 token per second with a good quant or usable speeds but a terrible model

[fiery gull]

I already think the qwen 4b 2507 has some intelligence, the glm 4.7 flash no thinking iq4nl would be enough for me

[wheat oak]

Hi... how can i make videos?

[undone hull]

https://arena.ai/video

[scarlet spire]

Disabling isn't possible. Reducing it might be inadvisable especially if you'd be restricting it to prohibitively small sizes. Check your firmware setup pre-boot executable environment. Either the setup itself or another pre-boot ex env tool might hold some answers.

[echo aurora]

Would note if you're looking for the Video Arena bot it's been removed from the server.

[toxic verge]

https://youtu.be/iQV-NXTRtAI?si=E5Rl_nclwWlSOKGx

Told you guy

[hollow imp]

@quartz pike

https://cdn.discordapp.com/attachments/1437882067554795611/1474595904139952318/image.png?ex=699a6bf4&is=69991a74&hm=b8ea079d1418ba0bc3301087da6a2b99e18e7202c309b0adcb0a80a17e8d8edd&

[toxic verge]

Claude 5 gunna be insane

Same with gpt6

[river whale]

yoo

[river whale]

we gonna have to wait more 6 months for it

[stray aspen]

Wheres geminine

[river whale]

theres no geminine

[toxic verge]

ILYa was right

[river whale]

?

[toxic verge]

Data-hungry A.I. developers, which have already sucked up mass amounts of online information from the internet, are starting to hit roadblocks from website owners. Between 2023 and 2024, 5 percent of all data and 25 percent of data from the highest quality sources were restricted across major A.I. datasets, according to a study from the Data Provenance Initiative.

Less gains but more expensive

Look at the games from ChatGPT 2 to 3 all the way to four

[stray aspen]

Wheres geminine tho

[toxic verge]

This is a classic diminishing return To get a 10% gain in intelligence, you might need 1000% more data, but that data no longer exists on the "one internet" we have. At least the little hanging fruit.

[cursive gyro]

Isn't this graph showing the opposite of that? That much of the largest task time success gains on METR are in the most recent period?

[toxic verge]

If Googleis spending $185 billion this year (which they just confirmed in their February earnings call), that graph is basically the receipt. It proves that the "Data Wall" isn't stopping progress it’s just making it incredibly expensive to "buy" your way through.