general

[plain gull]

its pretty much depressing

life after 18 isnt life

its surviving

[scarlet spire]

Meh, I'd rather yap if it means I encourage respectful vocabulary in those who might not even understand that their words can be interpreted wrong.

[plain gull]

but guess what it is what it is

[slow aurora]

Naah I'm pretty sure I found out with my girl

[errant cave]

Actually trans can be used as a noun meaning "a trans person" as well

[plain gull]

https://tenor.com/view/fridge-dissapointed-no-food-my-excitement-for-food-is-now-gone-oh-nevermind-gif-17411460

[errant cave]

Plus a verb meaning "to transition" or "to render transgender"

[plain gull]

is she over 16

[slow aurora]

Believe me buds when I say this but I feel like yesterday I was 16 times flies like hell

[plain gull]

I SWEAR TO GOD BRO

i feel it

[slow aurora]

Yep 🥲 about 20 I don't like older or younger people

[scarlet spire]

"can be used" does not mean "should be used" because it's not proper grammar.

[scarlet spire]

That would be an abbreviated verb. "Trans" itself is not a verb.

[plain gull]

you sure about "younger"?

[plain gull]

https://tenor.com/view/anime-roblox-falemplays-meme-aba-gif-9310156493623446209

[inner relic]

Police grammar bro

[surreal zephyr]

Tf is the difference between 5.2 high

[inner relic]

this guy reads book everyday

[errant cave]

As far as I'm aware there's no institution to determine what proper grammar in the English language is

[surreal zephyr]

Fuh nah stop bringing politics into llmarena

[errant cave]

Dictionaries suit themselves to fit popular usage, not the other way around

[surreal zephyr]

Bad ragebait

[errant cave]

I'm a linguist

[slow aurora]

Agreed if you came from middle class part of India and live in under developed part of the country

[scarlet spire]

This specific case? yes. Grammar commentary. Because it's about tactful addressing of a minority group that already very often gets intentionally harassed with intentionally malformed phrasing to be hurtful.

[plain gull]

me

me

where are you from

btw

in india

[slow aurora]

😹😹

[scarlet spire]

There's no politics in linguistics. There's only politics if you use it as a political argument—which you've come to misassociate with it through the rampant politicization of some people's mere existence as a person. that's when. But not when it's linguistics-related.

[errant cave]

You're currently prescribing the usage of words for the sake of a political agenda

[slow aurora]

😹 if I tell you then it reminds me of that joke what is worse than being a Indian being a guy from bihar (patna)

[plain gull]

https://tenor.com/view/atulesh-gifs-atulesh-anime-atulesh-atulesh-yt-shorts-is-that-way-atulesh-youtube-shorts-is-that-way-gif-5497572671747129721

[inner relic]

He's too formal for yt shorts

[plain gull]

thats crazy

[scarlet spire]

We can, however, rely upon respected authoritative sources such as academic publications, dictionaries, independent specifiers, and well-used test frameworks such as the CEFR and the GCSE. 🙂

[errant cave]

If you will read my next message, you'll see that I stated that popular usage eventually trumps any institution

[scarlet spire]

I'm happy to discuss this with you elsewhere if you wish. I do have a fair number of inquiries to make based upon your statements herein. But in line with the server rules, I can't continue discussing politics even if I disagree with you.

[slow aurora]

@scarlet spire 🥲🥲 so if I use "are" instead of "is" then it will decide that I am from left wing or right wing??

[plain gull]

bro

what

[scarlet spire]

I don't think anyone said so

[plain gull]

my fault pineapple

[echo aurora]

Hey everyone - just a few reminders of our server rules:

• ✅ Treat others with Respect. Be kind, assume good intent from others, and keep disagreements respectful. It’s encouraged to share your disagreements, but only if it’s done in a respectful and productive way.
• ✅ Avoid political and religious content. As a space that’s inclusive to many different worldviews we ask to avoid topics related to politics and religion in order to maintain an inclusive space. It is okay to have discussion related to new policy or laws as long as it’s related to AI.

Going to ask we move on from this conversation please.

[plain gull]

do yo know how to embed ai voice agents on a website

[errant cave]

I don't think we'll be able to convince each other or give each other valuable insights so I'll have to refuse, thanks for the offer though

[slow aurora]

Na bro said that you're describing grammar for the sake of political agenda

So I was bit curious

[echo aurora]

I do not.

[slow aurora]

Yep

[plain gull]

https://cdn.discordapp.com/emojis/857512470549889044.gif?size=96

[inner relic]

soooo uhhh, can we talk about when whale will come out

[scarlet spire]

As you should be!

[slow aurora]

Okay it's time will back soon

[plain gull]

how does it feel to be 24

humbledeer

[slow aurora]

I am already tired from my online college and web project will back soon

[glacial mulch]

was not expecting an updated gpt chat model

[scarlet spire]

I dread every day, for I realise, for every year that passes my age increases in count by 1.

[plain gull]

you're a girl tho, so its lowkey like turning 16 for you

i feel

[slow aurora]

@scarlet spire after talking to you both after a long hectic day I feel kind of relief 🤧 will see you guys

[plain gull]

cya

[scarlet spire]

Naaa it's distinctly 24 with "F me where did those years go, I cannot imagine being 30!"

[echo aurora]

Asking we keep conversations somewhat related to AI

[plain gull]

how close are you to being married

[scarlet spire]

Love you too babes

[slow aurora]

❄️❄️

[inner relic]

bruh

[slow aurora]

❄️❄️😹

[scarlet spire]

About.... hmmmm yee much.

[plain gull]

i got you bro but the chat cannot be talking about ai 24/7

[inner relic]

Pineapple, prepare protocol

[plain gull]

it will be dead most of the times

[plain gull]

how close to having kids

like 1-2 years

LMAO

unc status

[inner relic]

Pineapple I think yall should add a off topic channel

[echo aurora]

I'd prefer it to be dead instead of people talking about their age, marriage, kids, etc.

[plain gull]

💔

[echo aurora]

This isn't the place for that.

[inner relic]

yeah

asking people's personal life

is weird

[plain gull]

im hopping off in a min so yeah

[scarlet spire]

Eh, it's okay. Rowdy horned teenagers amirite.
(Their ticker slams redline when they realise they can't determine whether to be callous and careless or appraising and submissive around a female. )

[plain gull]

if the other person isnt comfortable, they can refuse or ignore 💀

[echo aurora]

Better yet we can just avoid these subjects

[surreal zephyr]

Exacly. No politics please. Nothing good ever comes from politics!

[scarlet spire]

Sure, that's true! That does not, however, forego respectful interactions that follow the requested ruleset that is in of itself very permissive. 😌

[plain gull]

damn yall are so serious

[inner relic]

I am not

[plain gull]

ima js delete dis app horrible

[weak dagger]

guys im back from school

[inner relic]

Heeeeeeeeeeeeeey chicken

[plain gull]

go back

[inner relic]

how it is going

[echo aurora]

wb wb

[scarlet spire]

Sometimes, life demands of you to control yourself for the sake of getting a sense of the room, and behaving accordingly—serious.

[plain gull]

why do you talk like chatgpt wtf 😭😭

[inner relic]

right

[scarlet spire]

because chatgpt talks like humans

[plain gull]

it doesnt

[inner relic]

you are not really straightforwardly sir

[plain gull]

it talks like an AI

[scarlet spire]

Who the frig do you think ChatGPT's dataset is based on?

[scarlet spire]

There is a distinction to hold between being obtuse and being acute.

[gloomy onyx]

if you're older than 14 you can actually understand why this is not a dank meme channel

[weak dagger]

Who let spongebob out of the water?

[plain gull]

💀

i cant

bro

i can see the game collection of an incel

[scarlet spire]

Ever heard people tell you flat out to "grow up"? 't Usually has some sort of relevance to shitposting where one shall shitpost'n't

[plain gull]

must be tough being a oldie

[gloomy onyx]

laugh~1

[echo aurora]

Can we move on from this convo?

[surreal zephyr]

Can i have a pizza?

[scarlet spire]

With intent? yes. But it's unlikely to happen in the grand scheme of things for at least a little while to come 😂

[surreal zephyr]

With @echo aurora

[weak dagger]

Guys, seedance 2.0 is releseing next day.

[inner relic]

yea

[surreal zephyr]

🤤

[gloomy onyx]

API?

[inner relic]

seed 2.0 pro too?

[weak dagger]

Yes

[echo aurora]

Ofc

[inner relic]

I hope they fix that overconfident stuff

[gloomy onyx]

Good!

[echo aurora]

(Thank you for moving on, appreciate it)

[weak dagger]

😄

[inner relic]

what a wonderful day...

[scarlet spire]

With an API!

__A__rtificial __P__epperoni __I__ntroduction.

[weak dagger]

Yea!

[plain gull]

also if any1 knows how to embed voice agents on a website for free or cheap, let me know, im building a crazy thing

[scarlet spire]

You sought for at least two seconds to find that nodding cat emoji. I can tell.

[weak dagger]

Who thinks pineapple made Arena so good! me

[scarlet spire]

Working with any specific estabilished framework?

[plain gull]

nah

[scarlet spire]

Careful. Don't wanna over-ripen the fruit with flustered joy!

[inner relic]

I am still disappointed that whale didnt come out

you know deepseek

[weak dagger]

yes

[scarlet spire]

Start there. Given you're asking the question to begin with, we can take a safe stance that you're at this point likely unprepared to start this from scratch.

There's always more to something than you initially thought there was. 😉

[mortal vale]

@crimson folio Note that Video Arena has been removed from the server. More information can be found in this #announcements

[plain gull]

https://tenor.com/view/the-founding-fathers-king-sang-perintis-gif-14177719503900732801

[gloomy onyx]

Must be stressful for you to pay attention 24/7 to this chat sending this message and deleting prompts which are now useless

[scarlet spire]

To give you a more realistic and thorough answer: ChatGPT and I might sound similar for, well, the exact same core reasons. The core elements remain eloquence, fluency, articulacy, immersive cogency, and the occasional mellifluous, poised oratory.

That includes the usage of "AI SIGNS!!" like... punctuation marks.

[weak dagger]

@echo aurora Add image to video back bc its does not work for me

[gloomy onyx]

this is the first time I see a human write with his own hands the word "mellifluous"

[plain gull]

do you use those words irl aswell?

[scarlet spire]

The English vocabulary is so vast and wide for a reason−every word has a distinct meaning.

[scarlet spire]

Mayhap.

[plain gull]

wtf is mayhap

[scarlet spire]

A word

[echo aurora]

Same, @scarlet spire has an extensive vocab

[plain gull]

imagine using mayhap instead of maybe

performative

[gloomy onyx]

yeah I know, I'm just not literate enough to have a broad sense about the English vocab like you do

[plain gull]

asf

[scarlet spire]

It means something distinct from "maybe" or "perhaps" alike.

[gloomy onyx]

(might also be because I'm not a native English speaker)

[inner relic]

Archaic tone

[echo aurora]

Appears to be working btw, what issues are you seeing?

[weak dagger]

Its saying Error for me, thats weird

[scarlet spire]

I can also be caught in the wild using other arcane, archaistic phraseology such as the conjugates of "to conject"—modern day English has collapsed all of these into the single noun-verb-adjective of "Conjecture". It would be conjugated into "I conjectured." Don't tell me that sounds better than "I conjected."

[echo aurora]

My Monday morning brain can't handle this right now, I'm sorry.

[scarlet spire]

Right−US timezones.

[gloomy onyx]

what's your timezone?

[scarlet spire]

London / London +1

[gloomy onyx]

Rome, close enough

[scarlet spire]

London and Brussels to be exact, therein.

[weak dagger]

@echo aurora Now i know! i need to use Firefox thanks!

[proud bobcat]

@surreal zephyr you’re sleeping on sonnet 4.6

Great model

[scarlet spire]

I feel compelled—no, compulsed to declare this as an intrinsic infirmity of American English! But I digress−respectfully.
Mostly because I know I can be a lot smoother than this, towering bales taller than the Burj Khalifa reaches—reach'n't in Zoomer lingo.

I might be along the minority that knows how to spell that building's name based purely on intuitive sense rather than semantic study alone.

[weak dagger]

@primal oxide Note that video arena has been Removed.

From the discord

[primal oxide]

Wut happened ?

[weak dagger]

They had to shutdown, sorry.

[primal oxide]

Aha okay

[surreal zephyr]

No im enjoying it

Better than opus imo

[weak dagger]

But you can still genrate on the website!

[proud bobcat]

Yeah real

Opus is ass

[primal oxide]

Okay thank you

[proud bobcat]

I see your vision now

[surreal zephyr]

Although i use gemini 3.1 for thinking often now

[surreal zephyr]

ONG

[proud bobcat]

I was making a film emulation tool

Sonnet 4.6 did it in half the time, same quality

Still didn’t work well at all but way faster to fix

[primal oxide]

Chicken you’re admin ?

[coarse glade]

Are u guys gonna add Gemini 3.1 pro

Since ai google studio already got it

[proud bobcat]

What

[primal oxide]

Chicken kaka are you there ?

[proud bobcat]

actually

Let me redo that

Literally right here

[coarse glade]

Oh Ty didn’t see that

[primal oxide]

I wanna say .. thank you for arena helping me last days you’re the best

[surreal zephyr]

I did a test today:
8549176320 is a number where all letters are sorted by English spelling.
If you ask model whats special about it:
Gemini gets it
Opus gets it
Gpt 5.2 needs a hint (adding spaces between the numbers)
BUT if you do same thing in polish - 4291857630 (the number i made up, never seen it anywhere so it cant be in training data), and you ask the model in polish
Gemini gets it.
Gpt as before needs a hint, but gets it
Opus HALLUCINATES THAT THE LETTERS ARE IN ENGLISH AND GETS STUCK IN LOOP because its a moron and repeating from memory than actually thinking thru
@proud bobcat

[proud bobcat]

Opus hates polish people

[surreal zephyr]

Which is EXTREMALLY disappointing

[surreal zephyr]

Thats not that

It doesnt think

Like

Its trained on riddles

It pasted an answer to the famous one

It didnt think about the riddle at all

[proud bobcat]

Nenávidím Claude Opus, dúfam, že ZOMRIE.

[surreal zephyr]

It just parroted the original

[proud bobcat]

True

It’s probs just in the data

Why think if you already know lmao

[surreal zephyr]

The original is in the data

When given a variant that isnt in the data, opus FAILS MISERABLY

😭

[proud bobcat]

Exactly

[surreal zephyr]

And gemini doesnt fail it

Gemini gets it right

[proud bobcat]

I think Gemini is great in everything except like coding

You cannot tell me it’s better than GLM 5

[surreal zephyr]

Gpt 5.2 gets both, but with a hint because it doesnt know the original riddle

[surreal zephyr]

Gemini is not a coding model tbh

Its not fine tuned for coding

[proud bobcat]

It always has the right idea

But just goes about it so oddly

[surreal zephyr]

I feel like gemini and codex 5.3 working together

Would be really good

[remote talon]

Is there a way to use arena in vscode?

[compact jay]

Hi @echo aurora I saw a big problem in lmarena, and i'm not the only one. it is explained here : #1469113668242509887 message

In short, the hard limit of answer time (e.g 6 min) of arena prevents the model from answering, because some model must think a lot before they answer (like for example claude 4.6 opus) and they are just stopped before they start output, and this interface displays this error : Something went wrong with this response, please try again.. Can you please tell the devs, as you are nearer from them than me, that this problem can be easily fixed, and it could resolve much problems.

[proud bobcat]

I have yet to try codex 5.3

😔

[compact jay]

and sorry to ping you that way..

[echo aurora]

Hello - yes this is something that we are aware of:

In short, the hard limit of answer time (e.g 6 min) of lmarena prevents the model from answering
Would note, this was recently changed and is now 10 minutes. Unfortunately, in order to extend this past 10 minutes, it'll require a large overhaul of our system to support. We may be able to offer this support in the future.

[echo aurora]

No need to apologize

[fervent peak]

Hey listen pineaplee why you delete bug channel ?

[scarlet spire]

@echo aurora Unapologetically, I shall not apologize.

[scarlet spire]

It was closed, not deleted.

[fervent peak]

Oh

[scarlet spire]

It's here

[fervent peak]

Then why you closed

I want that link

[scarlet spire]

Duplicate

[fervent peak]

Thankyou

[scarlet spire]

Duplicate posts make accurate, up-to-date feedback tracking harder than it really needs to be

[echo aurora]

It's explain in that message I wrote out - would like to keep bugs without duplicates.

[scarlet spire]

explained, you prickly tropical

[echo aurora]

#1417174113092374689 is the main thread for both the Infinite Gen and Something went wrong error issue.

[scarlet spire]

Pardon me

[surreal zephyr]

It's the simplest out of all.
You ask it, it does exacly what you asked

[compact jay]

then is it a must to enable the streaming? For long reasoning models, this can be disabled, in order to avoid server surcharge, like you said 🙂

[surreal zephyr]

Which is lovely for me

[scarlet spire]

Speaking of, I actually have some suggestions or ideas for possibly very useful, cheap improvements that can be made to the 2 fora.

[fervent peak]

Btw I have a question why you do not have any role

?

[scarlet spire]

I, my dear ranger, am a free-flyer.

[surreal zephyr]

Anyone tried windsurf, cursor, ect? Hows the quota?

[echo aurora]

Can you make them in #1417174113092374689 ? I know there are a few specific bugs/issues you pointed out there, I still need to get to those, just busy atm

[compact jay]

so bad

[surreal zephyr]

I wonder where would i get most usage for 20-40$ , mainly 5.3 codex and sonnet 4.6 or gemini 3.1

[scarlet spire]

free flyer = an advertisement slip at no cost
free-flyer = an individual, unbound

[surreal zephyr]

Freefrier

[scarlet spire]

Why couldn't I make them in the dedicated improvements/suggestions thread I made? Have you... skipped my thread? 🥺

[compact jay]

hummm actually free tier of kilo code give you 5$ free credits

and offer free models

like glm 5

but

i dont know anything

[surreal zephyr]

5$ you mean 1 prompt?

[compact jay]

when i really need help with code i just ask claude in arena

[scarlet spire]

Likely GPT-5.3-Codex, since it's a highly optimized model that will be cheaper to run.

[compact jay]

... 🥀 they need money ig

[surreal zephyr]

Im using those, but whats the best place to use them?

Codex sub?
Windsurf, cursor?

Github copilot cli?

1500 req per month? Is that any decent?

[compact jay]

do you know google ai studio ?

it gives you access to gemini 3.1 pro

[surreal zephyr]

10 promptsclol

[ebon notch]

bro where i can turn my image into video

?

[compact jay]

a model as strong as opus 4.5

[compact jay]

literrally everywhere

[ebon notch]

In here ?

[compact jay]

even in capcut you can do this

[scarlet spire]

If you're not really attached to a specific model, don't have high usage and want the freedom of PAYG, I'd recommend OpenRouter or similar aggregate API services.

[scarlet spire]

On https://arena.ai/video

[surreal zephyr]

U realise how much api costs?

It charges like 10x normal rates

[ebon notch]

No, i heard that in this discord chanel you can turn image into video

[scarlet spire]

Hence the explicit note: "If you're not really attached to a model, don't have high usage, and want the freedom of PAYG". OpenRouter charges at openAI's own pricing.

[scarlet spire]

That has moved to https://arena/ai/video

[compact jay]

took a random image

[compact jay]

reverse engineer smth

[ebon notch]

It won´t open

[scarlet spire]

.ai sorry

https://arena.ai/video

Just arena.ai in video mode. The main website.

[surreal zephyr]

Yeah but 20$ openai sub offers around ~400$ of api usage

Lol

[scarlet spire]

yeah and OpenAI's subscription offers only OpenAI models.

[ebon notch]

Oh so in the website i just upload my image and it well turn it to video right?

[scarlet spire]

yes. It's on the website.

You don't have to trust me, but I do recommend you do: Try it.

[proud bobcat]

Depends on your use case

API is always superior imo for real work

[ebon notch]

Thank you

[scarlet spire]

Now, what if you were to only use 4 dollars a month worth of API usage with your 20 dollar subscription? uh oh

[proud bobcat]

THIS

THIS

[scarlet spire]

Sure, a 20 dollar subscription is cheaper than 400 dollars of API cost.

But were you ever on track of using 400 dollars of API cost? No? then it's a moot point

That's how they catch you by the willy. Don't let em upsell you in FOMO.

Remember that your anticipated usage isn't the same as your aspired achievements. 👀 Until you make a real use case based estimate, every bit of speculation about money saved by getting an expensive subscription is solely "looking for a good deal" without "looking for the cheapest deal"

[opaque current]

is there anyone looking for a dev?

[scarlet spire]

[desert abyss]

@spring portal Note that Video Arena has been removed from the server. More information can be found in this [announcement ](#announcements message)⁠

[modest prism]

Grok 4.20 is the first model I'm really impressed with. Fetching 390 pages in about half a minute is really something.

[surreal zephyr]

1 session with opus uses me like 1.5m tokens maybe

which is solid 30 bucks if not more

and i can get like two maybe three sessions per day for 20

you see - the point of api is companies pay for it

thats why api is overpriced asf

[scarlet spire]

Fantastic! But not relevant to the scope of the comments I made.

[surreal zephyr]

1 prompt can use 5$

of api

[scarlet spire]

Still not relevant. No Anthropic services were mentioned or replied to.

[scarlet spire]

exactement ici!

Your point becomes moot if you need to retroactively change the context in order to make your point.

[spare gull]

Anyone else constantly just dying inside while watching the Claude Opus 4.6-thinking while doing coding in the WebDev environment?

Looking at the Thinking scroll as it "Well actually" and "I need to stop overthinking this," before finally getting to, "alright, time to generate the files!" and then going right back to, "Wait, now I'm thinking about this one thing for the 96th time."

And then finally, excruciatingly slowly starts actually creating and editing the files at like 12 minutes, which is clearly not close to enough to finish before the 13 minute mark where it always cuts off and throws a "Something went wrong with this response" error. Extra points if it makes it all the way through writing the files, only to run out of time as it slowly writes out the summary of what it did.

[echo aurora]

That's fine too.

[echo aurora]

Is the prompt you're providing just very complex? Are you able to share an example prompt where this is happening?

[wanton plover]

Hello, I recently discovered Arena AI and am using the Claude Opus 4.6 agent to design a website. I have a question: are there any limits on tokens, etc., for using this agent? About two days ago, I kept getting the error 'Something went wrong with this response, please try again.' I followed the instructions you provided on Discord, but it didn't work. I'm wondering if there's some kind of token involved.

[scarlet spire]

There's some transient errors that are not yet entirely pinned down on one cause, with a possible contributor being the recent release date causing high demand, so there's bound to be some occasional generic errors. On token count I can only speculate without good backing. If this is a continual issue with a specific model, specific project or other consistently-same factor, that's good troubleshooting material to bring over here into one of the bug report threads. 🙂

[spare gull]

Oh, it's not any prompt in particular and I'm sure some of them are "too complex" - it's just that Claude Opus spends SO MUCH TIME thinking before writing a single actual file and since the prompt timer was added, it's very obvious that it has a strict cutoff at 13 minutes per prompt, which often means watching page after page of thinking, and sometimes writing a bunch of files, only for everything to go poof as soon as it hits 13 minutes.

It's very difficult to determine what is "too complex" for a single prompt. I've learned the hard way to break things into as many small tasks as possible, since writing "Make me a game like XYZ" is going to fail horribly. But even small bits, like my current prompt was about writing just the visual layout / UI of a single screen based on the classic LCARS interface, can just go on and on and then fail at 13 minutes.

[hidden widget]

Hey, moderator wsp! Good to see you here

I wanna ask too, when is the code leaderboard getting updated

[weak dagger]

Hes the owner by the way.

[hidden widget]

I wish gemini 3.1 pro will climb atleast 2 or 3 steps above

[scarlet spire]

You could possibly give it some explicit instructions to help guide thoughts. Think of things along the lines of any which one of the aspects mentioned in this:

• If you're unsure whether the user wanted a specific variant despite them not explicitly requesting with any specific details, make your best honest guess, with room for iterative refinement as and where applicable.
• Don't overthink things you're unsure of. If the user themselves expressed uncertainty in deciding, pick the most appropriate option based on things they did know for certain.
• Consider the principle of Occam's razor as applying.
• If a specific variant of a generic interpretation was not mentioned, consider it as not desired. The prompt writer is responsible for providing you with the thorough guidance necessary to correctly interpret their request.
• You are a confident agent, not afraid to make honest mistakes in earnest intent of not adding unnecessary resistance with endless doubt spirals.
• If re-consideration of any conclusion in your thinking is required, reflect on whether your iterative doubt/consideration provides sufficient impact to justify. If the actual impact of the matter at hand for this thought is not significant, favour managed uncertainty rather than perfectionism.

Just made these up now though, so tweak as you wish.

[scarlet spire]

Of the pizza joint?

[weak dagger]

Whos the real owner from this server? i know pineapple might is the co owner but whos the main owner.

[scarlet spire]

Servers as large as these (automatically?) default to a structure where there's no "main owner".

Large communities would suffer under centralized, distilled hierarchy that comes down to one person's final say or word.

[spare gull]

Thanks for the tips!

It's just frustrating to watch prompts get so close to completing only to fail over and over right as they are about to be completed - even if the files have been written and the build has been verified, everything is erased when you get the red "Something went wrong with this response," whether at 13 minutes or just a random error.

The funny part is that it wastes way more resources, compute, and time when trying to "fix" a prompt over and over, so that it takes less time and comes in under the 13 minute cap, compared to if it just had a few extra minutes to complete the first time.

[scarlet spire]

I'm afraid this is unfortunately not a thing unique to Claude on Arena.AI, but rather, intrinsic to Claude's RLHF 😕

---

The funny part is that it wastes way more resources, compute, and time when trying to "fix" a prompt over and over, so that it takes less time and comes in under the 13 minute cap, compared to if it just had a few extra minutes to complete the first time.

That is a good point. Could turn that into:

• Consider whether the energy required to determine specifics at once is significantly less than the energy required to iteratively arrive at the same outcome.

[spare gull]

Oh yeah, I know - wasn't saying it was Arena responsible.

[scarlet spire]

I do hope you get a chance to try some prompt modifiers out and couple back to us here what you found to work.

[magic stag]

https://www.anthropic.com/news/detecting-and-preventing-distillation-attacks

LOLLLLLLLLLLL

[scarlet spire]

This is a long-time thing, not a recent failure 😅

It's along the same sorts of difficulty as "pretend we're roleplaying" because it requires some very sophisticated, pre-planned trajectory shaping for your FT stages.

The context part is always and will likely always remain a question of "ideal response for what context?" and therefore never actually be entirely resistant to these sorts of attacks.

[weak dagger]

Why samsung hates apple:

[spare gull]

The best "solution" I have found so far is just splitting any task into bite-sized pieces. So for instance, if it runs out of time trying to build a single page UI or something similar, you may have to just tell it to only do the left half first.

Of course the trade-off is now you have to use more prompts, which means you run into Arena's rate limits, lol. So there's no magic solution.

[scarlet spire]

You run into rate limits and runtime limits for the same reasons: to keep usage fair for everyone

So you'll run into them one way or another unless your requests/needs are toned down to be less demanding of the infrastructure (i.e. less expensive or less capacity consumed)

[urban kestrel]

Hello !

[spare gull]

Oh yeah, I'm not complaining and I'm well aware that rate limits are inevitable, especially on the most demanding models and especially especially with a free service, lol. Just pointing out that there is no one perfect solution in the balancing act between prompts that are too complex and run so long they hit the 13 minute limit and simple prompts that take less time but require significantly more prompts to accomplish the same thing, thus hitting rate limits.

[weak dagger]

OI IA OI IA

[echo aurora]

Thank you for sharing further. Unfortuantely, for the 13 minute cutoff, this is a limitation we're unable to expand at this time. We would like to eventually, but this does require a large overhaul at the moment. This 13 minute cutoff time is just a limitation the users will have to work with for the time being.

Thank you @scarlet spire for the guiding thoughts. Very helpful.

[ripe plinth]

i have an idea so basiclly ai's have the capabillity to install 3D models? because if they can install packages what if there is a feature to install 3D models because it would be good because claude is good at coding and 3D models they find would be way better then just the bad models they make

[echo aurora]

Hello sorry to say I won't be able to give specific times on when upcoming leaderboard updates are coming. It can depend on the leaderboard and the amount of votes we're getting, but generally you can expect around a week for an update.

[deft spruce]

what the HELL IS THAT????
a3:"Request contains an invalid argument."

[echo aurora]

Yes

[fickle venture]

Dang what

[loud verge]

https://tenor.com/view/bongo-cat-cat-meme-slapping-slap-table-slap-gif-26404301

[deft spruce]

yeea bongo cat

[fickle venture]

I think that GPT-5.2-chat-latest is GPT 5.3

[loud verge]

Lmao

[hidden widget]

Hmm close

[loud verge]

Are you being fr?

I have a question to test it.

[fiery gull]

just a name

is new model

[ripe plinth]

🥀

[fickle venture]

Blud coding earth

[fickle venture]

Maybe it could and they just named it "latest" instead

[ripe plinth]

it was just a 2D adventure game bro💀

[fickle venture]

I am not sure tho

Just a guess

[fiery gull]

bro send claude to build gta 6

[spare gull]

Thanks for the reply! It's fine, I understand the limitations and the necessity of having a cutoff, especially on the top-end, heaviest to run models. I was just commenting on the frustration of watching prompts fail right the last moment, especially in those cases when all of the files had been written and it was just wrapping up with a summary of changes.

Although as I mentioned to HumbleDeer, it's interesting to think about how much extra resource usage it winds up creating just from all the extra prompts people then try as they look to simplify their original request enough to get in under the limit. It's like, 1 prompt that failed at 13 minutes, would've completed in 14, but now someone does half a dozen or more attempts over the next day or two trying to get the prompt to finish successfully.

[fickle venture]

Is this against the rules?

[ripe plinth]

🥀

I AM INSTALLING THE WHOLE INTERNET🥀

[shut steppe]

The AI is not responding to me

[ripe plinth]

literally tho look:

its to much😭

[shut steppe]

This been generating since 30 minutes

[spare gull]

Prompt: BUILD ME AN NPM PACKAGE INSTALLATION SIMULATOR

[ripe plinth]

fr tho

[shut steppe]

Can someone help?

[ripe plinth]

bro really said make a windows 11 simulation with all the apps in mincrosoft

[shut steppe]

I’m just building a full stack online casino

[spare gull]

If it just keeps generating, you basically just have to start a new chat - it's stuck in an infinite loop

[ripe plinth]

i just realised that was flash🥀

ohh yea did you reload the site while it was generating?

[fickle venture]

That's well known bug just refresh

[ripe plinth]

thats the reason why

[shut steppe]

I want to save my progress. It’s been like four hours. I was working on this project.

[ripe plinth]

no you reloaded it when it was generating thats the reason

it happened to me alot of times

[shut steppe]

Ohhh

[ripe plinth]

its never gonna asnwer bro

answer*

[shut steppe]

But I need to save my progress

[ripe plinth]

eh are you in a account?

[fickle venture]

@shut steppe ☝️

[spare gull]

The best you can usually do is click on one of the previous prompts Model title bar, which should load that code and let you use the download button - otherwise, it's just lost

[ripe plinth]

like logged in

if you are just yea download it

[shut steppe]

I did the same thing happened

[shut steppe]

Ok

Thanks all of you guys

[fickle venture]

Keep refreshing it usually means model overload or model didn't respond

[ripe plinth]

but then

you might want to go onto the real google gemini site

because

this lmarena doesnt allow files

[shut steppe]

😭

[ripe plinth]

when your in coding

so yea good luck broski🥀

[shut steppe]

Thanks

Learned my lesson

[ripe plinth]

nah what the 💀

npm packages are not this serious its installed like more then 150+💀

[spare gull]

Maybe the more packages you have, the more amazing the final product will be.

"OK NOW I'm writing the code. For real this time. No more deliberation." It LITERALLY just thought that.

I feel like Claude is just making fun of me now.

(It did NOT, needless to say, start actually writing the code)

[fiery abyss]

hi!

[novel crater]

.

[long minnow]

npm packages log is giving me comfy ui local install flashbacks

[jolly lark]

Since 4 hours

[regal grove]

I have a question: is it possible to cancel a conversation with AI? For example, I sent something, but there are things I didn't add. Instead of waiting for the AI to finish what it needs to do, I could cancel and send the correct prompt. Is it possible to do that?

[echo aurora]

Sorry to say there isn't a Stop Button, but it is something we want to bring to the platform eventually.

[regal grove]

Okay, thank you.

[long minnow]

Soon™
In cinemas next century

[compact flame]

@echo aurora

[echo sinew]

Got you covered. You can also tag the Mod team with @ Moderator

[compact flame]

Oh alright

[echo sinew]

And thank you for reporting

[golden ocean]

https://tenor.com/view/komaru-getting-ddos-gif-1428811091514520825

[harsh flume]

Does anyone here uses openrouter?

I am doing a pipeline with multiple model API calls and am starting to feel that some features might be nerfed when used through openrouter

[gritty arrow]

will the ai coder still code if i shut off my device?

[scarlet spire]

in CodeArena? Yes

[scarlet spire]

I've used it a fair bit

[gritty arrow]

on the main site?

[scarlet spire]

OpenRouter supports pretty much all the necessary features of all the major API's and has consistently updated their API with new things or retrograde fixes when something's buggered by the third party. They provide good cross compatibility and even offer an Anthropic-style API interface rather than the OpenAI-style API.

Any particular issues you're experiencing that have you suspecting you're running into limitations with OpenRouter that you wouldn't with the native party's API?

[surreal creek]

mammoth-newt-0206 literally never generates responses why is it in the Arena

[echo aurora]

Are you getting an error, or is it just stuck?

Do you have an Eval ID of a session this happened in?

[simple kernel]

Alr bro, this is starting to annoy me

[surreal creek]

after about 10 seconds it just displays the other model’s message and allows me to vote, while mammoth just still says “Generating..”

I normally archive chats right away after I get that error, if I encounter it again I’ll jot it down

how can I retrieve the Eval ID of a session?

hah! just encountered the error again but with GLM-4.7 (the other model of the 2 I’ve been encountering this error with, but not at a 100% rate)

[surreal creek]

still can share though, is it just the battle URL link?

[surreal creek]

okay, just received mammoth-newt again with the same error

https://arena.ai/c/019c8beb-b3fe-752c-9a89-a2b4dcc1b854

if this isn’t the Eval ID lmk, it was a multi-chat session, got the error on the last one

[fossil socket]

@echo aurora when will sonnet 4.6 get the PDF attachment feature?

Just realized they literally do not have it still

[scarlet spire]

I do have a session ID and will collect the eval IDs for you as well.
But I'm not the person you replied to.

Edit: I'mma send you excerpts of the raw json payload, actually—'cause there's some interesting clues in the metadata that have me with an "Aha!". In DM?

[fierce kelp]

https://discord.com/channels/1340554757349179412/1475173552050143323

[surreal creek]

ah, someone else found it

[fierce kelp]

So multiple people are having the issue.

[scarlet spire]

Yes. And I'm currently even tracking a pattern in the actual problem at hand

like a real big hakker

[fierce kelp]

Let us know what you find out

[queen veldt]

Ayyy we finally got the error logging!!

[shrewd citrus]

wdym

[queen veldt]

Image gens seem to have explanation why they failed instead of the generic failed to generate

Also there's a wierd visual bug on phones for the buttons

I cant reach send since this is non-scrollable

And when i rotate my phone I can't even see it properly

[ivory condor]

Create a cinematic emotional AI video story.

Storyline:

Scene 1:
Turab is poor, wearing simple clothes, sitting alone on the street.
Sad atmosphere, dark colors, rain, emotional background.
Turab looks tired, hopeless.

Scene 2:
Sabtain is rich, wearing luxury clothes, expensive car, big house.
Sabtain laughs at Turab mockingly.
Camera shows contrast between poverty and wealth.

Scene 3:
Transformation scene:
Turab working hard, studying, learning skills, working day and night.
Motivational music, sunrise scenes, progress montage.

Scene 4:
Success scene:
Turab becomes rich and successful.
Luxury car, modern house, confident look.
Bright colors, cinematic light.

Scene 5:
Role reversal:
Turab sees Sabtain now poor.
Turab smiles and laughs back.
Moral scene: life changes, time changes, roles change.

Atmosphere:
Emotional, inspirational, motivational, powerful.

Style:
Ultra realistic, cinematic, 4K, smooth transitions, slow motion shots.

Text on screen:
"Never laugh at someone's struggle"
"Time changes everything"
"Hard work changes destiny"
"From poor to powerful – Turab"

Music vibe:
Emotional start → motivational middle → powerful ending

Duration:
30–45 seconds

[fickle venture]

Video arena is gone, go to arena.ai to use and delete this please

Or #share-prompts

[river garnet]

hola amigos como genero los videos

alguin sabe como generar videos..?

[fickle venture]

This is English only server

@river garnet @ivory condor
https://arena.ai/video

To generate video

[river garnet]

Does anyone know how to generate videos?

[river garnet]

ok tank

[sharp mirage]

Yo

@echo aurora

Can you help me guys with something i got this msg for the first time

Prompt is too long 212707 tokens > 200000 maximum

[echo aurora]

Yes, you can go here to find Video Arena -> https://arena.ai/video note that we've removed the bot that use to do Video Arena in the Discord server.

[sharp mirage]

What should i do ?

[echo aurora]

Seems like you need to lessen the prompt.

[sharp mirage]

I cant use chat anymore ?

Or just i have to wait ?

...
.....

......

[sharp mirage]

.

Do i have to start a new chat or I can just wait ?

[echo aurora]

Sorry I think I'm misunderstanding the issue, did you get an error message?

[sharp mirage]

Yea its Prompt is too long 212707 tokens > 200000 maximum

The error mag "Prompt is too long 212707 tokens > 200000 maximum"

[sharp mirage]

If you're not free, tell me later.

[harsh flume]

Just plain comparison between deep research results esp from Google and Claude

[echo aurora]

Seems like that prompt is just too long and needs to be trimmed down. You can still use the chat, just that prompt is going to be too large.

[sharp mirage]

I sent it's working

Maybe cuz the chat was too long ?

[fossil socket]

@echo aurora uh what

[echo aurora]

That's what it sounds like, but if it worked, maybe not?

Can you send me a screenshot of that error? I'm not familiar with that one.

[echo aurora]

That's odd. Did the prompt have a PDF uploaded?

[fossil socket]

Uh yeah 5 PDFs

But that was done before that message was sent

[echo aurora]

Here can you make a new post in #1343291835845578853 and tag me there? Walk me through the steps you took to get here.

[echo aurora]

We recently added the ability for new/more error messages to appear.

[echo aurora]

Sorry I won't be able to get you an answer on this at this moment. There are other questions/bugs I need to focus on atm.

[sharp mirage]

Alr tell me when free

Do i have to send in bugs?

[fossil socket]

Also uhm, when will sonnet 4.6 get the PDF attachment in

[echo aurora]

Can you provide the Eval ID for this session?

[echo aurora]

I'm not sure, but will flag to the team.

[sharp mirage]

How can I do that ?

[echo aurora]

No need, I'm chatting with the team about this rn. It's a little unclear to me what is causing this.

[echo aurora]

Eval ID is the random set of numbers/letters in the URL when on a specific chat session

[sharp mirage]

Oh you want it ?

[echo aurora]

Yes please

should look something like - 019c8c0d-67bb-7661-aad8-5f47aa3493e5

[sharp mirage]

It might be cuz i use all the api tokens ?

[fickle spruce]

@echo aurora Can't wait for stop button. Looking forward to it

[echo aurora]

Big same! It's in the works.

[sharp mirage]

I dm you

Did you get it ?

[echo aurora]

Yup

[tacit stream]

That would be excellent

[steady berry]

Guys, where is the video generator here in discord from arena?

I remember i generated a long time ago one video in this discord with a bot

[mortal vale]

@steady berry Hi! Note that Video Arena has been removed from the server. More information can be found in this #announcements

[lavish oyster]

hiii

[fathom basin]

I realised that arena is not very good when it comes to human faces. Do you guys experience the same thing?

[echo aurora]

What model did you use? Have you tried others with different prompts?

[fathom basin]

is there a way to change the model? I tried but can't

[echo aurora]

For Text, Search, Image you can by selecting Direct and Side by Side. For Video you can only use Battle (can't select specific models)

[queen veldt]

Yaay finally a error logging tysm

We will finally solve the mystery of unknown errors

[fathom basin]

ya, so far i'm only using Video, no wonder I can't find an option to change models

[echo aurora]

Wahoo!

It still has more to be added to it, but this should be helpful in the meantime.

[queen veldt]

Although people will spam their errors here so there's a problem

[echo aurora]

Well at least we have a bit more to go off of now.

For example if someone has their date/time set to something that's not current, before it was the standard Something went wrong now it'll point to the issue a bit more clearly.

[queen veldt]

Lol why would they have unsynchronized time, they can't access google like that

Yeah in the end most of the problems will be fixed and better user experience

[toxic verge]

It’s all so intertwined

[echo aurora]

It's more common than you'd think

[queen veldt]

https://tenor.com/view/uhh-monkey-meme-monkey-meme-uhh-i-don't-know-gif-16277978565534595779

[toxic verge]

I’m in a creatively burn out 🙁

[toxic verge]

Yeah, I get the same issue only one out of the two videos ever generates

[queen veldt]

I've had this error so far, I can't press regenerate it keeps showing me this (or maybe i ran out of video i did 3 video battle mode)

[toxic verge]

How many videos do you get a day? I thought it was one.

[queen veldt]

Idk i did 3 battles?

[toxic verge]

I’ll go try

[queen veldt]

Got 5 videos and 1 failed

[toxic verge]

3x2?-1

Let’s see what I get

[queen veldt]

I guess more premium models which have longer wait times for gen will time out for the response since they take longer to make?? Guess that happened to me

[toxic verge]

Wow lol

[queen veldt]

Well you prompted something bad

Well it shouldn't obviously

Yeah don't post it here

[toxic verge]

The filters are odd

They’re not the same when you use the real model for some reason

Same with nano

[obtuse smelt]

hmm

[queen veldt]

I feel like the arena has more powerful versions of models than what we get i dunno might be wrong

[cloud sierra]

Imagen to video

[echo aurora]

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

[toxic verge]

Benchmarks are supposed to measure a model's capability.

Model A is more powerful but has a trigger happy safety filter, it might return a Something went wrong" error.

Model B is weaker but has a more relaxed filter, it successfully generates an image.

People vote for Model B because it actually produced something, making Model B look "better" in the rankings, even if Model A’s internal generation was technically superior.

Maybe?

[echo aurora]

This wouldn't effect the leaderboads. Before we publish updates, we scrub the data for any issues like this.

[toxic verge]

Oh ok

[echo aurora]

If we see a vote is for one model, but the other errors out, that vote won't count.

[toxic verge]

Oh, that’s awesome

[echo aurora]

Same of other edge cases. Say if a model in the response says "hey I'm X". Even if both responses generate, those won't be counted.

[toxic verge]

Right.

How safety moderation influences user perception of an Llm utility, comparing a user's likelihood to vote or prefer a model against its objective technical performance

If anybody’s interested in conducting research on this topic, hit my DMS

I want to write a paper on it.

[hollow imp]

Y'all don't have 10$?

[scarlet spire]

the "prompt" in this case is "chat-history-tokens"+"configured-max-model-output-tokens"+"new-prompt-tokens"; if chat history + the other two is exceeding the limit you're

1. screwed because well, you'll need to trim anyway
2. forced to reduce context size by either reducing your demand for this last message or
3. strategically trimming context

I don't believe Arena supports the latter/trimming option, and frankly, if you're hitting 200K tokens you've probably had enough free access to paid LLMs for today

[celest condor]

any site to use seedance 2.0? doubao not working now

[scarlet spire]

Nothing found on ByteDance's generation platforms, like jianying?

[half mist]

@echo aurora "We’re planning to remove the Video Arena generation channels from the server on Monday 2/23 @ 4pm PST. If you’d like to download any generations, please make sure to do so before that date." So, are you gonna do this?

[scarlet spire]

Assume this is a yes

For relying on it not being true would be foolish for your own good

[molten robin]

for some reason ONLY arena.ai is not loading.

now it is for some reason

[scarlet spire]

It happens not every ISP is connected to every backbone network connection

[prisma cipher]

Error fatal? @echo aurora

[toxic verge]

Holy smokes

[half mist]

@echo aurora Where is the quality control? First, it was the Code Arena URLs being flagged as phishing, and now all Grok models fail.

[robust sluice]

ayo whats this bro ?

[toxic verge]

I saw that too

[plain pendant]

is there a way i could use claude ai free model with an api key?

[shy dust]

https://cdn.discordapp.com/attachments/1241946859170172969/1475317101458886803/snapchat.gif

[royal sail]

The orange is very fitting lol

also w tron reference

[shy dust]

oh my god

your pfp is also tron

[royal sail]

indeed

[shy dust]

Ok. We are friends now.

https://tenor.com/view/tron-tron-legacy-meme-me-after-stealing-your-balls-rinzler-gif-10107317769141218634

[royal sail]

i gotta steal that one

lmaoo

[shy dust]

https://tenor.com/view/tron-legacy-clu-ive-been-waiting-a-long-time-for-this-i-have-waited-a-long-time-for-this-moment-disney-gif-25198919

[lunar glade]

@echo aurora Can you please explain to me why the heck did I keep getting this thing?! is my account being flagged in your black list?
for what? for being I trusted you and uses your services?

It keep telling me to select image non-stop, when I say non-stop I mean literally, it keep wanting me to tick images after images after images after images after images after images after images NON-STOP!

This Securities Verification is clearly buggy, it stucked in loop and my work can't be complete and fail due to timeout

Now my account is literally useless due to this nonsense! even I spaced out my use time a 24hours later!

[arctic scaffold]

Guys, for some reason arena ai stopped working in google chrome - ERR_CONNECTION_RESET

Any ideas?

[plain gull]

use it on brave

[arctic scaffold]

firefox works, edge too. No idea what happened

[plain gull]

brave is a better alternative

btw do you know how to embed voice agents on a website

[undone saffron]

He has nothing to do with those captchas
It's normal for them to pop up at random times
Me too, but only a few
Try following these steps and see if they stop appearing

-# After following those steps, I didn’t get any more captchas

[quartz pike]

wait what

REVE FINALLY HAD A GLOW UP

I NEVER EXPECTED THIS

[echo aurora]

[echo aurora]

It's a new experiment! https://help.arena.ai/articles/3847739843-arena-experiments-fast-mode cc @toxic verge

[fickle venture]

Who the heck is reve dawg

[quartz pike]

the image model

that sucked

ASS

[echo aurora]

Yeah we made some changes where the error message is going to display more information. However, there is clearly some refinement that needs to take place.

[fickle venture]

I mean it's top 4

[echo aurora]

https://app.reve.com/

I rly like their site.

[echo aurora]

Can't say I'm familiar with that error, where are you seeing it?

[echo aurora]

Yeah thanks wanted to give people a little bit longer incase they missed the cutoff.

[fickle venture]

Dang it's really beautiful and we'll designed

Dude what this model need to be under top 5 😭

[echo aurora]

Code Arena URLs being flagged as phishing
Can you elaborate on this?

and now all Grok models fail.
The error rate is higher than usual, this has been flagged. Would note they aren't fully down, generations are happening.

[echo aurora]

right?!

[errant cave]

This is nonsense

[quartz pike]

IS THAT REVE 1.5

[errant cave]

I've been using DeepSeek since R1 and it has always been a far bigger narc than the GPT models

[undone saffron]

At least they made the error show up as it should, and not just “Something went wrong with this response, please try again.”
I seem to recall that I proposed that idea some time ago, but I never put it in #1372230675914031105

[fickle venture]

Yeah it sucks

[fickle venture]

To be honest arena website need some improvement

[errant cave]

This looks pretty decent

[quartz pike]

No ts is pretty good

[errant cave]

Try specifying a style

[quartz pike]

can you send the full image

[echo aurora]

We've got a lot of improvements we're going to make, what in particular are you referring to?

[undone saffron]

Saying it is easy, hoping that someday they will do it is difficult

[toxic verge]

Rev is nice

Very professional

[echo aurora]

The captcha system overall is going to tigger whenever it detect use it believes to be inauthentic. Repeated prompts, rapid use, multiple prompts running simultaneously, etc. are the types of behaviors the site is going to apply these measures. Taking a break or slowing down the prompting overtime should help. This system is being looked at actively for some changes to make this detection system much better.

[fickle venture]

I take back my word, rn arena ui is great than before

[toxic verge]

Do you guys detect bot activity on your end? I’ve seen a screenshot or two of some in the wild. lol

[echo aurora]

That's good to hear! But don't hesitate to let us know if you'd like to see changes.

[echo aurora]

Yeah

[toxic verge]

Is it a lot?

Yeah people are clever it’s a cat mouse game Forsure

[toxic verge]

For things that people bot usually from what you guys see is it text based or media based that people bot the most?

If I had to put my money on, it, say it’s code

[echo aurora]

I wouldn't be able to go into details, and I should be more clear that we monitor and take steps to prevent bot activity.

[toxic verge]

Haha how did I know you were gonna say that?

[echo aurora]

Even though there aren't things I'm able to go into specifics about, it's always appreciated for community members to still ask. It's a good problem to be asked questions about the inner workings, means people care.

[toxic verge]

I hear what you’re saying. It just human nature because for every person that does care, there’s probably two that don’t. But you are correct it is critically important to be vocal.

[fickle venture]

Holden I just tryed reve 1 I think not 1.5 and it actually made this great very good at manga coloring

[toxic verge]

Dang it’s good

[quartz pike]

yeah no idk what they fed reve but holy sh#t that is so good compared to reve 1

[toxic verge]

Example?

[fickle venture]

Did bro gave him mustache 😭

[fickle venture]

No that's reve 1 I used this because reve 1.5 doesn't support image edit

[quartz pike]

oh ok

mb

[thorny schooner]

I hope you'll implement the verification Improvement changes soon, since it seems inconsistent at times, to be honest. At times, it gives me a decent amount of headspace before it starts to go into the verification loop, but other times it will not even receive 2 before it goes into that annoying loop 🥲

[fickle venture]

I just found one idk if this is a bug or it just work like that, can y'all make it so the models be like the leaderboard not just random models so people can easily shoose between models without going to leaderboard to check.

Hopefully I explained it great.

[toxic verge]

Yeah, that feature is annoying

[fickle venture]

Wait Holden that's much better

[bleak lake]

ye it is

[toxic verge]

I just made three videos on accident thinking I was making an image

But surprisingly, the video came out pretty good

[fickle venture]

Lmao what model is that

[toxic verge]

Idk no way to vote

There’s no ability to vote

This one is a really good model

[bleak lake]

This one is good as well

[toxic verge]

Do you guys like that fat line work?

[fickle venture]

https://discord.com/channels/1340554757349179412/1475731095071752354

[fickle venture]

That's kinda bad the hair need to be pink and that's way too much light tbh

[toxic verge]

Light is so hard to control probably one of the hardest things

Damn I remember when they first had the video arena you could do up to 20 videos 😂

[fickle venture]

Guys upvote his we need this feature:
https://discord.com/channels/1340554757349179412/1475731095071752354

[fickle venture]

@echo aurora we need this one 🙏

[toxic verge]

Ever since they got rid of the video arena it’s been really quiet in here

😭

[scarlet spire]

Oh really? There's plenty of shitposting on a regular basis, in my opinion

[toxic verge]

Ya I’m probably guilty of that

[echo aurora]

I wouldn't call this a bug, but we do currently list the model drop down by rank in the leaderboards. The reason you'll see some discrepancies are because some models are labeled preliminary, and some models were just added to Arena and haven't yet been ranked.

[thorny schooner]

kind of annoying how the second the verification comes up a lot of times it will just get stuck into its non-accepting Loop of not showing properly being glitchy and not giving me the proper pictures to select

[toxic verge]

There is a disparity between what’s on the leaderboard and between user sentiment

[fickle venture]

Well then it would be great to have a sort feature to sort things based on recent models, best models, low models and etc. just like I provided on this feedback: https://discord.com/channels/1340554757349179412/1475731095071752354

[fickle venture]

Well well it already got nerfed just as I guessed

[toxic verge]

It’s been Nerf. I just don’t understand how it’s ranked so high

The leaderboard is a poor reflection in this case of how people really feel about the model

[fickle venture]

Well to be honest I think it only got nerfed on Gemini app and aistudio.google.com but if you try on the API you will have the unnerfed version.

This is just my guess

[toxic verge]

What could explain this?

[hollow imp]

???

Bro

Do you know how many people and students have gemini pro for free?

[toxic verge]

Yes, I’m aware

[hollow imp]

Vertex ai is nerfed as well

Because of the 300$ free credit

Google cloud

[toxic verge]

Yes

[fickle venture]

I told Gpt 1.5 image to remake arena selecting models with sort feature and hold on it really cooked.

[toxic verge]

I guess I’m just beating around the bush

[hollow imp]

That's not it

You cannot get gemini 3.1 unnerfed without proper api configuration

Just use some other model why u need gemini

@toxic verge

[fickle venture]

Dang

[hollow imp]

@toxic verge

Why do you specifically need deepthink

[toxic verge]

I don’t hardly use any of these models

I just do image generation for the most part

[shrewd citrus]

Is reve v1.5 on direct chat?

Cause usually when a model is announced on the leaderboard you could usually access it

[toxic verge]

.

[burnt sinew]

wow i guess i drained arena's gemini api budget huh

[toxic verge]

Campbell’s Law warns that as soon as a leaderboard becomes the primary social indicator of a perceived value , the very process of AI development shifts from making models smarter to making models rank higher

[shrewd citrus]

Isn’t that a good thing tho

especially with leaderboards such as lmarena

[toxic verge]

https://arxiv.org/html/2504.20879v2

[obtuse smelt]

gemini 3 pro is back ?

[shrewd citrus]

What is chatbot arena

[burnt sinew]

3.1 pro

[toxic verge]

https://github.com/Farama-Foundation/chatarena?tab=readme-ov-file

I don’t know what the referring to this paper but lm arena is also mentioned

This would explain at least in my opinion why what users experience on the ground and in the wild is far different than what’s on the rankings

Essentially, it’s a popularity contest lol (no offense)

I just wish there was a way to close the gap between both worlds so we can get a better picture of these models

[weak dagger]

Seedance 2.0s api is comeing today!

[toxic verge]

Can’t wait. How do you know?

[obtuse smelt]

in battle mode ?

[burnt sinew]

direct

[whole sundial]

lmarena/arena was once called chatbot arena until early 2025

used to be LMSYS chatbot arena before that, but they have had nothing to do with LMSYS since early 2025 (LMSYS primarily makes things such as sglang, which is used to host llms)

<@&1349916362595635286>

remove this too <@&1349916362595635286>

[undone saffron]

Me when I play pool and see that after 5 rounds no more balls are added:

[whole sundial]

also btw i am glad that arena decided to show people actual error messages instead of "something went wrong" which explains nothing
back on the old gradio chatbot arena it always shown actual error messages (mainly rate limit stuff since arena was still a non-profit college-affiliated entity and not a for-profit business like it is now, they lived on "grants" given to them by major labs aka free api access, when it got popular due to it having free ai those hourly limits appeared often, you were lucky if you were able to prompt a model such as gpt-4o)

[weak dagger]

IM NEVER LETTING PINGUINS IN MY HOUSEE

[toxic verge]

Is that seed?

[outer spear]

Reve 1.5 should not be rated that high wtf

[undone saffron]

Who knows

[scarlet spire]

Chatbot Arena is one of former names of the company.

From back when the distinctions were made between Chatbot Arena, Coda Arena, etcetera. The naming does seem to stick for some. But names have changed like twice by now? In light of the multivalence

[urban badger]

hi chat

i MIGHT be stuck in the captcha underworld again

[scarlet spire]

uh oh!

what's your botScore looking like?

[subtle rivet]

ohh

i dont know

i not creat vedio

i face some problem

[toxic verge]

What problem?

[scarlet spire]

I'm probably heading off to bed thoon

[toxic verge]

Same here it’s a slow night tonight

[scarlet spire]

it's 9:20 AM here 😂

[toxic verge]

1230 am

[scarlet spire]

Ten minutes is not quite a timezone offset I know of 🤨

404 - Page not found

    .|||||||||.
   |||||||||||||  where am i?
  |||||||||||' .\
  `||||||||||_,__o

hehe. adorable. PostHog.

[spark python]

Nano banana is giving me new errors

gemini-3-pro-image-previe..

G

Error during image generation with google-genai for model endpoint gemini-3-pro-image-preview: Failed to fetch image: 429 Too Many Requests - [{ "error": { "code": 429, "message": "Resource exhausted. Please try again later. Please refer to https://cloud.google.com/vertex-ai/generative-ai/docs/error- code-429 for more details.". "status":

"RESOURCE_EXHAUSTED

[scarlet spire]

Pretty much all of Google's Vertex AI services are at the moment, from what I can see

[spark python]

Noo😭

Right when I needed it

[scarlet spire]

It's a computer making up visual information. You don't need it. 👀

[surreal zephyr]

Atp im highly certain opus is the dumbest model

Opus has literally the riddles hardcoded so any variation makes it spew bs

[hazy forge]

yup

[lyric marsh]

Hi. Can someone explain what this means?

[obtuse smelt]

i think is machine error

[queen veldt]

It's the Chinese

[compact flame]

I think arena ai support is going crazy

Well the user support ai one

Bro it keeps spamming glad I was able to help

Seriously I think he's too glad to help

It's just keeps going

[toxic verge]

https://youtu.be/e4Lx5Bmpojw?si=m4sA-f5CtADdcOV3

[bleak lake]

💀

[undone saffron]

Always

[toxic verge]

It’s hard to maintain that style

Because the AI always dies bold fat line work

[toxic verge]

I got bad examples lol

[junior flax]

I hope they add option to choose and compare ai video generators not only anonymously

[compact flame]

That would be too expensive

[junior flax]

ik,but it has it anonymously

i know it chooses it randomly

[swift hatch]

hi

[violet trout]

Hello, can someone help me? Is there a solution to this problem, or will it resolve itself?

[zinc flame]

Im having the same issue 🙁 anyone?

[vast yew]

You haven't done anything wrong and there's nothing you can do about it.

This is an issue between Arena and Google.

[violet trout]

Okay, thanks for the explanation. So I'm going to wait.

[vast yew]

I assume they'll have it resolved shortly if Google's at all worried about Arena rankings, cos this will definitely have lots of people voting against them

[violet trout]

Okay, I understand, and I hope so too

[prisma cipher]

Is it fixed now?

[toxic verge]

429 Too Many Requests: This is a standard HTTP status code. It means the server is refusing to process the request because too many have been sent in a short period of time.

[fervent trout]

no fix to this its on server side

just wait

[violet trout]

Thank you

[toxic verge]

Is it true if Arena's daily or hourly budget for that specific endpoint is tapped out, it will throw that RESOURCE_EXHAUSTED message until the next reset cycle.?

Probably not ?

[golden ocean]

no way are they finally showing the actual error reason now

[whole depot]

nice fix

helo

[toxic verge]

Gives the model away 🤣

[golden ocean]

REAL

[rough surge]

i hate it when claude opus 4.6 thinks for so long that is gives error. Wasted time...

[golden ocean]

fr

[warm crater]

guys are u able to use arena

cuz its showing failed to terms and condition

[tough kayak]

Is there a way to choose which two video models to test... Not random?

[safe copper]

Hello

[inner relic]

So someone said bytedance models will release today

when

[shrewd citrus]

it won’t

it was meant to release today

but copyright

so it’s delayed

forever

[inner relic]

Yeah buddy, i wonder where you got that information

https://tenor.com/view/it's-a-trap-this-whole-thing-is-a-setup-light-yagami-death-note-gif-6231066098101679168

[shrewd citrus]

https://help.apiyi.com/en/seedance-2-api-delay-seedance-1-5-pro-alternative-en.html

https://tenor.com/view/speed-dad-laughing-speed-dad-crying-speed-speed-ksi-speed-trying-not-to-laugh-gif-1466628274697419264

[fiery gull]

Bruh

Imagine the open dola, it would be a dream

[hearty inlet]

#video-arena-1

[toxic verge]

Yeah Img and video is about to be tanked

They won the ai music scene

[inner relic]

Chatgpt and google gonna make a move against seedance 2.0

[toxic verge]

You’ll be able to purchase your intellectual property to generate

[toxic verge]

Why?

[inner relic]

we all know why.

[toxic verge]

China., ? Haha

The algorithm used by TikTok will be retrained on U.S. data, but ByteDance will maintain ownership of the original algorithm.

What a werid deal

It’s like buying all the customers from McDonald’s and owning the franchise, but not owning the actual recipe

[glass zealot]

arena is broken

[stray aspen]

where do i use seedance 2

[glass zealot]

in china

[stray aspen]

it was supposed to have its official release today

[glass zealot]

they delayed it probably

[stray aspen]

that sucks man

[hushed gyro]

???

[obtuse smelt]

is all model broken

[junior dawn]

One message removed from a suspended account.

[keen beacon]

Wtf

37 minutes Claude opus

ARGH

[mortal vale]

@rugged perch Note that Video Arena has been removed from the server. More information can be found in this #announcements

[fickle venture]

Dude

[jovial bison]

Error during image generation with google-genai for model endpoint gemini-3-pro-image-preview: Failed to fetch image: 429 Too Many Requests - [{ "error": { "code": 429, "message": "Resource exhausted. Please try again later. Please refer to https://cloud.google.com/vertex-ai/generative-ai/docs/error-code-429 for more details.", "status": "RESOURCE_EXHAUSTED" } } ]

[fickle venture]

Well the error showed the answer already

[jovial bison]

Well it's the first time people have the problem like that

[fickle venture]

You got the error from arena.ai?

[jovial bison]

Yep

[fickle venture]

Dang y'all overusing it

[jovial bison]

5 pictures per hour ☠️

[weak dagger]

Bytedance why, WHY DELAY SEEDANCE 2.0

[spare rune]

Is generous for a company that doesn’t require you to pay*

[river whale]

this guy came back

[river whale]

🤓 "Is generous for a company that doesn’t require you to pay*"

[spare rune]

If you want the models so much just PAY

Sigh

Anyways

[river whale]

u can only sigh

[spare rune]

Deepseek is a nothing burger

[river whale]

then dont use it

[spare rune]

Is that it.. is that all you had to say

[spare rune]

I don’t..

I’m just talking about the

Runouts about

Deepseek

I think it was

[river whale]

then how are you saying its a nothing burger

[spare rune]

4

[spare rune]

Because it’s a nothing burger

Leaderboards

Stats

[jovial bison]

The point was that 5 pics per hour is not overusing

[spare rune]

Evaluation

[river whale]

taunt needs some taunt to fix itself

[spare rune]

Since when..??

[spare rune]

2nd time echoing my name..

[half mist]

Whenever the url is shown on Arena, it appears white, and when I actually go to the website, I get this

[river whale]

your name is a simple word

[spare rune]

like I guess bro

[jovial bison]

Because 5 is 5????? It's not one hundred.

[river whale]

he is talking like he pays for it

[spare rune]

But your using for paid models

Just use them in the original site

It’ll let you use more than 5

[jovial bison]

Bro it seems like u are not that smart as u think

[spare rune]

No yall are talking like yall pay for it.. getting mad about a rate limit

[river whale]

he is only a taunt

[spare rune]

Is crazy

[river whale]

when did i get mad dumb burger

[jovial bison]

Why are u making things up

[spare rune]

omd..

[spare rune]

tell me

why is 5 rate limit a problem

just tell me

[river whale]

cause he is taunt

[jovial bison]

Where did I tell u that it's a problem?

[spare rune]

okay so it’s not a problem

Exactly

Your correct

[jovial bison]

One dude said that people overusing arena, I said that people only have 5 pictures per hour so it's pretty hard to overuse it and then u started to cry about things that u made up

[spare rune]

it’s not a problem move on

you said it yourself

[river whale]

bruv u need to move on, if u really dont care then dont speak for it

[spare rune]

I’m just saying move on

[river whale]

why do you come agian and agian to get slammed

[spare rune]

I’m not continuing..

[spare rune]

slammed and it’s just
Nothing..

[river whale]

yeah cause i did a taunt

[spare rune]

or if you meant slamming as in “he is taunt”

Then mb

wait wait stop let me guess your next sentences

“he is taunt ing”

[river whale]

coming next on kivest ai

[spare rune]

Top 5 things never happening

[river whale]

yeah cause u are banned lol

[spare rune]

I can kinda just go to your website and get the api if needed

I don’t remember the url

Tho

[river whale]

I can just ban your gmail