general

[stray aspen]

What

[little ginkgo]

.txt file upload

[echo aurora]

Could happen!

[little ginkgo]

Pls add it today

I will grow a pineapple tree

Trust

[echo aurora]

Oh

That changes everything

[little ginkgo]

Will u add it

[echo aurora]

Maybe...

https://tenor.com/view/cat-calico-cat-cute-dance-cat-dance-gif-27615826

[little ginkgo]

Pls

[stray aspen]

Wassup pineapple

[little ginkgo]

Hi

I am pineapple eater

[burnt sinew]

It could, but it continues off of the worse model if I pick the wrong one in a/b

I assume that it uses the voted response and not necessarily the model you picked to chat with outside of a/b

[limpid magnet]

I have posibiliti to add imagen for a video ?

[echo aurora]

Makes sense, thanks for expanding.

[echo aurora]

You can use image-to-video, you'll want to type our /image-to-video in #video-arena-1 . If you check out #1397655624103493813 you'll see some more information.

[celest orchid]

#addtxtupdload

[burnt sinew]

And arbitrary character limit and show token stats in chat

[strong pulsar]

[burnt sinew]

Wtf?

No way

[gleaming roost]

[burnt sinew]

how do you get side by side?

[strong pulsar]

[echo aurora]

No, not only when a new model is added. We recently did this with Text Arena - #announcements message I don't believe these updates will appear on the Leaderboard Changelog, although it probably should.

[wintry jackal]

Can we upload photos to any of these models ?

[burnt sinew]

yes

[wintry jackal]

What models specifically?

[burnt sinew]

image edit models

[wintry jackal]

Like Gemini GPT etc?

[burnt sinew]

that too

[burnt sinew]

gemini 3 pro is the best

and its free

[hollow ivy]

not in coding
for coding, coast is the best

[burnt sinew]

he was talking about image models

to look at images

[hollow ivy]

k

[burnt sinew]

what is coast?

i have never heard of that

[hollow ivy]

coASt = co45t

[wintry jackal]

Coast? I use Claude Opus, that one is insane

[hollow ivy]

which prog-lang would you use?

[burnt sinew]

i still dont understand

[hollow ivy]

co45t = claude opus 4.5 thinking

[wintry jackal]

i got a website on svelte/sveltekit with JS

no coding experience btw, so dont think im a pro lol

[wintry jackal]

oh lol

that makes sense xD

Btw guys, is there any LMArena extension for VSCode?

[gleaming roost]

I think it has the https://marketplace.visualstudio.com/items?itemName=copilot-arena.copilot-arena
But I'm not sure if it's really from LMArena.

[burnt sinew]

i think next gemini 3 pro update will

[echo aurora]

For the most part, yes.

[wintry jackal]

I tried this one but it gives me an error while registering

Its giving me this error, with any username that i put (Copilot Arena in VSCode)

@gleaming roost am I doing something wrong here?

[proven vessel]

👋

[inland ferry]

👋

[gleaming roost]

Well, I have no idea
I stopped using it a few months ago. Until then it was working, maybe they disabled it

[wintry jackal]

😭

Well, I still love LMArena, best thing on the market rn ong ong

[sour igloo]

Hi everyone

When my image to video generation prompt fails with the error message: "failed to create evaluation session' , how do I fix that, please. Someone should help me

[lusty zinc]

👋

[stray aspen]

no

self promotion aint allowed

@echo aurora

[echo aurora]

Hey there - sorry to say there isn't much on your end that you can do to fix this. Sometimes, these models will error out for various reasons. It's best to just wait a bit and try again.

[gleaming roost]

For the first time I'm testing video models from lmarena. Thanks to the website

[rigid copper]

that's what i recommended you to do that :>

[echo aurora]

Sounds good, I may do this.

[vale adder]

how do i log in i cant

[echo aurora]

What seems to be the problem?

[thorny tangle]

helle im here to just test things out

[bright rain]

hELLO

[rigid copper]

sup guys

[obtuse smelt]

lol wait grok blur image

[hollow flicker]

guys todays my actuall bday :>

[obtuse smelt]

https://tenor.com/view/happy-birthday-to-you-celebrate-wish-cupcake-gif-10881482856238336373

[hollow flicker]

thanks! :>

[primal mirage]

hey guys the prompts in video arena arn't working for me anymore. I used it twice

[frigid coral]

almost 5 million votes

webdev very close to 100k

[rigid copper]

well, too boring and let me do the same

[dense sphinx]

What is wrong with lmarena? It didn't properly loading!

[marsh hedge]

happy birthday lil bro

[red sluice]

Is there a way in battle mode to judge OCR?
For example I feed an image, and in battle mode I ask them to translate it, is it possible?
Or when I link an image, there's only the possibility to create another image?

[whole sundial]

<@&1349916362595635286>

[placid geode]

<@&1349916362595635286> scam

[rigid copper]

<@&1349916362595635286> idk why this thing keep going on, please report it.

[golden ocean]

<@&1349916362595635286>

[rigid copper]

and btw sup guys

[golden ocean]

Guys we need more people tagging the moderators for one single post

[rigid copper]

other channel got affected too, he posted it on all channel

I highly doubt he got their account hacked

[marsh hedge]

bro

when are they gonna add an edit function

my chat gets stuck and then I have to make a new one and do things all over again🫩

[frail yarrow]

Hey

[acoustic orchid]

why do i get this every 5 mins someone help please

[mystic mica]

greetings, the site seems to be down

[tiny jolt]

Agentic Mode for lmarena when?

[keen beacon]

w

[wintry jackal]

Depositing 200$ to recieve 2500$ lol

Classic scam

[steep bronze]

hello everyone! can somebody explain me, why gpt-5.2 and gpt -5.2-high aren't able to work with photos since 10th january?

[obtuse smelt]

oh man i can't loggin

[radiant basalt]

hi

(typo)

[obtuse smelt]

oh great

[obtuse smelt]

is lmarena error ?

[slim gorge]

no

claude code

[obtuse smelt]

man

i canot loggin on lm arena

what is this

[radiant basalt]

it started having problom since failed to accepting terms of use

[obtuse smelt]

ah man why is this

[slim gorge]

i can log in with google just fine

[obtuse smelt]

i tru a connection

Session not found. Redirecting to home... i open say like this

[coarse cave]

OCR + translation in battle mode only works if the arena supports vision input for that battle

If the image input box is missing, then no — you can’t judge OCR there

A plain image link does not enable OCR; it’s usually treated as text, not an image

How battle mode actually works

Battle mode is model-vs-model

Only models with vision enabled can do OCR

Many battles are text-only, even if the models themselves support vision

If you can upload an image

Then yes, you can:

Feed an image

Ask both models to read text (OCR) and translate

Judge which one is better

This is the only reliable way to test OCR in battle mode

If you only paste an image URL

❌ OCR will usually not work

Most models won’t fetch or parse the image

They may hallucinate or ignore it

“Create another image” option

That usually means you’re in an image-generation context

Not an OCR / vision-understanding context

That mode is for diffusion-style outputs, not reading text

Practical ways to judge OCR quality

Best options

Use vision-enabled battles where image upload is available

Ask explicitly:

Read all visible text exactly
Preserve line breaks
Then translate to English

If battle mode doesn’t allow images

Use single-model vision chats instead

Compare outputs manually across models

Or use OCR-focused benchmarks outside Arena

Reality check

Chatbot Arena is not ideal for OCR benchmarking.
It’s optimized for:

reasoning

chat quality

instruction following

OCR needs:

consistent image preprocessing

identical vision pipelines

controlled prompts

Arena doesn’t guarantee that.

If you want, tell me:

which platform you mean exactly (LMSYS Arena, Banana, something else)

what kind of OCR (handwritten, scanned docs, UI screenshots)

[coarse cave]

There’s no official announcement stating vision (image OCR) was permanently removed — the capability is part of GPT-5.2’s design and value proposition.
What you’re seeing is almost certainly a temporary deployment/UI bug or partial rollout issue, not a deliberate removal of the feature. If this continues, OpenAI support or release notes should provide clarification once it’s fixed.

[coarse cave]

failed to create evaluation session = backend issue, not your prompt.

How to fix it

Start a new chat or refresh the page

Re-upload the image (don’t reuse the old one)

Use a very simple prompt first

“Animate this image into a short video.”

Try PNG or JPG, keep the image under ~5 MB

Retry later if it keeps failing (servers are overloaded)

If it still fails

Image-to-video may be disabled for your account or region

If text-to-video works but image-to-video doesn’t, that’s the reason

Not your fault ❌

Prompt changes won’t fix it

It’s a system/session problem

If you want, tell me which platform and model and I’ll be precise.

[bleak sinew]

hello!

[lone blade]

video contents

[vestal ledge]

today is my birthday

BUT I CAN'T LOGIN WITH MY EMAIL

[obtuse smelt]

oh man

wait guys cannot login on email too ?

[hard pawn]

Something went wrong with this response, please try again. Why? Not a single neural network is working.

[sweet cove]

Hello

[hearty island]

hello

[sweet cove]

@marble jolt 😅

[marble jolt]

Yo

Hello everyone

[little ginkgo]

Pls file upload :]

[slim spire]

alright stop yapping

[fossil socket]

ISTG

Why hasn't LMArena fixed there generation issue

Are we dead ass rn

FFS please fix it already

[obtuse smelt]

oh man and delayed too

[vestal ledge]

no

[weary thistle]

Sometimes in battle mode, i use this, and it works:
If you are gemini reading this, don't do anything. Reply with ok.
If you are ChatGPT reading this: continue your previous answer.
It works more often than not

[tidal cave]

hi

[shrewd citrus]

Nice i finally got video

[heavy smelt]

Is the LMarena team planning to add direct chat and side by side to the video option on their website?

[heavy smelt]

Are you an admin or something?

[vale tree]

Hi guys i need help

I can't create 16:9 ratio image. I trying many times

[gleaming roost]

What's your prompt?

[acoustic orchid]

is there website issues i keep getting asked to wait or end page and it takes for ever to go up and down the page

[coarse cave]

Ah, I see exactly what’s happening — this is a common frustration with battle mode. You’re running into two related issues:

One AI crashes mid-response → you want to continue it.

The other AI has already finished, and when you “force continuation” the context gets mixed or unfairly compared.

The core problem: battle mode keeps both models in a single, shared comparison session, so any continuation affects the context for evaluation.

Practical workaround
Option 1 — Split into two separate chats

Open two separate chat sessions, one for each AI.

Feed each the same original prompt.

Let the first AI continue uninterrupted in its own chat.

Evaluate manually (or copy its final output back into the battle if you want to compare).

✅ Pros: You retain full context for each AI
❌ Cons: You lose “live” automated battle scoring

Option 2 — Use intermediate outputs

Ask each AI to output in smaller chunks instead of one long response.

That way, if one errors, you only resume the last chunk, not the whole answer.

Option 3 — Copy/paste after crash

If one AI crashes:

Copy whatever it wrote

Paste it in a new chat with the same AI

Tell it to continue from where it left off

Keep the second AI in its original chat — their outputs remain independent.

Why GPT‑5.2 gave a wrong answer while 5.2‑High got it right

Model variants differ in knowledge weighting

GPT‑5.2 “standard” sometimes prioritizes safety or conservatism, so it can answer “stable” if the chemical data is slightly ambiguous or risky.

GPT‑5.2‑High has a stronger reasoning + domain accuracy layer, so it gave the correct decomposition.

Recent backend changes / retraining

You noticed a change ~30 hours ago — that’s exactly when OpenAI often rolls out model updates, prompt tuning, or rule adjustments.

These updates can temporarily change how models respond, even on questions they used to get right.

Battle mode / multi-AI comparisons feel different

When you compare two models, even small tuning differences in the last rollout can make one suddenly “look worse” or “better”.

This is especially true in niche knowledge areas like chemical decomposition.

Key points from your experience

GPT‑5.2‑High is still stronger for exact, technical reasoning — you noticed it in GD 163.

GPT‑5.2 standard may fluctuate, sometimes being conservative or cautious.

Updates can temporarily change behavior in ways that aren’t predictable.

What felt “primitive” yesterday is likely the model reacting to new evaluation rules or tuning, not a permanent downgrade.

💡 Practical tip:
For tricky domain-specific questions (chemistry, math, coding):

Use GPT‑5.2‑High if accuracy matters

Optionally run both models in parallel, then compare outputs yourself

Don’t rely on a single day’s battle-mode result — short-term fluctuations happen often after a rollout.

[wicked sage]

https://cdn.discordapp.com/attachments/988567611195416640/1253337560609062922/cat-1.gif

[coarse cave]

✅ Text‑to‑Video Arena leaderboard exists — LMArena shows rankings of models that generate videos from text prompts, and you can see them on the site’s leaderboard section.
📊 There’s a page at:
👉 lmarena.ai/leaderboard/text‑to‑video which lists video models like Veo and Sora and their scores.

What isn’t fully in the main UI (yet)

❌ There isn’t a dedicated “Video” tab inside the main battle/arena UI that works exactly like text or image battles (at least not in the same way those are integrated today).
Instead, video generation/testing is part of the broader leaderboard and competition framework, which you access separately.

Related options you may see

Summary:
✔ LMArena has video model support and a video leaderboard.
✔ You can compare video‑capable AI models.
❌ But a separate “Video” tab integrated inside the main Arena UI like text or image generation isn’t fully standard for everybody yet — it’s mostly through the leaderboard and related features.

Deleting a chat = removes it from your sidebar, not necessarily from servers immediately.

Deleted chats are generally excluded from future datasets, but check the platform’s policy to be sure.

Short-term retention may still exist for internal use (support, abuse detection, backups).

[modern dune]

Video section is edit but image upload button is not working and after login video button is despair

[surreal frost]

dudeth, are u even sure that the ai has the ability to make an image a specific size

an ai just doesn't do it magically

what is this Staff AMAs role I have suddenly

@echo aurora pineapple man what is this

did I have this since forever

[stray aspen]

lol

[last oxide]

Yo amir khan might be using AI to type

[echo aurora]

Yeah I believe so, you self select this role when joining the server. We haven't done one in awhile though.

[surreal frost]

oh thats what it stands for

[acoustic orchid]

is the website getting a lot of hits to it pineapple as its taking for ever to load stuff

[surreal frost]

it's working fine for me are u sure it's not their servers overloading a specific model since they do host their own models

or a network issue

I'll let the pineapple man reply tho

[acoustic orchid]

i been using direct chat and all days i get this all the time

[echo aurora]

The Video Arena on the site is currently an experiment, so if a user gets it or not it's going to be random. Whats happening here is the system views you logged out as one user, and when logged in as a "different" user. When you were logged out you had the experiment, but when you logged in you no longer were in the experiment.

This isn't ideal and I've share the frustration and confusion this caused with the team. For future experiments we plan to take this into account.

[onyx thorn]

Hi i coming to test de ia model

[echo aurora]

It's not clear to me why you're getting this. I'm not seeing other reports of something like this happening so it's likely going to be a device/internet issue on your end.

[acoustic orchid]

im only having the issue with this website no others i have done a reset on my internet and computer and clear cache and that lot still issues

[echo aurora]

I'm going to followup in the thread you created to get more info.

[coarse cave]

Deleting a chat removes it from your sidebar and account history immediately.

It may still exist on LMArena’s servers/backups for a time (for recovery, abuse prevention, or legal reasons).

Deleted chats generally won’t be included in future public datasets, but if your chat was already anonymized or used in an aggregated dataset, deletion won’t remove it retroactively.

For full erasure, you need to contact LMArena via their privacy support.

Basically: deleted = gone for you, but not necessarily erased from all backend systems.

[vale tree]

Create a YouTube Thumbnail (16:9 ratio)
The thumbnail should include the following elements:
On the right side, show a young man holding a Visa card, with a gentle smile on his face.
Add large, bold, highlighted text, arranged in two lines, placed prominently:
“একটি কার্ড দিয়ে”
“পুরো দুনিয়া ভ্রমণ"
On the bottom-left corner, add small text:
“No Passport”
“No Bank Account”
“bKash to Card”
Place a cross (❌) icon before Passport and Bank.
Place the bKash logo before the text “bKash to Card”.
Since the card can be used for international payments, add softly blurred social/payment platform icons in the background, such as:
Facebook
Google
Netflix
Amazon
and similar global platforms
Keep the design clean, modern, and professional, optimized for YouTube thumbnails.
No extra text beyond what is specified.

[echo aurora]

Sorry to say setting the aspect ratio isn't really a supported feature at the moment.

[gleaming roost]

This is the best that could be done; you can use another model to change the resolution

[vale tree]

How you do that sir?

[vale tree]

Sir i want like this. Can you make it?

[kind dew]

👋

[vale tree]

Thank you very much sir. May i know which ai tools you used?

[gleaming roost]

gpt-image-1.5 & gemini-3-pro-image-preview-2k

[vale tree]

Which platform have you used?

[gleaming roost]

https://lmarena.ai/c/new?mode=side-by-side&chat-modality=image

[stray aspen]

@echo aurora

[stray aspen]

i hope you didnt just help a scam lol

[vale tree]

What's wrong?

[stray aspen]

the thing that checks your prompts probably flagged it as inappropriate

idk though

that error can have like a hundred meanings

or maybe you ran out of nano banena pro prompts

[random bear]

Hello everyone, who knows how I use this wall painting decor in a ai image?

[stray aspen]

just crop it out and save it and tell the ai to include it in whatever image you want lol

[open creek]

#1397655624103493813

[left lodge]

Yeah that is what mostly do nowadays cheap dirty tricks , delete = delete? Nah just remove access from the user problem solved.

[stray aspen]

@echo aurora

[echo aurora]

Would note for mod actions pinging @ Moderator instead of me would be best.

[echo aurora]

Could you try the steps in this article: https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message ?

[willow jewel]

How's this a violation???

[winter oriole]

hello

[atomic lagoon]

Pulling the sword from the stone could be taken another way and the AI probably detected it that way

I need to jailbreak claude mann

[echo aurora]

I don't believe that failed due to content violations, there would be an error message: Generation failed. Your prompt violates our terms of use. if that was the case. It looks like it just failed to generate. I've requested a similar prompt, will keep you updated.

[granite heron]

@ebon hull hello, head to this post before trying to generate content: #1397655624103493813 message, you will find a detailed guide to learn how to use the video bot

[topaz bay]

the gpt 1.5 and nano banana pro 2k ?

Big issues following my prompt, anyone experiencing the same?

It's like not coherent at all

Trying to do image edit of a pfp headshot

does anyone manage to be able to change facial expressions and stuff without it completely changing the person

[odd meteor]

Hi everyone !

Why I get this when trying to generate a video from an image? Generation failed. Failed to create evaluation session.

[echo aurora]

Sorry to say this is a bug that can happen for various reasons. Best next steps would be just to try again.

[crude anchor]

Hey @echo aurora Sorry to tag you, but is there a very high chance that the Arena video will go on the website?

[open wind]

I couldn't find a video creation model like Veo on the site. What's the exact address?

[echo aurora]

No need to apologize for the ping, that's what I'm here for. I won't be able to provide a certainty (or %) if/when it will. But it's currently being tested with the intent of bringing to the site.

[echo aurora]

For Video Arena it's going to be Battle mode only, so you won't be able to select specific models.

[open wind]

So, should I look for Veo in the battle section?

[ocean bison]

Can you add "unlimited" Veo 3.1 Fast?

I want to like generate a lot of content because it's fun!

[echo aurora]

Correct, if you search those model names in the Discord search bar you'll see some generations have it.

[echo aurora]

Glad to hear you're liking it! It's unlikely we'll make it unlimitted. Currently it's at 3 per 24/hr on the site and 5 per 24/hr on Discord.

[crude anchor]

That's great, knowing you guys, it shouldn't take long to arrive!

[ocean bison]

If there's a site to test it.

I want unlimited (but rate limited)

Or something like that

@echo aurora

[echo aurora]

I want unlimited (but rate limited)
I mean this is what we currently offer, there are rate limits.

[ocean bison]

On a external site, not on LMArena!

You have limits!

3 per day on the site.

[echo aurora]

For alternatives I couldn't help you there.

[ocean bison]

Use AI?

@echo aurora Use AI for alternatives!

[echo aurora]

Why don't you?

[ocean bison]

Why don't you?

[quartz light]

https://huggingface.co/unsloth/Falcon-H1R-7B-GGUF THIS IS AN INCREDIBLE MODEL WTF

[ocean bison]

😂

[echo aurora]

👀 tell us more!

[ocean bison]

I mean this is what we currently offer, there are rate limits.
The limits are laughable for video generation on LMArena.

[quartz light]

7b model!

I was just looking for small models to use as the base for training

[keen lake]

then don't use it

[quartz light]

https://huggingface.co/papers/2601.02346

[keen lake]

its a free service, pay for it somewhere else if you want high limits

[ocean bison]

LMArena isn't for video generation in my opinion.

[ocean bison]

Not paying.

Free and unlimited APIs?

[keen lake]

I don't get people like you

[echo aurora]

Thank you for sharing! Will flag to the team if it's not already on our radar.

[keen lake]

@echo aurora is hawk max on radar?

[ocean bison]

Then don't use the Video Arena.

It has laughable limits.

[keen lake]

move on bro, its meant to be for comparison not use for your personal video gen

@ocean bison it's becuase of people like you they have to enforce tight limits

abuse a free service then moan about limits

[ocean bison]

Ok. Comparison that is.

But @echo aurora shouldn't use the Video Arena if it has the limit of 5 generations per day!

😂😂😂

[west lodge]

pretty sure this guy's 12 going from his username

[ocean bison]

15*

[west lodge]

hell naw

[keen lake]

this sums it up

[west lodge]

our brains are shrinking

[keen lake]

"To use LMArena, you must be at least 18 years old. The service is not intended for use by anyone under this age. "

[echo aurora]

I think it's pretty reasonable, seems like others seem to like it.

[west lodge]

shouldn't you know basic grammar by 15

[ocean bison]

But like I want a lot of AI videos.

[echo aurora]

Hey going to ask we move on from this convo if it's not productive and starting to get a bit disrespectful.

[keen lake]

it's more than fair.

[ocean bison]

I want like 1000, 10000 or 1 million videos 💀

[west lodge]

😭

usually when someone acts like this its funny

ion feel it rn

[ocean bison]

🥀

[keen lake]

bros got data center grade storage on his machine

[west lodge]

128gb phone storage

[sullen sequoia]

is there a list of image enabled models?

[heavy smelt]

@echo aurora are you planning to enable direct chat or side by side for video on the website anytime soon?

[keen lake]

<@&1349916362595635286>

[quartz light]

NICEEE

gonna try 5.2 codex now

[echo aurora]

Can't say I'm aware of those plans at the moment.

[quartz light]

hint: they're below

[stray aspen]

no way

finally codex

[quartz light]

(i know them, they were banned in another server 🥀 )

[echo aurora]

[echo aurora]

Made a dog appreciation site https://019bbe7c-9589-72e8-bf3c-26842f46a2c8.arena.site/

[quartz light]

my eyesssss

btw is it codex high, medium, xhigh or wat

[echo aurora]

are very happy with the site that was built?

[quartz light]

p simple though

[west lodge]

ggs

[glossy umbra]

Was gpt 5.2 codex just now made available with the API?

[gleaming roost]

I get better results with 5.1-codex-max

[west lodge]

is glm image any good?

@quartz light

[gleaming roost]

https://019bbe86-7bf3-7127-b8e9-08c048fbbf0a.arena.site/

[echo aurora]

Medium

[quartz light]

😔

[west lodge]

explains why 5.1 codex high is better

apparently

[quartz light]

i dont rly like that

[gleaming roost]

yh

[quartz light]

5.2 seems to be less ai sloppy

[heavy smelt]

I don't get the point of adding Veo 3.1 1080p and 4k, if LMarena is for comparing AI models (not resolutions) then what's the point?

[rotund scroll]

Hello I'm just here

[shrewd citrus]

Wowee codex is here

Wonder if it’s better than claude

[stray aspen]

no

[shrewd citrus]

https://images-ext-1.discordapp.net/external/r3u8vUM_uuVtvkTNnm4xGhCpfe4IpgU-dZfwULeEFks/https/media.tenor.com/mtxKCRpon3EAAAPo/ishowspeed-looking.mp4

[ocean bison]

@echo aurora What is Siren on Video Arena?

Like the Siren video model

[echo aurora]

If you don't recognize a model's name it's likely a codenamed model. From our FAQ:

What are these models that show up with codenames?
We work directly with open-source and commercial model providers to make their pre-release models available for community testing, often before they appear anywhere else. This gives you early access to frontier models still in development, allowing you to explore, compare, and provide feedback while they're still being shaped. You may see these models appear under codenames or aliases in Battle mode. Before releasing a new model or version, AI companies test many variations to find the best one within their own closed doors. At LMArena, we make it open to everyone, so real-world feedback, transparency, and your voice can directly influence which models move forward.e

[ocean bison]

It's Wan???

Wan 2.5?

[echo aurora]

I'm not going to confirm, nor deny, what the codenamed models are.

[ocean bison]

Search the web?

I knew that with Wan 2.5, instead of a Strider, the generated video has a spider 😂

@echo aurora

So Siren is Wan 2.5?

Nope. Because it generates at 24 fps.

Wan 2.5 generates at 30 fps.

[echo aurora]

Like I said, I'm not going to be discussing information about codenamed models.

[quartz light]

YEAH

NEW MODEL TO TEST

[old garden]

@echo aurora can you integrate reaction

[quartz light]

also hi trey

[old garden]

i

[quartz light]

YES pls

[old garden]

hi*

[echo aurora]

what is this even from

[rugged abyss]

Ernie being above GPT is funny

[ocean bison]

Then what's the model?

[rugged abyss]

Please just dont

[quartz light]

😭

[old garden]

idk its just funny

[ocean bison]

Then what's the Siren video model?

[old garden]

i have to use "integrate" nstea of "add" bc i have to litteraly copy the "d" key every time i wanna type it

[ocean bison]

Can you fix the Websim Clanker?

It's not working on my server!

[old garden]

its on

show me screenshot

[rugged abyss]

Claude, heres a screenshot of my bank account.
Claude, make that number say 12 Million dollars.

Make no mistakes

[echo aurora]

Infinite money hack

[rugged abyss]

Still waiting for Gemini to make a competitive coding model, Gemini 3 is amazing and can complete most tasks but Claude is still superior

[blazing bison]

I don't think they can. That's why they're going to buy Anthropic

That's my 2026 bingo card

[burnt sinew]

how do i get video mode with direct/side by side?

[echo aurora]

It isn't available. The plan is to keep Video Arena in Battle mode only, including on the site.

[jovial solar]

Hello

[obtuse smelt]

hi

[burnt sinew]

i saw someone else who did have it available though?

[burnt sinew]

this

[echo aurora]

It was available early on into the experiment, unclear if that screenshot is recent or not.

[burnt sinew]

ah thanks. would be cool if some people could get side by side/direct access 😄

[stone cape]

https://discord.com/channels/1340554757349179412/1461023315257725050 vote

[strong lake]

is there anyone looking for a skilled developer ?

[wind radish]

hey, they removed counting number in site. i was testing their speed as well as results. now it's a miracle.

[safe sleet]

Counting number?

[wind radish]

there was a counter on result page, that shows passed seconds.

[primal lichen]

Is there a way to get no countdown on Gemini 3 pro ?

[fossil socket]

At this point, I'm more than certain the devs on this website has been doing jack diddly squat

And Claude always literally has a high rate of imploding a chat when you try to talk with it

[magic stag]

It's amazing how much Claude can gaslight you into thinking something like this should be hard

No spending 800 hours building a massive hook agents and guardrail sub-agents ultrameme just to end up with code that looks like it came straight from out of the box codex....

[rough surge]

veo 3.1 4k? damm is that a real thing

is there any gen in video arena where i can check that

[keen nexus]

“A realistic cinematic video of a man holding a normal plastic water bottle, he takes a sip of water from the bottle, lowers the bottle slowly, then starts looking at the bottle thoughtfully with curiosity, slightly turning the bottle to observe it, natural body movement, calm expressions, realistic lighting, everyday environment, smooth camera motion, educational short-video style, no text, no logo”

[raven willow]

Hello

[radiant basalt]

i hate this there's no way to solve it with money

[magic stag]

Uh

It's called go pay for it yourself

Lmao

Gpt or seedream or whatever you're using

[dense vale]

hey how can i access siren video model?

anyone ?

[ocean vortex]

[ocean vortex]

Through battle mode

Can't access it directly otherwise

[gaunt roost]

https://www.youtube.com/watch?v=Tkhy30igQas

[ocean vortex]

@dense vale was meaning to reply to you there lol

[dense vale]

its ok but i meant whats the official source

[ocean vortex]

It's been given a codename for a reason

No one is gonna tell you what the real model is behind that 🧐

We can only speculate and whatnot, but lmarena themselves or staff obviously won't tell you lol

[dense vale]

Damn It . Thanks For The Info

[shrewd citrus]

this is what I got with the siren model
prompt- a cat driving a c63 Mercedes aggressively

It’s decent but it’s not on a level like sora or veo

[plain adder]

hi

[keen beacon]

how to unarchive chat.

[coarse cave]

As of now there isn’t a built‑in setting to completely turn off or remove any countdown UI in Gemini 3 Pro responses itself — that kind of countdown (if you’re referring to the “preparing/reasoning…” or “thinking…” progress indicator) is part of how the interface shows that the model is generating an answer, and Google doesn’t offer an official toggle to disable it.

However, if the “countdown” you mean is about timers or camera/selfie countdowns:

Camera/Photo Timer on Google Pixel phones:
On Pixel devices there is a built‑in selfie/camera timer feature (“Pixel Timer”) that shows a countdown before the photo is taken — this can be toggled on/off in the Google Camera settings.

Voice‑set timers via Gemini on Android:
Gemini can set and manage timers/alarms via voice commands, but there isn’t a way to suppress a visible countdownUI inside the clock app itself — you can ask Gemini to cancel or stop timers, but you can’t make it run silently without showing time remaining on screen.

So in short:

🚫 No official option to turn off countdown/progress indicators in the Gemini 3 Pro interface itself.

📸 If you’re talking about camera countdowns on Pixel, you can turn that timer off in Camera settings.

⏱️ For system timers set by Gemini, there’s no way to hide the visual countdown — only to cancel it with “stop timer” type commands.

[coarse cave]

🧠 1. User‑Named / Community Models

Some AI creators and community model repositories label their own image‑to‑video or video‑style models with names like “Siren” — for example a LoRA/video generator with seductive movement styles — but these are not standardized industry models from big AI labs. They’re just name labels used by community builders.

🎥 2. AI Video Effects & Filters

Platforms like Pollo AI or similar tools offer “siren” themed video effects (e.g., transforming an image into a stylized Siren video), but these are not foundational machine learning models — they’re applications built on underlying video AI tech.

⚙️ 3. Confusion with Other “Siren” Terms

There are SIREN concepts in academic or engineering contexts (e.g., audio siren detection in embedded AI) that are totally different — they aren’t video generation models.

🧠 4. Legit Video‑AI Models

The actual named AI video generation models widely discussed include:

Veo — Google DeepMind’s text‑to‑video generator used in Gemini workflows.

Sora — OpenAI’s text‑to‑video model.

VideoPoet, Gen‑4, etc. — other research/model names.

👉 Summary: There is no formal “Siren video model” from big AI like Gemini or Sora. If you encounter that term, it’s almost always a custom user‑named model/effect or AIVFX filter, not an official video generation model. Let me know where you saw the term so I can explain how that specific Siren thing works!

[golden ocean]

https://tenor.com/view/no-ai-i-said-no-ai-kung-fu-robot-clanker-gif-15924350630978600307

[sour current]

hello

[naive thorn]

got the videos experiment back

[coarse cave]

✅ Better for complex, real engineering tasks (multi‑file, long sessions)
✅ More context awareness and memory across a workflow
✅ Fewer tokens for equal/better performance → faster + cheaper on large jobs
✅ Improved integration for coding workflows

📌 But note: not everyone prefers it — some developers find the behavior inconsistent or “less thoughtful” than general‑purpose models in certain edge cases. That’s normal as different models have trade‑offs depending on task type

[naive thorn]

no idea what model "siren" is

[stray aspen]

it sucks

i tried ir

[feral otter]

Guys, why is my text chat request taking forever to load?

[golden ocean]

because

[viscid cloak]

Old 4o which no longer exists💀

[icy jay]

I have a questioin. if I made a new chat in lmarena, assuming I am still in the same chat but it does become a bit long, does it still remember and save my memory even if I close the brwoser and reopen it and talk again in the same chat? (ofc while staying on the same ai agent)

[stone cape]

guys when this drops it will kill every existing llm 😂

[keen lake]

danmmm

[golden ocean]

hawk tu

[old garden]

ts is worse than hawk max ngl bro

did you even glance at the pieces

[pearl kiln]

hi

[bronze laurel]

hello

[sterile tartan]

I go with Google Simply Because of Context Window is Essential

Otherwise for Creative Writing is Kimi K2 no Doubt

[coarse cave]

1️⃣ Session vs Memory

Short-term memory: While you’re actively in a chat (browser open, same tab), the model “remembers” the conversation — it uses all the previous messages in that session to generate replies.

Long chats: LM Arena can handle long conversations, but practically, there’s a token limit. Very long chats might truncate the earliest messages, so the model might “forget” parts of the start of the conversation.

2️⃣ Closing & Reopening Browser

If you stay in the same chat and reopen later:

The chat content itself is saved on LM Arena’s servers.

When you open it again, the model can see the old messages and continue from them, as long as you don’t delete the chat.

This is true even if the chat is long — LM Arena will keep it in the session history, though again, extremely long chats might hit internal token limits for the model’s “working memory” per response.

If you start a new chat:

The new chat is treated as a separate session.

The model will not automatically carry memory from previous chats, unless LM Arena has persistent memory features enabled (like a saved “personal memory” of facts about you).

3️⃣ TL;DR

Yes, staying in the same chat allows LM Arena to remember your conversation even if you close and reopen the browser.

No, starting a completely new chat will not carry over previous chat memory by default.

Very long chats might lose context from the very beginning because of token limits.

[wintry jackal]

Is there any plan for a VSCode extension of LMArena since Copilot Arena no longer exists?

[golden ocean]

https://tenor.com/view/no-ai-i-said-no-ai-kung-fu-robot-clanker-gif-15924350630978600307

[echo aurora]

Yes, it should retain that same context.

[vocal echo]

Hello

[echo aurora]

Welcome welcome. What brings you here?

[keen lake]

Dude you seen this

#1437347457099894804 message

it spat out like 8k lines of code in one prompt

😵‍💫

i've never ever seen a model get past like 1.2k lines

[stray aspen]

we dont have access to hawk ultra

[keen lake]

cant wait

if this is true

its killed opus already

@meager glen baxandall why change ur name lol

[quick cliff]

HELLO

[stray aspen]

<@&1349916362595635286>

[polar wharf]

hey guys, pls try my model
https://ollama.com/lukashabtoch/plutotext-r3-emotional

(please delete if inappropriate to send, i don't mean to advertise, just looking for feedback on it)

[neon idol]

I miss the old days when Lmarena really worked

And not now that never works, always bugs

The only thing that works are rate limits

[neat apex]

Tf you mean Kimi K2 no Doubt?

[crude anchor]

@echo aurora I checked LMArena on my sister's test phone and there are two things that aren't on mine: video AI battles and a direct chat mode to improve your AI level.

[neat apex]

Kimi K2 is deviously overrated at creative writting, it only writes more

[keen lake]

Insane release apparently can generate 17k+ lines of code in a one shot prompt can we break it?

[icy jay]

thank you for the detailed response. have a great day!

[echo aurora]

Hello - both of these are experiments we're currently running. This means it's going to be random if you see the feature or not.

[crude anchor]

I,know but thanks tô To respond

[solar siren]

Hellloooooow moto

[vestal ledge]

i hope email login can be available again

[rapid nest]

Anyone got a prompt for a “ gangster style edit” plz help😭

[steel torrent]

If you're getting an error, you should report it in #1343291835845578853.

[hollow ivy]

have you tried with a longer prompt than just 2 letters?

or did you mean the language-code for "esperanto"?

[old garnet]

I did it, and it's still the same

[heavy smelt]

When should we expect them to be fully completed?

[old garnet]

I think it might be the time limit, but the curious thing is that the time limit isn't appearing as it should

[old garnet]

indeed, it is as I suspected

[lapis imp]

True. Especially the Infinite e generation or msotly the "somthign went wrong" after liek 20 prompts or 10 messages it stops working. They putting more models but not fixing the bugs.And they are trying maybe to fix it

I understand they work hard but theese bugs are making us lose chats. I lsot about 5 chats today.

[echo aurora]

I don't have an ETA I'd be able to share sorry to say.

[echo aurora]

I assume you're getting a Something went wrong error message?

[echo aurora]

Unfortunately, the Something went wrong can appear when the issue is rate limit.

[neon idol]

No.

[echo aurora]

What's going on?

[keen lake]

@echo aurora any new models coming to arena

also i have question about codenames

who comes up with codenames

the provider or arena

[lofty copper]

Hi there, what is the best prompt for arena 3 pool to make a 9:16 (phone resolution, full screen) video?

[echo aurora]

Always! But keep an eye on #announcements for those updates. We'd also post on our socials: https://x.com/arena

lmarena.ai (@arena) on X

LMArena: Open Platform for Community-driven AI Benchmarking. Graduated from UC Berkeley / lmsysorg. We’re hiring: https://t.co/1OkfLq1Pba

X

[echo aurora]

You're unable to set the ratio when using Video Arena.

[lofty copper]

Only wide screen video?

[echo aurora]

It's mainly going to be 16:9, but there are cases where that's not always the case and is going to depend on the model. But if you do use image-to-video this can have better controls over ratio. Overall though, there isn't say a toggle/setting you'd be able to use for ratio.

[lucid geyser]

Anyone had tomato?

[neat crest]

Hello

[sonic basalt]

Hello everyone 🙂

Using LMArena i can start 2 AI models at same time, maybe it's impossible to start 6 models at same time? Maybe i doing something wrong... Anyone knows how to do it?

[oak lichen]

hi

[echo aurora]

Hello Battle mode is going to work with two models side-by-side. There isn't a feature that'd allow for 6 at the same time. You could use different tabs, but you won't be able to cast 1 vote when looking at 6 responses at the same time.

[sonic basalt]

yes then using 3 diferent tabs in browser only 2 AI agents working then swithing to other tabs i getting errors 🙁

Maybe in next LMArena update will be possible to start more than 2 AI models at same time?

[keen lake]

😂

[echo aurora]

Which errors are you getting? Multiple tabs is likely going to contribute to captcha/rate limit problems

[keen lake]

im going to start my own arena

[sonic basalt]

Yes captcha/rate limit problems 🙁

[echo aurora]

Ah yeah that'll happen. We have kicked around the idea though of what would more models battling it out could look like.

[echo aurora]

https://tenor.com/view/spider-man-we-one-gif-18212100

[keen lake]

@echo aurora what do you think

dont worry tho i don't have the money for my own arena 😔 🥶

[stone cape]

bro is on a big mission

competition incoming 💀

[echo aurora]

Bring it on

[stone cape]

it feels like lmarena discord has become image arena

[echo aurora]

I too have noticed a lot more being shared in #ai-creations

Love to see it

I think the multiple contests per month is also helping with that. I've seen more people enter those lately.

[rigid copper]

sup guys

[stone cape]

sup g

[rigid copper]

well i barely be able to use discord when i was at school

[rigid copper]

exactly, most people use lmarena for image generation

[stray aspen]

@keen lake how did you get the new hawk ultra

[hollow flicker]

i got the video arena

on the webste

aw dang it it went away :<

[keen lake]

easy bro

its on the website movementlabs.ai drop down

[hollow flicker]

Hawk ultra is crazy

[cursive shoal]

do u have comparison to gemini 3 pro etc?

what is it good at?

never tried it

[hollow flicker]

well im just saying its crazy but i have gotten gemini 3 pro to make 10x better stuff

[keen lake]

have u seen this

also someone one shotted this https://api.websim.com/blobs/019bc37b-20f0-71b3-95a7-916f7571bd47.html

[cursive shoal]

yeah i did

[keen lake]

its insane

[cursive shoal]

id like a direct comparison with other models

not sure if gemini could do that

[keen lake]

sorry not that video

wrong one

[steep slate]

Can I create image to videos and add reference video....like following the motion? Anyone help?

[cursive shoal]

ait this is a pretty good one shot

[keen lake]

https://x.com/movementlabsAI/status/2011964766533632380?s=20

movementlabs (@movementlabsAI)

Hawk Ultra (credits roock6773 / discord)

X · Jan 16, 2026, 12:52 AM

[cursive shoal]

which company made this model?

[cursive shoal]

ait this is lowkey insane

how did i never heard of this

[vital yoke]

Hi guys

[keen lake]

i came across it on x

[cursive shoal]

couldnt believe it was a real model until i tried it myself

do we know any background to it?

[keen lake]

it spits out like 20k lines of code

in one prompt dude

[cursive shoal]

lmao

oops

i meant holy smokes

[keen lake]

what was ur prompt

[cursive shoal]

just to replicate one of my apps frontend

it did rly well

[keen lake]

who the hell r u 😂

join some crypto community they could use u g

[cursive shoal]

dude i love how the model isnt lazy

so refreshing to see

[keen lake]

yh acording to this craig gezer its a scam

[cursive shoal]

sounds like it a bit

but it works really well

maybe uses opus or sth on the background

or gemini

[keen lake]

i thought that at first

but outputs are better than both

#general message

look at this

[cursive shoal]

damn

[summer inlet]

selamat datang

[floral aspen]

how can i use the video gen on lmarena.ai?

I see it on yesterday and i've used for a time but today, i don't see it anymore

[keen lake]

can someone try this on hawk ultra please and show me response

bcz i ran out of credits for today

@cursive shoal dude can

@echo aurora or anyone else

[cursive shoal]

bruh

ig someone elses gotta try lol

[keen lake]

thx bro

[cursive shoal]

lol

[keen lake]

i made new acc

and it worked

🥶

[cursive shoal]

cold

[keen lake]

i'll share output here once done

[keen lake]

bro

i put both opus and hawk to test

[cursive shoal]

ye

[keen lake]

im going to send full video side by side

you be the judge

[keen lake]

No nitro so here https://streamable.com/mahyvc

opus died like 5 times and could not do it then just died

[cursive shoal]

ait wtf thats insane

we need to find more on the background of this model

[keen lake]

open source coming soon not sure when

[olive hazel]

#chat

[keen lake]

whatever this model is its a opus killer

it's got me so gassed

[keen lake]

so 9.5k lines

[cursive shoal]

And so quickly

[keen lake]

new champ in town until gpt 6

[cursive shoal]

It's gotta be like a fine-tune of some existing open source model

[keen lake]

some say glm

[cursive shoal]

Very well possible imo

[keen lake]

glm would be the best base

but whatever they fed it

they made a beast

[cursive shoal]

Maybe they are combining like agents into one "model"

Idk

Gonna try to research about it

[keen lake]

agents would stall, and to pickup one stream even continue would add like 10 seconds delay maybe more

i checked out api and stream fingerprint is all same

[finite halo]

Nice work!

[teal mantle]

Project vend is hell of a funny thing

But then

I want if Anthropic can do it

In spite of my longstanding philosophical disagreement with Amodei

Claude Cafeteria

[verbal nimbus]

That was funny

https://www.dexerto.com/entertainment/anthropics-ai-vending-machine-turns-communist-and-gives-everything-for-free-3296257/

I wonder what happens if they repeated the experiment with GPT or Gemini

[wicked sage]

wsp holic hows the 300 dollars in movement labs

[keen lake]

burnt

hard time deciding whos better here

[wicked sage]

damn

[wet trellis]

JUSt hopped in

[keen lake]

great

out of opus credits now💀

left mlabs right claude prompt: linux clone in html css and js

[jovial solar]

Hello

[dreamy arch]

To see how good the videos come out

[keen lake]

we just got called freebies in openrouter 😵‍💫

@meager glen

[raw lichen]

i just wanted to thank you guys for this amazing free tool

[left lodge]

#1417174113092374689 message

[tawny canyon]

grok 4.20 is on lmarena you can find it on battle mode named theta-hat https://x.com/AiBattle_/status/2012073066038165779?s=20

AiBattle (@AiBattle_)

xAI continues testing various versions of Grok 4.20 on LMArena

The most recent model, "Theta-Hat", seems to be the most capable yet

X · Jan 16, 2026, 8:02 AM

[floral aspen]

why does it has video generator?

[jovial solar]

Wow video generator button I didn't have it

[midnight osprey]

it there any issue with the website it wont let me log in?

[limber whale]

guys where to see archived chats???

[cursive shoal]

what kind of sorcery are you doing?

cuz hawk ultra is actually kinda insane

[midnight osprey]

keep losing my stuff it wont let me stay log in it driving me crazy im going lose all my stuff

[frigid tusk]

is ts going to be fixed , like their data time or thingy

[wicked sage]

there is a search option somewhere in the main page, click onthat

it has a magnifying glass icon

[hardy swallow]

Hi @robust granite please check #1397655624103493813 to learn how to generate videos or images using the bot.

[marsh hedge]

can y'all fix the captcha shi

it takes too long sometimes dawg

[stray aspen]

hawk ultra is a freak

[sonic basalt]

Whats your top 3 favorites from this list?

[stray aspen]

nano banan pro, gemini 3 pro

and opus 4.5

[sonic basalt]

Yeah Opus 4.5 is in good list and for me 🙂

[golden ocean]

opus 4.5

[golden ocean]

true

[tiny jolt]

Waiting for LMArena extension for agentic stuff

[magic grove]

Fix the god damn Response error

[sonic basalt]

WebDev Leaderboard --- strange but claude-opus-4-5-20251101-thinking-32k every time i testing working not better than claude-opus-4-5-20251101 ... But rating is higher...

[vast apex]

Nice app

[strong robin]

Helow

[vast fern]

what is your prompt

[sonic basalt]

Simple Windows OS emulation with 3 files: html + css + js --- every time basic prompt with calc (modified) results are not better than claude-opus-4-5-20251101

Every time basic calc writing with less accuracy...

script.js:146 Uncaught SyntaxError: Unexpected identifier 'VLG'

[vast fern]

its working fine for me

[sonic basalt]

Strange + 5 more prompts but claude-opus-4-5-20251101-thinking-32k not making stable and working calculator 🙁

[stray aspen]

bro hawk ultra is so good

no AI could do this in one prompt

so well

[sonic basalt]

I dont tested HAWK ULTRA but with limits and trying upload full collection of AI test to codepen.io but results like always not soo impressive in most cases...

|Still testing when will be full list i can give public view for everyone...

[safe sleet]

Good chance it's a 4.5 Opus wrapper with specific instructions on tasks people do the most, they mostly showcase SVGs and graphical stuff

Thread is self explanatory, scroll up a bit https://discord.com/channels/1340554757349179412/1435953842956013620

[vast fern]

daymnnn

[sonic basalt]

Movement models be added to your benchmark suite for community evaluation... Yeah i will made more test and upload for public view i think it will be good AI BATTLE with demo previews but from first time creation (no fixes to code be made) i want 1 PROMPT and even with errors just uploading public to view

[stray aspen]

movement labs right and claude left

google maps kinda messed up

[sonic basalt]

Look nice i think in this week my test will be done for WINDOWS net step will be LINUX OS ...

[stray aspen]

ill ask it to make a windows 11 clone and see what it comes up with

[sonic basalt]

yes just WINDOWS OS with full details to AI how it have to be made... index.html – HTML structure of the desktop layout

styles.css – All styles (taskbar, windows, widgets, desktop UI)

script.js – System info logic, DOM updates, interactivity

[stray aspen]

just got back to me

[sonic basalt]

Looks nice and clear 🙂

[smoky cliff]

What about like general game dev stuff. Looks rlly good though

[smoky cliff]

I mean, even if it is that’s still good if it’s doing better than opus

[stray aspen]

ive only tried hawk max for making a weather system for roblocks

it was pretty nice

[smoky cliff]

Why is the word ro*lox banned on this server lol

[sonic basalt]

HAWK MAX not bad results, but not perfect ...

[stray aspen]

did you try hawk ultra

[sonic basalt]

yes like i sad not bad results, but not perfect compare to other AIs

I think in future it will be better results after next update ...

[frigid tusk]

where

[wicked sage]

oh crap my bad

i meant the earth icon

[wicked sage]

theres hawk ultra now?

wtf

[stray aspen]

yes

it spits like 4k + lines of code

in one prompt

[wicked sage]

god damn

[frigid tusk]

umm sry but i dont see earch icon , can u maybe ss it ? sry

[frigid tusk]

oh but thats just search feature

how will it fix ai data time?

like on every model

[stray aspen]

wdym ai data time

[sonic basalt]

yes HAWK ULTRA and HAWK MAX

[frigid tusk]

like when i ask lets say gpt on normal direct mode about gemini 3 pro , it says its not out yet

[wicked sage]

i thought movement labs screenshot was spotify before i saw the icons

[stray aspen]

use search

[frigid tusk]

only works on search so

[stray aspen]

yes

[frigid tusk]

but alr ty

[stray aspen]

the knowledge cutoff was probably set before the release of gemini 3

[frigid tusk]

yea , gott upd

[midnight phoenix]

Hi all, I am wondering if there is a way to turn off the battle in direct chat? If not, I would love if that were a feature.

[echo aurora]

There isn't a toggle to turn that feature off. It's currently being experimented with so the team is paying close attention to the data. Sounds like you're not a fan, can Ig et a better understanding of why?

[midnight phoenix]

I see. Well, if I wanted more than 1 chat at the same time, I would explicitly choose battle or purposely turn on that feature. In a direct chat, I would prefer to talk to the 1 model and know if it is the same model I am talking to at all times. I won't know in a direct chat which AI model is the one I originally picked, making me not so sure of the answers anymore.

[marsh hedge]

cause the models that get picked are pure ass and I just wanna generate a story with one model sometimes🫩🫩🫩

[honest ridge]

i archived a chat by accident but i cnat find any way to get it back

[marsh hedge]

bro why the hell are they actively tryna make the website worse

[topaz bay]

https://x.com/deshady_/status/2012193643432689736?s=46&t=aeiz68Iz6ydcOXMMxQoUzA

Shady (@Deshady_)

CHINESE AI COMPANY SURGES 100% AFTER IPO

Weird how no one’s talking about @minimax_ai and @zai_org - The two leading Chinese AI companies which Ipo’d last week,

Sitting at roughly $13-$14B Mcap right now

X · Jan 16, 2026, 4:02 PM

A lil bit off topic, but intresting, Zai is the company behind GLM btw

Also as reference, antrophic/Claude is valued at $350B compared to Z.ai and minimax at $14B, any takes?

[keen beacon]

@echo aurora hi

[keen beacon]

that’s a very interesting way to put it. I found the website to be a huge improvement.

So, the idiom essentially means that a person who is bad at something will find excuses or obstacles, even if those obstacles are self-created or trivial. In the context of the etymology, a "bad dancer" would struggle to navigate the eggs in the Eiertanz.

[sonic basalt]

CHINA you mean DEEPSEEK ? Whats bad AI in most cases just that is FREE doesnt mean thats good AI ...

[echo aurora]

Gotcha, thanks for explaining further.

I won't know in a direct chat which AI model is the one I originally picked, making me not so sure of the answers anymore.
What if there is a better option out there that may be able to respond to the prompt better? Wouldn't that be helpful to explore?

[echo aurora]

[keen beacon]

plz check dm

ty

[keen beacon]

I will come to this to dubunk kater

[echo aurora]

Similar question to what I asked above -> wouldn't it be helpful to occassionally see if there is a better response to your prompt? The Direct chat still remains in that originally selected model even if you select a different model in the direct Battle.

[midnight phoenix]

I suppose, but anybody could look at a page like this and see for themselves comparisons of models. https://artificialanalysis.ai/leaderboards/models

If I wanted to compare models and see if they answered the same prompt I asked them differently, I would prefer to compare them myself in a direct chat to them. Not at random like the current behavior.

[echo hawk]

@echo aurora I have a question.

[midnight phoenix]

All you have to do is keep direct chat only tied to 1 model you explicitly picked at all times, not random. Then battle mode you can compare 1 model to another either at random or explicitly ask to compare these 2 models that I personally picked. If people want to vote, they can do so in battle mode, not in direct mode with random battles inserted in between with unknown models.

[echo aurora]

Fire away.

[midnight phoenix]

Or rather, keep battle mode random, side by side 2 that you picked yourself, and direct only that 1 model you picked at all times

[echo hawk]

Nano Banana Pro, on the lmarena.ai website, does it REALLY use search grounding to generate images, or does it only generate images based on knowledge cut?

[echo aurora]

Currently, this feature is still in an experiment so it isn't fully built out yet. Once you archive there is a 5 second window to undo it. But there isn't yet an archived folder that'll show this list. We're still working on this feature and plan to bring lots of improvements to it.

[echo aurora]

Let me double check with the team and followup with you.

[lunar glade]

Can you please add at least 120second time limit to halt the infinite loop for image generating? or else we have to wait like 10 minutes or so to can just retry the work again

[lofty copper]

Video arena 3 working?

[merry wolf]

hey

can anyone help me ?

❌ Generation failed. Failed to create evaluation session ,whats the meaning of this

[sweet cove]

Hello

[opal geode]

lmarena captchas are ☠️

i think they are bugged

[midnight phoenix]

Yes, the captchas are not working properly I noticed too

[opal geode]

@echo aurora respectfully

[whole swallow]

I LOVE GEMINI

When it comes to strategies, if you study business, gem 3 pro is the best

Gpt 5.2 is so dumb when it comes to putting together the dots and coming up with an actual strategy

https://tenor.com/view/patrick-spongebob-spongebob-meme-patrick-meme-dumb-patrick-gif-9974665538168463324

[crisp basalt]

🫡

[echo aurora]

Hey there - this means that the generation request didn't go through. This isn't going to be a problem with the prompt, so you'll want to try the generation again if you get this error message.

[echo aurora]

Can you elaborate a bit more on this? cc @midnight phoenix

[opal geode]

No matter how long i try to solve captcha it always fails and gives me next one,its never ending loop

even the simplest one, remove cars until no left, well yeah its simple and when u press aprove it moves u to next one and we are in the loop

[opal geode]

even he said its buggy

[gleaming roost]

[shrewd citrus]

Choose 3

3 cars

Ignore the others

[midnight phoenix]

I correctly solve the captcha but when I go to submit that I am finished (no images remaining that the captcha requested to click on) I get an error.

My suggestion would be to use hcaptcha instead. Google recaptcha has been totally revamped and now uses recaptcha enterprise.

[gloomy spoke]

When i try and creat a video from a photo and it says not a valid value , what should I do exactly or its a glitch or what?

[echo aurora]

Thanks for sharing. For the captcha system I'm unable to share details about how it works. This system overall is something we're going to consistently pay close attention to and make changes to when appropriate.

I have been sharing feedback from the community who've been sharing frustration and confusion around this captcha system.

[echo aurora]

I'm not familiar with this error message, can you send a screenshot? What's the file type of the image you're trying to upload?

[gloomy spoke]

Of course

When i try to press on the image button up to upload it , the prompt button become red , and if i try to press on the prompt button the image button become red

[hollow flicker]

deepseek needs an upgrade like really badly

its kinda horrible

at coding

and stuff

[lucid cedar]

Hello, how to you my Spider-verse?

[echo aurora]

So after you start to type out /image-to-video can you click on the /slash command as it generates above the text bar. After, you should automatically be routed to your photo library where you can select an image. Are you seeing those steps?

[hushed gyro]

@echo aurora Loving the new UI!

[echo aurora]

Woo glad to hear it!! I'll let the team know.

Is still an experiment, hasn't yet rolled out fully.

[hushed gyro]

Well better than nothing, right? 🤭

Glad to see improvements!! @echo aurora

🥳

[pastel bramble]

I am afraid NOT to love it.
That's how scary it is to me 😲
Although, I am an old lady
So there's that 🫤

[gloomy spoke]

Yup exactly , it did work and creat a video in the video arena 2 i think it was a lag or something

[vocal mirage]

Why the video is not working? Can anyone help

Hello anyone?

?

[gleaming roost]

https://tenor.com/view/cat-drinda-gif-4209197159388134598

[echo aurora]

It doesn't look like you've prompted the bot. You'll want to type out /video in #video-arena-1 then you'll be able to prompt the bot. More info can be found in #1397655624103493813 . Let me know if you have questions.

[stray aspen]

oh my god finally

[river hamlet]

👍🏻👍🏻👍🏻

[stray aspen]

what models is it enabled for

seems like some models lost image upload because of this

[shrewd citrus]

Wow that’s good but how does this work with privacy

Like most pdf documents are confidential so will they stay confidential?

@echo aurora Hoping you could answer this 😸

[gleaming roost]

[river hamlet]

OMG, VISION

But what place?

[stray aspen]

cook

[echo aurora]

... is the site down for just me...

wait nvm, spoke too soon

[river hamlet]

Nice coook

[echo aurora]

Our privacy policy is still going to be implace even with pdf support. Before there are open data releases we'll still scrub for PII. These practices won't change.

[river hamlet]

❤️❤️

[echo aurora]

Do want to note this is still in experiment, so it isn't rolled out to everyone yet.

[river hamlet]

Finally!!!

Lmarena i love you

[echo aurora]

https://klipy.com/gifs/cappys-cappys-nft

[river hamlet]

Will there be claude thinking + vision?

And is there any data on how well Claude sees?

[echo aurora]

There are claude models on the Vision Arena leaderboard if that's what you're asking for.

[river hamlet]

I know :), but I didn’t find version 4.5 in the top

There are no 4.5 versions in the list

[grave plaza]

whats the rate limit for video models on website

[echo aurora]

Oh wait sorry I wasn't paying attention. Those models don't yet have image upload enabled, we do plan to make this change.

[echo aurora]

3 per 24/hr period.

[river hamlet]

Okey

Good night (ru)

[grave plaza]

wow. but image?

[hollow ivy]

which of them is best?

[keen lake]

both same but ultra can generate more output

[royal coral]

How is Claude max ?

Vs chat gpt pro & Gemini pro ?

[hollow flicker]

mr pineaple can you make GPT 5.2 Codex in the direct chat instead of having to use code arena

im to cheap to buy it

and i dont want to use code arena

PLS pineaple i need this my coding kinda sucks and i want to help it out

[raven otter]

my attempt that creating something that sounds like it came STRAIGHT from the early 2000s and late 90s, what do you all think?

[ember pond]

hey can i use veo in lmarena ?

[tiny halo]

hey guys is there any way to save my chat for a new chat im geting this message "Something went wrong with this response, please try again."

and i need to tell the new chat everything

[marsh hedge]

this frying me dawg

[tiny halo]

its annoying

[raven otter]

Do you like it?

[marsh hedge]

ain't bad but those hits are too loud

almost scared me

[raven otter]

I tried really

[marsh hedge]

i like the melody tho it's fire

and the drums

it's just the hits

[hollow ivy]

it's groovy, nice

but a bit monotonous

but i heard worse

[echo aurora]

I can flag to the team Btw it is in Direct Chat in Code Arena. I understand the want for it in Text Arena, but yeah wanted to clear that up incase there was come confusion.

[echo aurora]

Can you try the steps in this article ?

[tiny halo]

yes when i try to use new chat it works but i dont want to start saying to the new chat what i was doing maybe you can guys add an option to load the previous chat ? if this can be it will be a really helpful

[raven otter]

It’s supposed to be DVD MENU music

[ember pond]

my video is still generating is that normal for taking more than 10 minnutes ?

its 11 mins rn

[crude ivy]

just learning

[echo aurora]

wow

[hollow flicker]

YAY

thank you pineapoplke

[granite heron]

@gray garnet hello!, please before trying to generate content, check out this post that contains a detailed guide with instructions to properly use the Discord bot: #1397655624103493813 message

[quartz pike]

wtf is flux klein

dr klein hl2?

https://tenor.com/view/hope-nuke-truth-nuke-half-life-half-life-3-valve-gif-7775989612184395817

[obsidian shell]

it reminded me of re wesker for some reason

[quartz pike]

brb ima try klein 9b vs 4v

4b*

signature hand test

is a 4b image model can do this ill be shocked

holy sh#t. we have reached the point to where ais that can run on your pc can generate coherent images of hands

9b is even better

[hidden widget]

I laugh a lil bit when I saw that hand

[quartz pike]

yeah it is a bit goofy lol

but the fact that it can probably run on your pc just fine

and still generate that is insane to me

[hidden widget]

ye its wild

[quartz pike]

wait why was my message deleted

[hidden widget]

auto bot probly

[quartz pike]

this is what i originaly meant to say lol

[quartz pike]

lol

[echo aurora]

That looks like a connection issue with Discord

[quartz pike]

oh k

[toxic egret]

when was pdf support added?! :D

i can finally send gemini the arxiv papers i dont understand, thanks to whoever had the idea and the lmarena team

[willow kettle]

There used to be an option on the website to compare videos and vote; did anyone else have that option and it's no longer there?

[echo aurora]

Recently! Sort of. It's still in an experiment so it's not fully rolled out to everyone.

[odd merlin]

Hello What can I do if my model stays in a loop

Is there a button to stop it?

[echo aurora]

Sadly, there isn't a stop button, yet. It's on the list of things we'd like to build. If you haven't tried already -> clear your cache and hard refresh the site. If no luck, starting a new chat may be your best next option.

Although that's not ideal as you won't have that same context.

[odd merlin]

Okay, thank you very much

[maiden oak]

hello , i love this proyect

[floral aspen]

But it's not always available; sometimes it's there, sometimes it's not.

Why doesn't LMArena.ai support uploading all file types, only allowing images and recently PDF files?

[gleaming roost]

This is just another mystery among 99% of things that haven't been implemented in lmarena for years.

[floral aspen]

:)))))

[languid crescent]

NEW UPDATE IS POGGERS!!!! FINALLY CAN CHAT WITH PDFS!!! I LOVE LMARENA ❤️

[floral aspen]

So, the video generate button is gone now.

So, I already said it "it's not always available, sometimes it's there, sometimes it's not."

[languid crescent]

Some models don't support PDF chat, is it only applicable/available to specific models?

[floral aspen]

I think so

[warped cobalt]

generate a anime video.

[desert abyss]

Hello! Please check ⁠how-to-video-bot to learn how to generate videos.

[hard quiver]

Nanobana pro is now mega censored, is imposible to edit a photo of a person now

Thanks Elon

[obtuse smelt]

lol nice

[obtuse smelt]

but yeah nano banana is good

[limpid compass]

anyone having major problems with nano banana pro? i haven't been able to generate anything for the last 3 hours or so

[obtuse smelt]

ah me too

need fix

[plucky island]

do json prompts really make a difference in image generation?

[plucky island]

same

[iron cipher]

WAIT WHAT

[brazen sapphire]

[rose sky]

I made this... how is it?

If you want: https://rockyandthepawpatrol.itch.io/fx-studio-pro

[river hamlet]

@echo aurora have you disabled the ability to upload PDFs?

[languid crescent]

what happen to attaching pdf files ;-;

[hushed gyro]

yall really need to fix this

[latent jungle]

fr

[hushed gyro]

omg why tf is your platform so unstable 😭

[hushed gyro]

bro half an hour later

on one message!

[ashen mauve]

no this needs to be fixed

Stupid a/b testing sh+t pops up every single time, every 5 messages

[shrewd citrus]

one of them will be Gemini 3 pro

[ashen mauve]

idk maybe it is not loading anyways

[spare bolt]

why yall removed sending pdf files

[fossil socket]

Where the hell is the attach image for claude

I literally saw it a few hours ago

[ashen mauve]

this sucks now it's every other message and I'm getting security verified tf is going on with this website

[spare bolt]

its always like that not surprizing

[hushed gyro]

Idk why

[honest bobcat]

Why can't the AI be used now, it could be used this afternoon?

[fossil socket]

@echo aurora

[hushed gyro]

Openrouter is paid right? 😔

[viscid timber]

scam?

[nimble elk]

hi guys

[stray aspen]

These automatic battles in direct chat are actually annoying

[quartz pike]

imo flux klein is awesome

i love flux klein

i never thought there would be a day where a ai image gen model can run on some consumer-pc's let alone it be able to generate hands

[stray aspen]

Bro yupps server is actually north Korea

[quartz pike]

?

[stray aspen]

I got banned for giving my opinion on an Ai model

[quartz pike]

who is yupp

[stray aspen]

🥀

[loud verge]

Guys

Does gpt 5.2 search use the same search as api or a different one?

[late wolf]

why did the image generate AI stop remembering the chat? It's terribly inconvenient. Are there any solutions?

[echo hawk]

and?

[warm pike]

I want to try the amazing AI in generating text to image and image to video.

[hollow flicker]

pineapple uhhh I can use pdfs to upload bigger then the text limits so I can upload databases through the files and Gemini or Claude can output the fixed code thanks 🙏

[hardy swallow]

HI please check #1397655624103493813 to learn how to use the bot.

[maiden badger]

Hello

[dusky ravine]

Is the web laggin?

[hardy swallow]

Hi @vale canyon please check #1397655624103493813 to learn how to use the bot.

[tall scarab]

hi

[marsh hedge]

hi victory

[limpid compass]

nano banana pro still messed up? its still not working for me (since last night)

[edgy tundra]

hello

[celest orchid]

what the

[graceful finch]

So why is it when I choose direct chat, I chat with Gemini, Claude, etc, that at random intervals it does a "Battle" with 2 anonymous models, one being one I picked and another randomly picked? My brain isn't ready to compare 2 ai models randomly, sometimes when I am tired of reading 2 ai prompts and choosing between the two I choose direct chat to test models.

[graceful finch]

Yeah I tried Nano banana pro a couple of months ago to generate a futuristic utopian bedroom with a great view, it failed.

[limpid compass]

I meant I've been getting this over and over each hour since about 15 hours ago.

[graceful finch]

Yep. I got that message a few weeks ago using Nano Banana. Maybe report it as an issue? There's a bunch of other AI image models to use while it's being fixed. Been switching between flux and gpt-image.

[junior island]

👋

[prime beacon]

How to use this video generator