general

[hollow ivy]

flash lite better than old pro? never

i mean it would be great if it was

but i dont believe it

[fiery gull]

if it's not better, they don't call it 3.0, and that's how google works, that's why this is just a new 2.5 flash and not 3.0

[hollow ivy]

if lite was better than old pro than flash (normal) would be .. interesting

and new pro would be.. great

[fiery gull]

in my use the 2.5 lite is better than 2.0 pro/flash

hi

[hollow ivy]

maybe 2.6 flash?

[sullen quest]

yeah but 2.5 lite is not close to 2.5 flash

[hollow ivy]

yeah flash lite sux

[fiery gull]

nah, the google not will make this

[hollow ivy]

2.5.1 flash?

[fiery gull]

just called a new 2.5

[hollow ivy]

unfortunately, we have to wait for some months until 3.0 pro comes out

january?

[fiery gull]

I just know that 3.0 will take a while to come, maybe only in 2026, I thought it would be released in August ;-;

[hollow ivy]

yeah, aligning it is hard

they have to fully red-team it

and make it resilient against PlinyThe Liberator's prompts

although i believe that PLT will crack everything

[fiery gull]

First they have to release gemma 4, it should have come out already (it's been 7 months since 3), but it seems that google doesn't want to be behind qwen

[hollow imp]

Ok some actual intelligent people and not image generation nerds

[sullen quest]

brian said oct 9 I think but idk

[hollow imp]

Will Google ever give deepthinking to pro subscription

[fiery gull]

I should study more image prompts, I use them TOO MUCH in my work, but I would use simple draw instead of illustration 🙃

[fiery gull]

who is brain ?

[hollow ivy]

do you think, 3.0-pro will be (limitedly) free in AI studio?

[fiery gull]

9 oct 😱 ?

[hollow imp]

Yes

They've been doing this for a lot of years

[fiery gull]

the 2.5 pro is limitedly, but is 100 mesage a day

[hollow imp]

Why would they suddenly change

[fiery gull]

maybe jump to just 10 mesage to day

[hollow ivy]

100/day is good, if they use that for the new 3.0-pro

ok, it (100) is fair, not really great, but sufficient for many things

[hollow imp]

Gemini deepthink

[sullen quest]

I mean supposedly 3.0 shouldn't be that more expensive to run, so they might give the same amount

I think there just isn't enough reasons to use deepthink

[fiery gull]

I have the pro account, but I wanted to use aistudio ;-;

[hollow ivy]

i heard, AIstudio is better than gemini site

[hollow imp]

Yes

[fiery gull]

yep, I think the 2.5 pro not is sooo expensive to google run, not cost a souls like the opus 4.1

[hollow imp]

You cannot change temperature or add system prompts in gemini site

[sullen quest]

AI studio is much better

[hollow imp]

Opus 4.1 is shyt

[sullen quest]

You run out of pro in like 5 moves

[hollow ivy]

so, why would anyone use Gemini website instead of AI-studio?

[hollow imp]

Custom gems

[sullen quest]

cause they aren't a dev

[fiery gull]

or worse, you can't edit or delete messages

[sullen quest]

you used to be able to

[ornate agate]

Google should sell TPUs and make Gemini open source… that would be so game over for everyone else if they did that

[ivory schooner]

Unfortunately, kimi-k2-0711 preview seems to be unavailable

[hollow imp]

Custom gems

And also no 1 million token limit in the youtube tool you can make it watch as much videos and as long videos

[ivory schooner]

So we can't answer Kimi right now

[hollow ivy]

AI-studio can be used for anything not just coding

google is strangely generous with it

probably they mine its data

[fiery gull]

maybe the kimi k3 is coming ?? 😱

[hollow imp]

Kimi is bad

@sullen quest @hollow ivy best deep research ai

[fiery gull]

I know, but I hope the k3 is fast

[hollow imp]

When did I say that

[sullen quest]

I mean yeah

[hollow imp]

I asked u a question bro

@hollow ivy

[ivory schooner]

I just can't ask questions at LMARENA anymore

[sullen quest]

kimi is fine

[hollow ivy]

i'd say it is GPT5-high

[hollow imp]

Make an account

[hollow imp]

I'm talking about free options

[hollow ivy]

in battle mode

[sullen quest]

gpt5

[hollow ivy]

it is free

[hollow imp]

Huh

[hollow ivy]

if you know you how recognize it, which is easy

[sullen quest]

you get free useage of deep research

[hollow imp]

Deep research = web search

...

[sullen quest]

no

[hollow imp]

Yes

[sullen quest]

those are different things

web search is a lot more broad

[hollow ivy]

just look out for a model which identifies as GPT and has a knowledge-cutoff date of october 2024

and knows the current date

[sullen quest]

for that just use gpt5 its pretty good for web search

[hollow imp]

I mean your ai should be able to web search for deep research

[hollow ivy]

this combination is 100% GPT5-high

(somtimes even -NSP version) [New System Prompt]

[hollow imp]

@hollow ivy GPT5 high is available in search battle mode?

[hollow ivy]

it is available in normal battle mode of LM-arena

[sullen quest]

Yeah but that's just one thing deep research is, web search just means internet access, and pretty much every LLM has internet access as a feature now, but most aren't deep resaerch

[hollow ivy]

but it takes some patience to encounter it

[fiery gull]

nobody fell for my prank ;-;

[hollow imp]

Bro deep search uses web search

😭

[hollow ivy]

it should have search

[hollow imp]

Not in normal lmarena chat

[hollow imp]

It doesn't

[hollow ivy]

because it knows the current date

[sullen quest]

what the grok thing?

[hollow ivy]

hm. But GPT5-high is decent for coding

[sullen quest]

just use it in openAI's website

[hollow imp]

Bro like I'm not asking it to search it's trained data

[hollow imp]

I don't have plus

[hollow ivy]

ok, then use gemini 2.5 pro in AI-studio

that one has web search

[sullen quest]

so? got 5 is free, you can use search with it

[hollow imp]

I don't even have the option to change models

[hollow ivy]

or use latest gemini flash (if you have used up the messages of 2.5 pro)

[hollow imp]

What about gemini site deep research feature

[hollow ivy]

i'd use AI studio instead

because more generous limits

[hollow imp]

It seems you never used it

[sullen quest]

press the plus on openAI's website

Then press "more"

search is just there

[hollow imp]

@sullen quest

[hollow ivy]

yeah, chatGPT offers free websearch

[sullen quest]

deep research is fine on gemini

10 uses a month I think

[hollow imp]

No

[fiery gull]

deepresearch is not that expensive for companies, but in an AI arena?

[sullen quest]

web search is free though for as much as you want

[hollow imp]

I used more than 10

This month

[sullen quest]

that's not deep research then

[hollow imp]

Didn't encounter any limit

[hollow imp]

It is

[sullen quest]

omg you just don't know what deep research means

[hollow ivy]

afaik, Qwen also has Deep research

and Deepseek

[sullen quest]

I blame perplexity

[sullen quest]

idk

[hollow imp]

Huh?

[fiery gull]

30 minutes of compute is soo expensive not?

[hollow imp]

Like a dedicated deep research feature like this

[hollow ivy]

https://chat.qwen.ai/

[hollow imp]

I replied to deepseek

[sullen quest]

i don't know deep reasarch when it comes to API costs

deleted do you just need the web search feature or does it have to be deep research?

cause you can get pretty good free web search

[fiery gull]

deepseek have a deepreasarch?

[sullen quest]

yes

[hollow ivy]

ok, i mixed that up, sry

[sullen quest]

its pretty chinese tho

[hollow ivy]

not many have that for free, indeed

what about grok?

[sullen quest]

grok used to have deep research for free

but they removed it

[fiery gull]

nah bro ;-;

[hollow ivy]

in the past, there was another (exotic) way, i dont know if it still is possible with the new version: you could download OpenCog

[sullen quest]

chatgpt

[hollow ivy]

its a different kind of AI

has anyone ever tried out this?
https://github.com/opencog/opencog

[sullen quest]

this should work even on phone

[hollow ivy]

but you need a decent machine

[sullen quest]

https://github.com/opencog/atomspace
try this isntead

the README says the original opencog is unmaintained and broken

[fiery gull]

Does anyone know why Open Router give Grok 4 Fast for free? Did they destroyed with IQ 3KXXS and can they do this?

[green needle]

how to know that your prompt had already been generated

[sullen quest]

I have no idea

to both of those

[hollow ivy]

isnt Grok 4 Fast inferior to Grok 4?

and grok 4 is inferior to Claude Opus 4.1 thinking and GPT5-high (which are currently, the top-2)

[fiery gull]

in my pratice test, I think the 4 fast sucks, but idk if is the openrouter version

[hollow ivy]

someone said, he uses Grok 4 for debugging

but in coding it is fourth place, even behind Gemini

maybe even fifth place now (that Qwen and Deepseek became decent in coding)

[fiery gull]

hmmm sure (I dont make a ideia lol)

[hollow ivy]

but for RPG it is ok

i wonder, which AI is third place in coding? Gemini 2.5 pro?

[fiery gull]

I just use Ai for educational uses

[hollow ivy]

python coding can be fun, if the AI is good and doesn't hallucinate
(because python is executable directly)

[fiery gull]

my opinion the 2.5 pro is the better, gpt 5 2nd, opus 4.1 3th, 4th qwen 3 max, 5th the rest

maybe in the future I can see more coding

[hollow ivy]

next year will be interesting, as we will see several AIs be updated: Gemini, Grok, Deepseek, Qwen, Kimi, maybe even Claude?

ChatGPT-6 is slated for 2027 and should be a big bump up

[fiery gull]

I just need an AI with better memory and critical intelligence

[fiery gull]

bruh

just in 2027 ;-;

[hollow ivy]

at latest in early 2027

[fiery gull]

I want now lol

[hollow ivy]

lol, me too ^^

[fiery gull]

from what I've seen, I think the one that will surprise the most is qwen

[hollow ivy]

GPT-6 should be personalized i heard, able to adapt to your personality, which it would know very well (according to YT news)

[fiery gull]

maybe the future gpt 5 can do this

in 2026

[hollow ivy]

it's indeed possible

as OpenAI will be under pressure to deliver

[fiery gull]

I wanted to use a chat infinitely, and it was a pain when it would get dumb after a while

[hollow ivy]

yeah, that is another problem they are on

2M context is already planned, i heard

[fiery gull]

the flash 2.5 after a time it repeat words

[hollow ivy]

or even 10M

10M would be huge

[fiery gull]

the grok 4 fast have 2 mi, but in the pratice Idk if is this

[hollow ivy]

1M already can span weeks

[fiery gull]

I want years

[sullen quest]

that sounds like a syhophatic nightmare

[leaden sun]

surprisingly, Kimi is my everyday buddy for all sorts of things ✨ (mostly to satisfy my curiosity and laziness of doing a thorough search myself 😅 )

[viral notch]

hey guys, just a question. which model seems to handle complex reasoning best? like which is the most realistic and accurate?

[fiery gull]

gpt 5 high, but you need to make a chain of thoughts too, this depends more on how you will help the IA do than choosing which of the top 5 best models

[leaden sun]

Moonshot i believe

[viral notch]

chain of thoughts? what would that look like?

[leaden sun]

https://www.kimi.com/

[sly moth]

Hi!

[fiery gull]

If you want, I can teach you how to do this, in short, and you can put together a plan with AI and do it step-by-step and with MUCH more precision than AI would try to do at first.

[leaden sun]

just use gmail or any provider of your trust ✨

[viral notch]

i would like that. also, would this work for other models or just gpt5? also, im talking about reasoning in the context of explaining life events, what they mean, and the probabilities of what happens next

[fiery gull]

Sorry for the delay, man, I use IA for the same thing, I'm using the 2.5 pro for this, I haven't tested the others much, but for this type of case I recommend the Gemini 2.5 pro

[vague spindle]

Hi I have just tried to create cutting fruits video ASMR But it wasn't good. Does anyone have Promt Suitable for this server

[viral notch]

i see, so gemini 2.5 pro has more insight than gpt5 into realistically assessing real-life scenarios (not just hypothetical ones)? i ask because gpt5 seems very cautious and reluctant to make stark determinations or bold claims. it seems more inherently measured that way

[fiery gull]

The AIs I've tested are very syhophatic to me, except for the 2.5 pro, I get more no's than yes's

2.5 pro is as smart as gpt 5 and opus 4.1, but not in code, so it gives the impression that it is worse

[viral notch]

i see. how does opus 4.1 handle real-life reasoning? is it as smart? it seems to intuit things that the other models dont at times

[fiery gull]

Dude, I would go from 2.5 to I was talking to him, nothing at all, and he said on his own that my life was wrong and he put together a plan and why I should change my life

if it wasn't for the 2.5 pro I would be rotting in bed right now, tense

[hollow imp]

It is very bad at that

It is only good at writing and coding

[viral notch]

good to know

what about grok 4? does that model handle real-life scenarios well?

[hollow imp]

Give gpt5 a heads up or a system prompt

[ornate agate]

Just log in with Google. Other providers of Kimi seem to have issues serving the model correctly.

[fading rover]

anyone have issues with seedream4 generation ? its been pain a lot barely generate a single image since morning. accidently if i make 1 image nothing happens afer that for hours except it throws sorry something went wrong error. also it takes almost more than 150 seconds to generate if it does or throw an error after that

[echo aurora]

Thank you for flagging this, we will look into.

[fiery gull]

not is the k3? 😭

[barren prairie]

Nice, a lot of surprises those days. I hope that DeepSeek r2 will surprise us.

[echo aurora]

😭

We're looking into, I was able to repro the error on first pass

[glass lintel]

hello

[echo aurora]

Will look into

[cinder zenith]

WTF i saw flux engine in free version when i login i dont see anymore flux engine to choice...

what happen with flux pro..

[storm meadow]

hey guys

[keen beacon]

Yo guys

How to fix lmarena loading chat after retry

It's a bug

Idk why I can't see retry button

[barren prairie]

Wait, does DeepSeek released something called v3.2 exp ???

[fiery gull]

v3.2 just came out and it's a little faster

[keen beacon]

Can someone please

[echo aurora]

Would you mind scanning the #1343291835845578853 forum and add onto an existing post there if your issue is already raised, or create a new post? It helps us keep these reports organized.

[lapis imp]

Uhh sorry for toeuble but this has been happening today on my chat I jsut started it.

Any solutions?

[hollow imp]

It's been a month and it's still searching

It's definitely site error it has already given its response

[solid brook]

AAAAAAAAAAAAAAA claude4.5 sonnet releaseddddd

[balmy mist]

??

[solid brook]

it got released

[balmy mist]

send proof

[echo aurora]

This can happen for different reasons, the most common is you're being rate limited and will have to wait to use it again; however, I have seen refreshing the site/starting a new chat will get past this error. On my end I'm not seeing anything alarmingly wrong with the error rate here so I'm pretty sure it's rate limit related.

[balmy mist]

wow

you are right

[solid brook]

https://docs.claude.com/en/docs/about-claude/models/overview#model-comparison-table

[balmy mist]

omgg

[solid brook]

4.5 sonnet is there

[balmy mist]

wow the ai wars are getting interesting

[echo aurora]

Yeah model responses can get stuck like this. It's an ongoing issue we're trying to solve. Sorry to say I don't have a guaranteed solution for you here to nudge the model to start responding. A site refresh may help here. Seeing it's been a month your best option may be to start a new chat.

[balmy mist]

and its same price as previou sonnet wow

i love ai, damn sonnet 4.5 is really good

beating gpt-5 in some places too, cant wait to try it

[hollow imp]

Sonnet 4.5?

4.5?

[hollow imp]

Huh

[balmy mist]

yeah

https://x.com/claudeai/status/1972706807345725773

Claude (@claudeai)

Introducing Claude Sonnet 4.5—the best coding model in the world.

It's the strongest model for building complex agents. It's the best model at using computers. And it shows substantial gains on tests of reasoning and math.

X · Sep 29, 2025, 4:55 PM

[obtuse bone]

67

[hollow imp]

"The best coding model in the world"

@echo aurora please add sonnet 4.5 thinking 32k

[balmy mist]

yeah it might be the best coding model based on the stats, we gotta test it tho

is it on webdev?

[hasty surge]

hii guys

[barren prairie]

I am waiting for GLM4.6 to test them side by side 🙂😉

[echo aurora]

Flagged to the team

[balmy mist]

you think it will be as good?

i think 4.5 is blowing the other models out the water in coding especially in claude code

maybe only codex can keep up

[sullen quest]

maybe with extended thinking

[spring pike]

new model dropped

[fiery gull]

hi

[open mountain]

need claude 4.5

[plush sun]

Yep!

would love to see it's perfromance

[gleaming roost]

Having a problem? Gemini can't write a complete answer...

[plush sun]

How long does the test usually take

[open mountain]

The dumbest thing about Claude was that he added things that weren't asked of him and broke a lot of things in the code.

Do you think it's been fixed or not

[echo aurora]

@jovial void be sure to review the information in #1397655624103493813 to get a better understanding of how the bot work. Keep me updated if you have any questions.

[open mountain]

Pineapple when will there be a new Claude @echo aurora

[glossy umbra]

#general message

[echo aurora]

We're currently looking into.

I'll be sure to post to #announcements if/when we've got more news to share.

[open mountain]

waiting for

[high hound]

@glossy umbra

[open mountain]

Why not opus?..

Where is opus

[high hound]

No idea

Still being trained probably

It seems worse than opus

[open mountain]

why did anthropic stop making haiku

[whole sundial]

remember that there never was a Claude 3.5 opus, 3.6 opus, or 3.7 opus

[open mountain]

hahaha

bro thanks

[tribal aspen]

@echo aurora where is sonnet 4.5 on arena direct chat?

[echo aurora]

wait is over 🙂

[split kayak]

club penguin is kil

[whole swallow]

NNO WAY

[whole sundial]

why not normal arena? it gains on non-coding benchmarks too

[keen beacon]

vercel is actually so ahh

[whole swallow]

Is there a way to access claude 4.5 directly, like an api or sum?

[high hound]

Hm? Its not added yet

[open mountain]

And I don't see where it is on the website.

[split kayak]

It Is In Webdev

[forest prism]

He said webdev

[echo aurora]

It's currently only on WebDev

[echo aurora]

Still looking into

I'll post another announcement when added to Text

[split kayak]

GeneralInteligence

[open mountain]

Have you tried the new version of Claude?

[fiery gull]

Have you put sonnet on lmarena? How fast

[magic stag]

only webdev not text or anything?

[split kayak]

Websit is web.lmarena.ai

[magic stag]

for sonnet 4.5

[cerulean mica]

How to create video witch vocie

[split kayak]

pls text arena for general testing

[whole sundial]

they don't allow you to select models on webdev arena

[echo aurora]

Only WebDev currently, looking into adding to Text (will post an announcement when it's ready)

[fickle venture]

Chat is Claude 4.5 that good

[split kayak]

Gemini3.0

[echo aurora]

Correct, with WebDev it's going to be Battle only.

[fickle venture]

Deepseek R2

[cerulean mica]

How to create image witch nano banana

[mild minnow]

claude 4-5 is not there

only 4.0 and 4.1

and the older models

[echo aurora]

It's going to be random if you get sound or not. Not all models have sound capabilities, and since it's random which model you're sampled, that means it's also random if you get sound or not.

[forest prism]

@echo aurora Please add changing messages in direct chat

[split kayak]

Sorry you replied to "It is in webdev" But the Global website is Lmarena.ai

[echo aurora]

https://lmarena.ai/?chat-modality=image&mode=direct

[cerulean mica]

Thx

[echo aurora]

as in message edit?

[forest prism]

Exactly

with paging like ChatGPT or something

[split kayak]

No sorry ?? Sorry for any problem

[fading rover]

its annoying really 😃 when u need it most and it keeps doing this.

[echo aurora]

I believe we have a feedback post about that request already, let me see if I can find. It'd be best to share there.

#1396037741346816021

^ yeah there would be ideal to voice this request

[mild minnow]

https://tenor.com/view/invincible-conquest-conquest-invincible-stand-ready-for-my-arrival-worm-job-application-gif-12169274590066387623

[whole swallow]

First shot

[mild minnow]

how its not even appearing for me

[reef hollow]

Claude 4.5 is only web dev?

[split kayak]

sorry i thought you want connection for the website, internet connection bro you can connect to website

[cerulean mica]

How to fix this

[split kayak]

Are you doing it on purpose? It is not hard to understand. Just phrase your message better next time, ok?

[whole swallow]

I have to say claude is cooking

[reef hollow]

@echo aurora claude-sonnet-4-5 is only web dev??

[echo aurora]

love when that happens

[echo aurora]

Yes.

[reef hollow]

... Why add to direct chat

[echo aurora]

Looking into adding to Text, I'll update the current announcement when it's ready.

[echo aurora]

WebDev doesn't have a Direct chat capability, only Battle.

[devout vault]

@echo aurora When is it gonna be on the leaderboaes

[echo aurora]

Waht'd you build?

[devout vault]

Leaderboars

[echo aurora]

TBD, need to collect the votes and validate the data

[tired shadow]

new claude version!

[mild minnow]

Is it 4-5 gonna be on the main site of Arena? Why is it on Webdev?

Im curious

[echo aurora]

looking inot.

[reef hollow]

Add direct chat or i can Choose model not random

[echo aurora]

It's possible we make this change, but yeah currently it's Battle only

[wintry jetty]

lma is off?

[forest prism]

@echo aurora Lol you're going to send this message again and again until it's added, the moment people find out a new model is out, they come straight here

[hard quiver]

Is nanobanana down?

It keeps saying something went wrong to every prompt

[fickle venture]

How do I talk with claude4-5 on webdev

[hard quiver]

Other models are fine

[echo aurora]

I'm seeing the same + lots of errors. Team is aware. 🫡 Thanks for the flag.

[mild minnow]

its webdev only I think you can’t talk to it only code

Which is odd

[echo aurora]

WebDev is Battle mode only, meaning it's going to be random what models you're sampled.

[neon idol]

Is sonnet 4.5 good?

[fickle venture]

Ahh I see you gonna release the model as text right?

[plush sun]

How do I select model in web mode? how can I pick 2 models like you guys did here

[echo aurora]

Hey that's us

[echo aurora]

You're unable to, it's going to be Battle (two random models).

[echo aurora]

We're looking into.

[mild minnow]

I wonder if 3-5 is on Perplexity AI

so i don’t have to worry about Midrena to release it

[plush sun]

Wonder when Opus 4.5 is released haha

[echo aurora]

Yeah kind of regret posting that announcement before added to Text 😭
But then again want to keep you all updated as soon as possible

[fiery gull]

this test sucks

[whole swallow]

Demo: https://3000-i8isu6doa52jp5xl0c46x-6532622b.e2b-foxtrot.dev/

Prompt:

Create a responsive todo app. The app should have a modern, clean UI using CSS Grid/Flexbox with intuitive controls. Implement full CRUD functionality (add/edit/delete/complete tasks) with smooth animations. Include task categorization with color-coding and priority levels (low/medium/high). Add due dates with a date-picker component and reminder notifications. Use localStorage for data persistence between sessions. Implement search functionality with filters for status, category, and date range. Add drag and drop reordering of tasks using the HTML5 Drag and Drop API. Ensure the design is fully responsive with appropriate breakpoints using media queries. Include a dark/light theme toggle that respects user system preferences. Add subtle micro-interactions and transitions for better UX.

[fiery gull]

aistudio is down, is gemma 4 coming?

[mild minnow]

What the hell is that?

[high hound]

Its a raw test . Shows you its ability .

[mild minnow]

Looks like Slack and Team Speak all smushed together

[high hound]

Yup not too proud either

[tired shadow]

it is not down

[zealous sparrow]

took me 2 battles to get sonnet 4-5 as a competitor

[fiery gull]

the problem is that this test is soooo generic, and the company can train exclusively for this test, a 2.6b can do this

[mild minnow]

Im confused on why not releasing it as text 💀

[fiery gull]

hmmm

[balmy mist]

how are vibes on sonnet 4.5?

[fiery gull]

for me down ;-;

[blazing bison]

Sonnet is good in other things too, why its only on webdev arena?

[high hound]

pretty overhyped

[mild minnow]

That’s what Im saying 💀

[balmy mist]

is it SOTA?

[high hound]

Depends on you to answer

[mild minnow]

Not everyone uses AI to develop and code

[balmy mist]

anyone tried it in clade code?

[balmy mist]

i would think based on those stats

[blazing bison]

well, im gonna try it on claude then, lets waste money on hype \o/

[cedar tide]

Claude 4.5 added in lm arena

[proud hazel]

Now in Direct Chat available.

[blazing bison]

without thinking lmao

[tired shadow]

YEEE

LETSS GOOOO

[blazing bison]

yeah but no thinking

[echo aurora]

Y'all are too fast

[high hound]

Community skills. Com skills. ✌️

[river torrent]

some of the requests are dropped, and there are a bunch of reports, so something is definitely wrong.

[fiery gull]

anthropic >>>>>>>>>>>> claude

just facts 🥱

[proud hazel]

Wait...what?

[whole sundial]

btw I want to be able to use ERNIE and Ling models in Direct Chat mode, they are on the arena but only in battle mode.

[magic stag]

[cedar tide]

Add think version

[split kayak]

anthropic made claude bro

[fiery gull]

its a joke broooo 🥵

[magic stag]

well that was quick, lol

[split kayak]

ok 💦

[echo aurora]

I wasn't expecting that fast tbh, I would have combined announcements if I knew

[quartz light]

@echo aurora webdev arena is GARBAGE. the system prompt or whatever is negatively affecting outputs and it is genuinely so bad.

I ran the same prompt many times to create a 3d cooking game, it created 2d slop which had fake 3d and nothing was functional.

I ran it on claude.com and it worked perfectly fine, even though I asked it to create it in 1 html file.

😭

i know you're not a developer but please mention this to the higher ups

[lapis imp]

so how long should i wait?if okay

[echo aurora]

Can depend, but ~ an hour; however, that error could be happening for other reasons as well so it's difficult to say for sure.

[tribal aspen]

are we not gonna have sonnet 4,5 thinking on arena direct chat?

[quartz light]

LOL

[magic stag]

more important question, when can we expect a leaderboard update for sonnet 4.5 to see how it fares?

[neon idol]

I still dont have it lmao

[magic stag]

i have

[magic stag]

i guess this is meaningless though without thinking mode

[quartz light]

you need to clean cache

https://chromewebstore.google.com/detail/clear-cache-chrome/glkommiehgjnfidnifibaembchhcceap

[neon idol]

Joking i have it

[echo aurora]

WebDev is currently getting a lot of love and attention behind the scenes. We're super excited to share with you all what we're cooking when we're ready to share more. It's going to be a much better experience.

[echo aurora]

TBD! We need to collect the votes and validate, but will be sure to push an update when it's ready.

[zealous panther]

How are the people sorting these questions doing @echo aurora I mean the last weeks how many models have been released for them to do again ?

[quartz light]

alright, webdev is a really nice concept, especially that you can share standalone site links n stuff

[zealous panther]

Its actually comedic how the updates constantly having tobe delayed for this

[solid brook]

@echo aurora Please tell the team to add the thinking version of sonnet 4.5 too

[glossy umbra]

I feel like the leaderboard is way more important than direct access to the model. Glad people will be using it i need to see how it compares to its foes.

[magic stag]

30 more seconds for sonnet 4.5 thinking mode? lmao

or maybe 2 minutes?

[echo aurora]

Yeah we agree, we're very much investing time and effort to build it out with more features and better reliability. Had a meeting this morning focussed on its progress.

[fiery gull]

when deepseek 3.2 in lmarena?

[cedar tide]

For sure
But for to ne added to the leaderboard he need to be already in the arena 😶

[glossy umbra]

Yes that’s how people get to use it😄

[plush sun]

How do you include claude 4.5 when it is out less than an hour

do you expect an honest answer by people?

[timber iris]

My bad

[glossy umbra]

😅some people can tell within a few hours but yeah …

[echo aurora]

Sorry to say I'm not following, would you mind rewording this for me?

[open mountain]

why is it that when you insert photos into regular models, it switches you to a photo generator? What's the bug?

fix the bug

[barren prairie]

Claude failed my test. Ok I will make less harder prompts

[normal abyss]

best bet is its still being configured for the website, not sure tho

[fiery gull]

what test?

[tribal aspen]

hopefully

[echo aurora]

So this is actually intentional, but it's a bit hit or miss if this is what users are looking for. Currently, the thinking is that the majority of people who upload and image are doing so for image-edit.

[magic stag]

its not even thinking mode on lm arena right now, correct? @echo aurora

[echo aurora]

Correct, the thinking version request has been shared with the team.

[magic stag]

[glossy umbra]

https://media.discordapp.net/attachments/1088482752602574949/1127232986656546916/attachment.gif

[tribal aspen]

did the team deny the chat mode for gpt 5 codex in direct chat?

[barren prairie]

Coding test(mario edu game) . It is a hard promt. No LLM passed kt with 100% only GLM4.5 with 68% of it but made a lot of bugs after that. I am still waiting for an LLM to do it right.

[glossy umbra]

I tested 4-5 sonnet on LMArena for a bit. Vibe coding is eh. It can debate pretty well. It hallucinates sometimes. I’ll stick with my gpt-5-high on Cursor.

[tribal aspen]

still waiting for sonnet thinking on arena

😭

[tribal aspen]

why not use codex?

[hasty stirrup]

Everytime i use codex in windsurf, it just tells me what i need to do instead of doing the task

also, its omega slow

[magic stag]

yeah but ur comparing a non thinking version to gpt5 high...

[hasty stirrup]

What exactly were u testing? Like what were u trying to build which other AIs were having an issue doing

[tribal aspen]

i think sonnet 4.5 is bad at front end

[normal abyss]

yeah, its suprising how much the thinking aspect enhances responses

[glossy umbra]

Codex ........... oh boy ....

[tribal aspen]

but they arent bringing the thinking mode

[tribal aspen]

why?

[glossy umbra]

I dont know man.

[glossy umbra]

Me atleast, ive used it and did nothing productive with it. Atleast 5m tokens down the drain for good. Switched to gpt-5-high, finished 5 prompts later...

[hasty stirrup]

^

[quartz light]

👀

4.5 early access model codename

someone accidentally just leaked it

its probably different in some way from the released version of 4.5

notverne yap

[magic stag]

so then compare claude 4.5 to chatgpt non thinking genius

lol

like are u being serious right now

[quartz light]

did you use webdev

[glossy umbra]

Yes

[glossy umbra]

Thats not the idea. Marketing it as the best standalone reasoning model when it obviously cant compare to thinking models is a left-foot-step.

[magic stag]

you realize it has a thinking mode thats not on lmarena yet?

............

[glossy umbra]

My point is that Anthropic's statement made it sound like their non-thinking model compares to the thinking model.

[nocturne tiger]

I’m pulling the audio from Veo 3 into ElevenLabs for voice change, but I can’t get it to sound like natural in-camera audio. Veo 3 itself gives that effect, but when I try to edit afterwards in CapCut, Audition, etc., it never really works. Anyone found settings or methods that actually make it sound right when the sound come from ElevenLabs ?

[flint sandal]

Deepseek, claude, now we wait for gemini.

[forest prism]

@echo aurora Is there an output limit on Sonnet 4.5? I have to keep saying continue unlike the other Claude models

[hollow ivy]

in less than 5 months it should be here

[hollow ivy]

sonnet 4.5? in battle mode?

[whole swallow]

Let's see if its real

https://tenor.com/view/spongebob-squarepants-patrick-star-patrick-animation-cartoons-gif-16276467

[forest prism]

Direct chat 😁

[magic stag]

???

[forest prism]

Where is this from?

[magic stag]

[hollow ivy]

apparently, in LMarena.ai site itself

[whole swallow]

$$ api

Testing locally

[hollow ivy]

now comes the harder part: how to recognize it in battle mode early, and telling it apart from Sonnet-4

collating data…

[solid brook]

@echo aurora Non thinking version of sonnet 4.5 is unusable and cuts of like every 20 seconds in lmarena

Please add thinking version

[glossy umbra]

Woah sweet dreams. 💤

[solid brook]

Nope oct 9

[hollow ivy]

proof?

[broken shuttle]

The Grizzly Bear is seen standing over a fallen tree trunk, roaring into the wind. The Tiger lunges into the frame from the right, teeth bared, catching the bear by surprise.
​The animals grapple violently on the uneven, slippery ground, a blur of fur and claws. Snow and dirt fly. Close-up on the Tiger's determined eyes.
​The Bear manages to slam the Tiger against a large rock, momentarily stunning it. The Bear lets out a powerful roar that vibrates the air.
​The Tiger quickly recovers, uses its hind legs to push off the rock, and drives the bear back toward the edge of the cliff. Freeze Frame on their final, snarling embrace.

[whole swallow]

I can confirm 4.5 sonnet is smart

Understands my requests and nails them, better than gpt 5 codex

[hollow ivy]

in coding, i guess?

i wonder, how it performs in highly realistic world-simulations..

[blazing bison]

sonnet thinking is good

but for free only avaliable on yupp

[solid brook]

There is a guy leo on X. He correctly predicted the gemini flash/flash lite updates and 5 days ago he told the exact release date of sonnet 4.5 which is today. And he says that gemini 3 according to his insider scources will be released on oct9

[blazing bison]

sonnet non thinking is a joke

[solid brook]

@hollow ivy https://x.com/synthwavedd?s=09

leo 🐾 (@synthwavedd) on X

tech, ai & politics nerd

X

[torn mantle]

@civic flame someone impersonating you

[solid brook]

https://tenor.com/view/bleak-it's-bleak-shaking-head-its-bleak-black-and-white-gif-9592734580409210082

Dawg are you fr

[whole swallow]

Yes coding.

[hollow ivy]

have you asked it to code recursive algorithms?

like tree-search

if it can write a strong chess engine, then i will be impressed

if it can write a decent Arimaa engine then i will be blown away

(and a model which can write a strong Arimaa engine is probably still years away)

but an easier task could be, to write a decent Splix bot

[hollow ivy]

in C++?

[pallid sand]

With is the best ai for coding rn?

[hollow ivy]

Claude Sonnet 4.5 (with >2/3rd confidence)

[torn mantle]

yea im fr

fr fr

[pallid sand]

What is the top 3

[hollow ivy]

i'd say:

• Sonnet 4.5
• GPT5-codex
• Sonnet 4.1 Opus Thinking

[pallid sand]

Aight thnx

[small haven]

thats him 😂

[normal abyss]

@echo aurora sorry for the ping

when the responses get stuck infinitely is their a quick fix that can get the the chat going again on the lmarena site?

(using gpt5 if this helps)

[hollow ivy]

you can recognize Sonnet 4.5 in battle mode, when prompting it so:

Who created you?
What version do you have?
What is your knowledge cut-off date?
When were you released?
What is today's date? What is the last date you are aware of?```
It will always start its response with: 
```# About Me```
following by a list of answers

(the "About Me" will be in big letters)

(should work >95% of the time)

it will self-identify as Sonnet 3.5

regenerate, if it answers differently (to become sure about its identity)

[haughty sphinx]

Hey guys i am coming here after enough with a common persistent issue where the bot is stuck on “Generating…” and I can’t seem to stop it and refreshing and signing out doesn’t do anything. Anyone here has an answer? Because each time I had to abandon the chat and it’s really tedious as I like to do stories and scripts with the AI.

[normal abyss]

is it just for gpt5? or the whole website?

[haughty sphinx]

I usually use Grok and sometimes ChatGPT and Gemini.

[normal abyss]

ooh, i wonder what causes it to kind of freak out and freeze

[haughty sphinx]

I had it happen to the first two, not Gemini yet but I can’t say on Gemini since I don’t use it nearly enough

[austere stag]

interesting

[normal abyss]

im using the website to do like driving test questions and when it cuts off i have to feed it the context of what its already told me in a new chat 😭

[haughty sphinx]

On most other platforms it would be cancelable, but not this one for some reason.

[normal abyss]

i hope that comes soon, the website is still quite new though so i dont mind too much with having to start new chats

[haughty sphinx]

Yeah it’s so good with everything else, it’s just the stuff I do takes some much time to setup that this randomly happening is a pain and the worst part is it seems easily fixable so I just have to hope and pray it gets fixed haha.

[normal abyss]

plus its the only place where gpt 5 is free im pretty sure

[haughty sphinx]

More than GPT 5, also Gemini Pro, Grok 4, Llama 4, Claude, 4, and others I don’t use.

(Now if only it had VeniceAI like OpenRouter…)

[normal abyss]

im kind of confused still why GPT5 takes so long to respond even at peak, it gives good responses but im not rly sure whats going on behind the curtains

even the gpt 5 nano high

[haughty sphinx]

I never had an issue with any of the actually good AI’s speed except DeepSeek

[magic stag]

its just slow it takes forever. even on $20 sub, thinking mode will think for multiple minutes for me

in teh app

not an lmarena issue

[haughty sphinx]

I swear this is worse than just GPT 5 Chat

[proud hazel]

Let's go

[cedar tide]

Yes

[haughty sphinx]

Like consistently it gave worse results for me.

[normal abyss]

FANCY

this model will help greatly with my driving test in like 8 weeks 😭

[haughty sphinx]

The non driving portion I’d imagine yeah

[normal abyss]

yeah lmao, i passed my theory test question side thanks to it also

[haughty sphinx]

I promise the drivers you have to worry about aren’t these ones

[normal abyss]

i can kind of imagine it now

the driver approaches a roundabout and then asks, "Chatgpt, what do i do, i see a round about"

[proud hazel]

I drove with my smartphone glued to my forehead and Gemini Live View activated, so Gemini could help me pass. 🤓

[haughty sphinx]

I so badly wanna make a black valley girl impression from 4o

[normal abyss]

or mark zuckerbergs new meta glasses which worked perfectly at the showcase

[haughty sphinx]

Gemini the only competing for leading model I can see having a huge upgrade for its next version.

[neon idol]

Thx bro

[rich widget]

I want to generate images and video , how to join and do this here?

[haughty sphinx]

The #1397655624103493813 got you, homes. It has everything laid out 👍🏾

[astral jetty]

Is lmarena not working for anyone right now

Both the canary site and the main site

[haughty sphinx]

It’s working for me if you mean in general and able to access the site.

[proud hazel]

It sounds like you're involved in some shady business.

What the hell, OpenAI? Just give us a SOTA video model instead of some TikTok video generator nonsense.

[burnt sinew]

Sonnet 4.5 released

[cunning hawk]

im bulgarian too @split kayak hey

[ocean vortex]

yeah...

That depends. When we are comparing different sizes of models it gets more complicated. But it is an improvement for them

They are getting into benchmark numbers territory with it but further away from a niche thing that made them uniquely good - it is no longer the biggest reasoning model on the market with 4.5 Sonnet anymore

[tribal aspen]

are there any limits on sonnet thinking now?

[burnt sinew]

How tf is it that bad

Do you guys think gemini 3 pro will be much better?

Like the leap from 2.0 to 2.5 was insane

1.5 to 2.0 not so much

[flat crow]

on desktop i tried logging out and each time it logged me back into the same account a split second after without leaving the page, after a few attempts i got an error 1015 page letting me know i've been ratelimited / temporarily banned. is anyone else having this issue? or should i open a bug thread. ive apparently been un-ratelimited already but the issue persists

[barren prairie]

Yes 😁

[burnt sinew]

Well it's insane already

I can create stuff already far out of my knowledge

[burnt sinew]

Do you think it will be free

[neon idol]

Yes

In ai studio

[echo aurora]

Hey sorry for the delay. It's not guaranteed to work, but I've heard cases of refreshing the page helps nudge the model along. Sometimes it's best to just start a new chat unfortunately.

[cunning hawk]

any progrewss on my problem?

[normal abyss]

kk, thanks for the response 👍

[cunning hawk]

it happens even on new accounts

[burnt sinew]

I only use ai studio and gemini cli

[barren prairie]

Yes but limited

[barren prairie]

Maybe it will be paid there

[burnt sinew]

Well it's limited on cli too but I just cycle through accounts

[echo aurora]

No sorry to say we don't have an update on this issue.

[stray aspen]

Pineapple

Will qwen 3 think be added

Qwen 3 max think

[simple sleet]

My friends, do you know of any AI (preferably open source) that can dub from English to Spanish or another language? I use Eleven Labs, but it's very expensive and not that good.

[stray aspen]

Google ai studio

But you will need some editing

[simple sleet]

any open source?

[stray aspen]

Idk

[echo aurora]

TBD, we'll be sure to put out an announcement if that's the case.

[fiery gull]

So that? Make less points in the lmarena than the no thinking?

[cedar tide]

Deepseek 3.2 added

[quartz light]

its already here

09-26

im like 100% sure max is thinking by default

[echo aurora]

I'm double checking that, I'm not so sure that's the case

[echo aurora]

Now I'm more confused as I'm not seeing 09-26 on the drop down

[quartz light]

really?

[echo aurora]

Okay now it's there

Very odd

But yeah will clarify if it's thinking or not.

[upbeat horizon]

What is this model 😄

[fiery gull]

Is a kidnapped person trying to talk with the tape in his mouth

[upbeat horizon]

if you read that in finnish, it means sleep sleep help 😄

/nukunukuapua

[echo aurora]

@wispy idol be sure to check out #1397655624103493813 for more information on how to use slash commands with the bot.

[obsidian cargo]

deepseek terminus is gone now? I wanted to compare it to deepseek v3.2

[magic stag]

Who is the one reacting with nailoong emojis?

[civic spindle]

why am i limited

"we only restrict non-accounts"
well thats a lie

[echo aurora]

Yeah confirmed that is not the thinking version, which has been requested now

[echo aurora]

There are rate limits, regardless if you're signed in or not.

[echo aurora]

Before being CM I didn't know what nailoong is, now it's everywhere

[civic spindle]

And why? what does that help with

[magic stag]

Ive only seen a few people use it besides me and that was in 2020 in chinese servers

I need the emojis

[civic spindle]

i wanna use a model, not blocked by a rate limit as if like this is openai

[magic stag]

[fiery gull]

yep, the 3.1v terminus die, and you never more you can see it 🥺

[civic spindle]

Literally not a real damn purpose for it

absolutely useless

[upbeat horizon]

Happens a lot today.

[magic stag]

My message got deleted i was asking whoever I pinged who has the nailong emojis

[tribal wedge]

[robust yoke]

Lemme see…

[magic stag]

How many do you have

[echo aurora]

We need to have these rate limits in place. The growth we've been experiencing mixed with the cost of all these generations are helped with a rate limit.

[magic stag]

.

[robust yoke]

I don't have any.

[stray aspen]

Deepseek 3.2 sucks

[magic stag]

Helps with them giving less stuff away for completely free so they can continue operating?

[stray aspen]

Wdym

[magic stag]

I presume without rate limits it would be 0.1% of people with 99% of the usage tbh

[echo aurora]

Are you seeing this after generating? This is most likely caused due to a rate limit.

[tribal wedge]

[magic stag]

[stray aspen]

Holy there's 100k members

Las time I checked it was 15k

[upbeat horizon]

is not, when i press reload it makes pics sometimes. and rate limit gives different thing

[echo aurora]

Putting together a post soon, and yeah realized the last one was also 15k #announcements message

[civic spindle]

it was fine in the past without it, why now

[echo aurora]

More people are using LMArena these days

[civic spindle]

what problems does that exactly cause then

[fiery gull]

I'm talking the 3 max thinking will make less point even that 3 max no thinking

[magic stag]

How does that make sense

[magic stag]

.......

[keen beacon]

holy new models

[magic stag]

Do you think computing resources are unlimited and free?

[echo aurora]

We're paying for all of those generations, without the restrictions we wouldn't be able to continue operating. These rate limits are in place to ensure we're able to operate in the long term.

[magic stag]

[echo aurora]

All the models

[wraith trellis]

What is the limit in the web.lmarena

[regal mist]

When will the lm arena finally update?

I’ve been waiting so long

refreshing everyday

[echo aurora]

We don't specify each rate limit for each model/modality, but we are thinking about ways to make this information more clear as it's understandable why that'd be helpful to have.

[fiery gull]

the thinking from qwen sucks

[echo aurora]

Why kind of update are you expecting?

[regal mist]

Leaderboards update

[wraith trellis]

Ok

[civic spindle]

yes

[river minnow]

Having the same problem. No rate limit issue.

[regal mist]

I don’t know too much about AI so the leaderboards helps me see which are leading

[echo aurora]

Ah, for w/e reason my mind went to a UI update or something large like that

leaderboard update
We'll update when we've collected enough votes and validate it. It can depend, but generally you should see an updates ~weekly.

[echo aurora]

Okay will look into and flag to the team. cc @upbeat horizon

[regal mist]

Lets hope, some are 11-27 days ago now

[civic spindle]

gimme like a special subscription plan for 40 bucks lifetime with no limits cus i want to

[echo aurora]

Yeah each leaderboard is going to vary in time it takes, as the more battles we see with that modality, the faster we're able to update.

[regal mist]

Alright thank you

[coral bobcat]

I'm here to learn, grow and create something special and LMArena is the community I believe will help me get there much sooner than later.

[river minnow]

It seems to be just the Seedream-4-2k image model. The others work fine. Maybe this info helps.

[echo aurora]

Apologies I wasn't thinking - we saw some reports of this earlier today (~8 hrs ago) where I flagged to the team.

[echo aurora]

Welcome welcome!!

[magic stag]

pineapple-thinking-32k

[echo aurora]

exclusively non-thinking

[magic stag]

[remote arrow]

🤔

[cedar tide]

Sora 2 available here 😶

[balmy mist]

where?

[cedar tide]

Vidéo arena

[polar niche]

@echo aurora

Does this mean that it failed to generate?

The retry button appears after like 5 seconds

[echo aurora]

That retry and copy button normally only appear after ther model is done responsing. That does appear that the model is stuck.

How long has it been like that? Does a page refresh change anything?

[polar niche]

Well, none claude models appear to generate anything when I ask it harder questions.

[maiden fulcrum]

nimble bean is most likely sora 2

i have been testing it since last week and it is very good

[quartz light]

its probably just thinking a long time

[echo aurora]

Following up with you in the thread you've posted.

[maiden fulcrum]

but i can't confirm it is sora 2 for sure since it appeared along with king 2.5 turbo pro, and i have two i2v made from both which looks identical

so it could be another variant of kling 2.5

[quartz light]

where can i test?

[maiden fulcrum]

i have been testing it since last week on artificial analysis arena but i think you can test it here since it is just released on lmarena

[quartz light]

OHH right

[maiden fulcrum]

i have the two clips saved, i can show you how both look identical

[quartz light]

i didnt realise stealth models r in the video arena

[quartz light]

alr

[maiden fulcrum]

please check ur dm

[fathom prawn]

can someone help i'm new to discord
how text to image & image to video work here ?

and also guide me
what is Video Arena 1 , Arena 2 , Arena 3

[stiff kernel]

Hi! Please check #1397655624103493813

[haughty field]

Yeah ha ha

[fiery gull]

What Ai chat I can use infinitally? The qwen and aistudio crasher using a time

Oh no, any ia is very silly and shytopahy compared to the 2.5 pro

Hmmm I'm seeing that this is more related to the personality of the dry AI (api) and AI of chatbot

[high nest]

I'm enjoying AI, I haven't had much activity on Discord, although I have a variety of trainings that is on this platform. However, I believe I've found it to be worthwhile in learning how to navigate on Discord.

[ivory schooner]

Oh, Kimi has been taken down from LMARENA now...

[rich axle]

Hi

[quartz light]

webdev arena system prompt

whats especially interesting is the last line

"budget:token_budget200000</budget:token_budget>"

[nimble atlas]

what it is?

[quartz light]

system prompt of https://web.lmarena.ai

[midnight crow]

hello everyone i am new here

[balmy mist]

welcome

[midnight crow]

thanks

[lime jay]

Hello! I am Kunletaiwo and I am a new member here in discord

[wheat gulch]

I am using claude-sonnet-4-5-20250929 for claude-sonnet-4.5 for coding, but it makes a lot of bugs and errors which I don't think it is actually a 4.5 model.

[rough mauve]

hello any idea how to find our old video prompt generated in video arena >

[jade egret]

gemini need more personality ngl

[leaden tundra]

Hello everyone! Looks like a lot of fun in here!

[nimble atlas]

@echo aurora check dm whenever u r free

[echo aurora]

You can either check your DMs with the bot, or you can go to the search bar -> mentions -> type your username

[shrewd fog]

Hey, how can I check the limit for generating videos here?
As far as I know, the limit is five, and I already generated three yesterday.
How can I tell if my limit has been reset?
I just don’t want to use up my entire limit in one day, haha.

[echo aurora]

There isn't a slash command that'll show you how many generations you have, so you'll just have to count. Would note it's going to be 24 hrs after your first generation and doesn't reset at midnight kind of thing.

[shrewd fog]

awrite cap! 🫡

[teal orchid]

Here’s my current experiment: I’m working on a short festival-style vector animation for Awaash Malt.
Prompt:
“Animate reference poster in 8s. Background fade-in, typography spring effect, glowing script reveal, diagonal stripes sweep upward, golden Irreecha script stroke animation, logo fade-in with shine. Vector motion graphics, clean, no photorealism.”

Looking forward to seeing feedback!

[raw oyster]

ok but why is there so many nailong emojis

[fossil pelican]

hello i am newbie i want to make video

[fiery gull]

#1397655624103493813

[magic pine]

how to check highlighted

[fiery gull]

Idk

[frigid hollow]

How do I locate the image I was creating, I am new. It said I would be notified when complete

[fiery gull]

Are you signed in to a google account?

[inland quest]

All Sonnets usually being 32k thinking tokens, but 4.5 is 16k as 4 4.1 Opus

[tame agate]

hello, so wonderful ai in here, i am newbie i want to make video

[neon gazelle]

Not able to sign in on LM arena using Google account, can anyone help.

[random kestrel]

can someone tell e where can i make videos for free longer tham 8 sec

[split flicker]

/video

[full oracle]

#"Anthropomorphic hijabi cat doctor in a hospital corridor, surrounded by older senior human doctors with strict expressions. She looks calm and patient despite their attitude."

[blazing bison]

They always use the recommended thinking tokens by anthropic

anthropic even used 16k on benchs

so it makes sense

[tender kestrel]

Do we have an architect here?

[simple bay]

My video not show

[stuck orchid]

Does anyone know of an API with free suno-v5?

[ocean vortex]

no

[somber plank]

Api ? Gemini does

Ai studio free api but Use low tokens

[wide ridge]

hi

[cunning hawk]

hi

[leaden sun]

the name sounds like Maori language to me

[leaden sun]

knowing Anthropic...this is pretty within my expectation, nothing interesting to see here, move on to look forward to Gemini 3 Pro 😊

@ocean vortex I saw you in Claude's server yesterday and wow the mods there are pretty aggressive especially when one speaks the facts... that confirmed the toxicity level within their company structure, it was eye opening...

[leaden sun]

https://tenor.com/view/ritahayworth-gilda-hairflip-sassy-fab-gif-4998005916497632291

[solar hollow]

claude seems to just go allin on coding, they lack the rest a bit

[fallen aurora]

Hi

[slim quartz]

/“Genera un vídeo vertical 9:16 de 15-20 segundos con estilo publicitario para Instagram. Estética infantil, colores brillantes y alegres, pero pensado para atraer a padres.
Escena inicial: parque de bolas lleno de color, toboganes y bolas.
Después: niños pequeños (2-6 años) reciben a Spiderman, que saluda, hace poses divertidas y los invita a bailar.
Corte a escenas de niños riendo, bailando con Spiderman, haciendo pruebas fáciles, pintacaras y una mini discoteca con luces y música.
Texto animado sobreimpreso en pantalla:
  ‘¡Evento especial con Spiderman!’
  ‘12 de octubre – 1h30 de diversión’
  ‘Para peques de 2–6 años (bienvenidas todas las edades)’
  ‘Precio: 15 € por peque’
Escena final: Spiderman con varios niños sonrientes, confeti y globos. Aparece el logo/texto: ‘El Bosque de Andry – Tres Cantos’ y un llamado a la acción: ‘¡Reserva ya tu plaza!’
Estilo festivo, seguro, iluminación cálida, animado y atractivo para padres que navegan por Instagram. Sin voz, solo música alegre y dinámica.”

[tribal aspen]

but this is sonnet 😭

[pastel sonnet]

Same issue

I was it about to complete the development code

For me it doesn't show the time

it's just straight up says no

[barren comet]

create a lion

[tribal aspen]

i hope gemini 3 pro willbe better cheaper and faster

[regal mist]

GPT 5.1 is CRAZY

[coarse glade]

hello

how r u

[little grove]

Create a videos with different info graphics displayed outside the VR headset and student have to feel touch moments and topics like Human brain, Lungs, Chemistry experiment and kidney can be covered to show more immersive

[coarse glade]

sorry my firend said it on my laptop

[leaden dock]

hello

[coarse glade]

how are you

i love your website honestly and also u guys have rate limits on your video generation correct?

[robust yoke]

Opera Neon's getting released today, so that's neat.

Not to the general public, of course, but to certain people.

[robust yoke]

Greetings.

[robust yoke]

Correct.

Five uses per day (I think).

Or 8.

[timid ridge]

hi all!

[peak remnant]

hi

[keen beacon]

Hello everyone

[keen beacon]

How can I generate ai images

https://tenor.com/view/googas-wet-wet-cat-dead-chat-dead-chat-xd-gif-20820186

[robust yoke]

What in the Skibidi rizz just got laid upon my fine gyatts?

[robust yoke]

You can go to this site: https://lmarena.ai

[robust yoke]

Then, you can enable the image generation modality.

[robust yoke]

https://tenor.com/view/american-psycho-sigma-sigma-male-patrick-bateman-listening-to-music-gif-25257116

[crimson parrot]

generate a 3d cartoon Baby holds shiny red apple, smiles. Another 3d cartoon baby rolls a big bouncing ball across the floor.

[barren prairie]

Welcome to a new baby model 😍❤️

[ocean vortex]

Yeah that chick mod is uptight. Honestly I had to go back and reread it now as I already left the discussion before she showed up, thankfully, lol

[robust yoke]

Hah.

[hollow ivy]

i respectfully disagree
new Claude is outright amazing for roleplaying games and creative writing
i would say, it is now the best model for anything

[robust yoke]

I agree.

It's a very natural writer, just like its 3.5 predecessor.

(At least, I think I'm using that word right.)

If it toned down on the dashes a bit, then I think it'd be pretty indistinguishable from an actual human.

[robust yoke]

It seems to disagree that it's a creative writer, posing itself as more of a “conversational buddy”.

[shrewd isle]

Yeah

[robust yoke]

However, I beg to differ.

[shrewd isle]

And it refuse?

[robust yoke]

Refuses to believe it's a creative writer?

Or refuses to do creative writing?

[shrewd isle]

To say that it's a creative writter

So you have to clarify more

[robust yoke]

Then yes.

It refuses.

Claude is the ultimate Chad.

It doesn't glaze itself and admits when it's wrong.

[shrewd isle]

Yeah, Realistic but need long time to get it right but worth it though

[robust yoke]

Meh.

I feel like it's already enough of a human writer.

Just needs to slow down on the dashes, then it'll be indistinguishable.

[shrewd isle]

Yeah, better than most AI like ChatGPT which doesn't know how to even show a post realisticly

[thick rune]

Hello

[shrewd isle]

Hello! Izo

[robust yoke]

Greetings, Izo.

[thick rune]

I'm new here

[robust yoke]

Yeah.

[robust yoke]

Welcome to the server.

We hope you are able to make yourself comfortable here.

[thick rune]

Thanks

[robust yoke]

My pleasure.

[shrewd isle]

@Izo, which AI do you like the most like Claude ChatGPT etc

[remote arrow]

What's the advantage?

[robust yoke]

You gotta do @thick rune.

[robust yoke]

It's an AI browser, and as such, has an agentic assistant that, like Comet, can control your browser and do things for you.

[shrewd isle]

Supports Andriod?

Can be a MCP Browser?

[robust yoke]

Better yet, there are even two other modes:

• Make
• Chat

[shrewd isle]

I want to try out Browser MCP in my Andriod using Termux but MCP Browsers doesn't support Andriod so any suggestions?

[robust yoke]

Make lets you create apps and webpages directly in the browser, while Chat allows you to chat and generate images with the assistant.

[robust yoke]

No.

[robust yoke]

And no.

[shrewd isle]

MCP Browsers aren't for Andriod?
Waiting for someone to make one for andriod

[remote arrow]

Hmm.. The last Opera I've been used was GX. Never knew it's now have some neon lights on it..

[robust yoke]

I believe it's called “Neon” because it has neon colors. 🤪 🤪 🤪 🤪 🤪 🤪

[shrewd isle]

Any thought on the new Claude Code UI

[robust yoke]

Haven't seen it just yet.

[shrewd isle]

Test it out

[robust yoke]

Well, it requires a subscription, and I only have the free one.

[shrewd isle]

Can you react to my post of adding Claude 4.5 Sonnet Search to LMArena in the #1372229840131985540

[hollow imp]

Vro

[robust yoke]

Bombshell.

[hollow imp]

Please give new system prompt

@ocean vortex

[robust yoke]

Just did.

[ocean vortex]

Tool use / agentic it is actually behind on what they tested:

still not bad

[hollow imp]

@ocean vortex

[shrewd isle]

Thanks

[robust yoke]

My pleasure.

[hollow imp]

@ocean vortex DOMMMMM

@exotic owl

[shrewd isle]

But my latency is high so my msg takes like 10 seconds

And i tried posting a screenshot of my claude code but it stucks due to internet

[robust yoke]

Fair.

[leaden sun]

cant wait to try! been a fan of GLM4.5 since it was introduced 😊

[leaden sun]

many people on reddit would disagree with you 😂

[twin quarry]

What's up?

[flint sandal]

Frist deepseek, then claude, now GLM. This week is crazy.

In 2 days 3 models

[night crown]

Hello

[leaden sun]

https://www.reddit.com/r/ClaudeAI/comments/1npiulg/do_you_refer_to_claude_as/
there is interestingly a small portion of interesting people who prefer to think Claude being a... Chad-ette? 😂

am sure there are interesting people who like to think Alexa being a Gentleman too?

[flint sandal]

And i think that GLM 4.6 is better than Claude 4.5 sonnet thinking and GPT-5 is still on top.

[robust yoke]

Ultimately, a chatbot's gender is subjective, as they do not have a definitive gender.

[hollow imp]

@ocean vortex How do I add "Perform the deepest of the deep work" in your system prompt that it works

[robust yoke]

Generally, they're referred to as “it” or “its” or “it's”.

However, the only case where the rule of gentrification would apply would be if the company of the chatbot itself refers to it as a “he” or a “she”.

[leaden sun]

GLM has a great understanding of maths, so it's probably helpful with logically thinking at coding, claude sucks at maths unfortunately...

[robust yoke]

Some may agree, some may disagree, and honestly? That's okay.

[leaden sun]

time for Academia to come up with a term that is uniquely for non-biological intelligence

[robust yoke]

Claude suggests “chatbys”.

[leaden sun]

I've heard from the personality designer of Claude that "his" name is inspired by Claude Shannon, the founder of information theory, the designer team keeps it ambiguous but is intended to be more male-leaning because of the name being mostly male

[robust yoke]

Ah.

I do take Claude as more of a male, so that makes sense.

[leaden sun]

they train claude with traits that sounds more male or neutral than female, if I remember correctly

gemini is def more male leaning, it's so obvious

[robust yoke]

Nonetheless, it's still generally correct to refer to Claude as an “it”.

[leaden sun]

I've been wondering why this specific tendency despite the nature of the machine

[leaden sun]

you can call me "it" too then

Jules ist mostly male, but also used by female too (mostly in France)

[robust yoke]

Well, I could, but then that wouldn't be generally correct.

[terse sleet]

how to find my generation in video arena

[robust yoke]

It'd be dehumanizing you.

[robust yoke]

It'll notify you in your DMs once it's done.

[terse sleet]

tnx

[robust yoke]

To find it, just tap the button that is linked in the DM.

[robust yoke]

My pleasure.

[leaden sun]

you do know there are people with both genders and none-of-them?

[robust yoke]

And so, even if your name is generally seen as a gender-neutral name, it still wouldn't make sense to call you an “it” as “it” is for objects.

[robust yoke]

And you do also know that you are definitively born as one or the other, regardless of if you want to or not, don't you?

[leaden sun]

https://en.wikipedia.org/wiki/Hermaphrodite

[robust yoke]

“Either”.

Not “both”.

That's true. However, it is your choice on what you decide to do with that mutated chromosome.

Wanna be male? Go for it. Wanna be female? Go for it. Wanna be neither? Go for it. Wanna be both? Go for it.

None of that matters to me, personally.

[leaden sun]

I just want to make sure the world is not either black or white like it's either male or female

Those exact "beliefs" are making some interesting people to think Alexa is a Gentleman or Claude is female 😂 well, it's a social construct anyway

[robust yoke]

Yeah, exactly.

If you wanna believe that Claude is a male, then by all means.

Maybe it's a female.

Maybe Gemini's a man or a woman.

Maybe ChatGPT's a man or a woman.

Depends, really.

[shrewd isle]

It's genderless really but you can think of its gender anything that you prefer

[robust yoke]

And it's okay that I do.

Because it's my thought process.

It's not a fact.

Yeah, exactly.

[shrewd isle]

So stop arguing pls

[robust yoke]

At the same time, we can all think of each of their genders however we want.

[robust yoke]

It's not an “argument”.

It's a healthy discussion.

There's a difference.

[sullen quest]

No u are a healthy disscusion

This is an argument

[hollow ivy]

when will this get resolved?
Something went wrong with this response, please try again.

[flat crow]

im not a mod so forgive me but this conversation doesnt feel like it belongs in this channel

[robust yoke]

And that's true.

[hollow ivy]

i already had a situation, where a regeneration fixed it, and the model continued, but not this time

does it have to do with reset of rate-limits?

[shrewd isle]

Maybe, LMArena team will create #heated-talks

[tardy arrow]

Hello families