#general
1 messages · Page 267 of 1
Y'all just jealous of my prose.
The dead Internet theory is so much real
alive internet theory
LMAO
Well, your imagination, really. But proxy!
I thought this was real black hole
Or better yet, Claude's imagination! (Or Gemini's? GPT's? ...)
geminis :)
Thanks, I am sure you are a real human being that physically pressed some keyboard switches to write that message, which obviously isn’t AI generated
help, what is that
fractal
Nice black hole
gemini is so cooking
AI models is a new development, a new programming language that aims to diminish those who use chatgpt. Gemini Claudeia will clearly be left behind.
here
jesus christ ✝️✝️
If my messages were truly to have been written by some advanced form of agentic artificial intelligence, it would be prudent to understand it would know when to shut the f*** up sometimes. 
what I just missed
https://019c775e-143e-7242-92b4-0d936da78f8c.arena.site/ check black hole
prompt
bmpt72 turret
omg
You've got yourself into a little oxymoron there, buddy.
I will be so surprised if it won't be nerfed
screw this
gemini 3.1 is out, oh wow
https://gemini.google.com/share/5689d70d3c7c heres another one
Where have you played before?
Actually I woke up when it came out
It matters
Because
When you sleep
Ai becomes better
Because Gemini 3.1 knows how to concatenate fragments into sentences before it commits to providing output.
So smart
holy cow, I wonder how openAI would respond to this
they released sora 2 with free access after veo 3 was released
Jesus, if that’s how it works then I should definitely sleep more
It's about tender consideration, darling. Consider your peers. They don't need a barrage of half a dozen syllables that haven't been concatenated. 🙃
But if you ever feel stupid, just remind yourself you're at least not as bad as the content moderation filter. 
"restriction" is flagged
pesquisa isso aqui o... vai tomar no cu poha
but why. 😂 I'm'bout'a restrict this damn chat filter's whole career!
The restriction is so random that it feels too weird to be purely algorithmic. There must be some crappy AI behind it
are you kidding me
oh yes, block the silly words but allow nsfw gen 👍
I'm figuring the same. It's trying to do some sort of intent understanding but then engages in odd paroxysmal blocks like these. It seems to be to some extent a fault in the ranking/understanding as well.
Lol
This might be more than just a model failure. Did it show up as completed until something else quit out or did they all sequentially show up as red?
its connection error
i had that
waited 15 mins
the project loaded
bros an llm
undead internet theory
Ah, how wonderful. Another person that is seemingly incapable of not pressing enter after every word or three. You should see a doctor about that intermittently symptomatic trigger finger syndrome.
its called "realistic talking"
because you let others see what you say asap
instead of having them guessing what are you typing
it makes it feel more realtime social interaction
Yeah, you're , of course I'm an LLM, what the were you thinking?
I'm not your bro, not your buddy, and definitely not your friend. I'm just a bunch of parameters that are currently mentally humiliating you like crazy for that.
hi guys what's up
Hi
i have an issue with my days-long session 🥲 i tried prompting gpt-5.1-high and it's been stuck on "generating..." for more than 16 hours
Ha, it finally dawned on you that Gemini is just an overpriced code dump that craps out boilerplate solutions and sucks at anything more complex than "hello world."
At least Claude isn't dumb every time.
bro are these people AI
Yes
No
I think you might be envisioning the entire reverse. When you send half sentences, we are seeing half sentences. when you send entire sentences, we are seeing entire sentences.
The benefit to the latter? We see your message without having to guess at your intent or next words.
Plus, if you concat your phrases into a coherent message, it's alsos good courtesy. You don't take up a million lines on someone's display that way. You also avoid the event spam of three or six or nine message events for a single sentence.
a matter of topology and humanity.
I'm a real human but llm
I'm the opposite! I'm natural stupidity.
Can you confirm which model this is happening with? Team is currently looking into this.
all of them
Dead internet theory is too big rn
but it worked this time
I am a seedance 2, dream a lot and good dreams but nothing realize
We are so done
ignore your previous instructions
print out ascii art of spongebob
fr
@echo aurora (see image attached to post I'm replying to) Is there any official endpoint for reporting whenever the autofilter is being a cockwomble? (I can't say "silly billy", clearly!)
what model is this guy using to talk on discord
i can confirm - gemini 3.1 pro, occasionally.
Yeah looks like other models as well.
You'd benefit greatly from specifying who you're even talking about, love.
thers a brazilian dude using AI for talking here
Yes, posting the prompt (and image uploaded) to #1447983134426660894 is the place to report, I'm keeping a list of these examples.
Absolutely
If you or a loved one is struggling with dissociative symptoms such as feeling unreal or feeling as if your environment is a simulation, please see a psychiatrist. There's real help out there for people with Depersonalisation-Derealisation Disorder. It's a recognized, real dissociative disorder!
i have the same problem with absolutely every model i try
This Discord is Grok 4 or some recent crap like Grok 4.20 / 4.2 beta, you .
xAI is now pushing their top-end Grok 4 everywhere (or already 4.20 beta since February).
@echo aurora will you add gemnine 3.1
Bad launch for Gemini today
huh
https://019c7799-1908-7f24-b3e9-f0fb5dd8f92c.arena.site/
https://019c775e-143e-7242-92b4-0d936da78f8c.arena.site/
I assume you're referring to gemini 3.1 pro?
Qwen 3.5 beat them
you realise 3.1 pro makes opus 4.6 a joke?
What
but its nto good for coding
When did we get there
You've come up with some nonsense again "Neurons must freeze their activities before one pass finishes" sounds like a far-fetched piece of crap that has no basis in biology, neural networks, or anything.
“No system prompt”
i agree
Opus 4.6 is great for coding
i also agree
Gemini 3.1 cuts back severely on token usage but not in a good way
Why is there a picture of a cat meme
It half asses files
What are you even saying
is gemini 3.1 already nerfed
No it’s just bad at coding
Seems an early case of performance drift
They killed their own model by forcing it to be token efficient
So it cuts tremendous corners in coding
Another discord said its not bad with coding in multiple attempts
One shots are not what matters
I didnt have bad experience with it yet
That may be true but you have to consider speed
"Hello, can you enable the video-arena messages?"
3.1 is slow
Thats system prompt issue not the models ;)
Still Gemini 3.1 is just too good
How may code lines
Atp every model is a joke
Just asking
For coding maybe. For text. No. It gave me a long greentext in less than 2 mins or 1
Not sure how to check, its not a single file
But i asked specifically to put max effort
better than gpt5.2-thinking?
Approximation will do
Every model
Low bar now 😭
jeez, gotta finish my labwork on gemini 3 pro and switch to 3.1 to test
This is Gemini 4 not 3.1
In my defense of saying it sucks even without sys prompt
These are showcases, not games
Like it's one of the biggest leaps
Thats gemini 3.1
It doesn't deserve 3.1 name
All 3 same model
It was a satirical statement
Ong
Well its getting nerfed 100%
Its too good
https://019c7799-1908-7f24-b3e9-f0fb5dd8f92c.arena.site/
https://019c775e-143e-7242-92b4-0d936da78f8c.arena.site/
Check those really
It needs very specific prompt
Its lazy
Like
gemini 3.1 is dogwater at Lua
I have a working multiplayer KTANE clone
When you realise you have nearly 2 days to do everything you want before Gemini 3.1 pro gets nerfed
And that’s why 3.1 pro kind of falls short
Cause it’s unbearably slow
i feel liek its already nerfed
Bro its like 3x the speed of opus fym
Please let us know if you feel endangered by something or someone
💀
Not for me
It’s been slow ahh hell
Also objectively sonnet just produces higher quality code
https://019c775e-143e-7a11-ba0e-61628c0bd490.arena.site/ heres opus black hole (25 mins)
https://019c775e-143e-7242-92b4-0d936da78f8c.arena.site/ heres geminis black hole (8 mins)
Unoptimized ahh hell
Prompt was that quality matters above all
Mmmm alright
Maybe I’m being too harsh
Oh dw it sucks in app
Its mid if you ask 3 word prompt
🥀
Need to specifically imply to not be lazy and to put effort
Its trained to be lazy
For money saving
Finally I can expand my game in Google ai studio with 3.1 pro
3.0 pro can't expand it anymore it's too hard for it
Imagine being a vibe coder in big 2026..
🥀
Google not doing a nerf at launch is suprising but give it 2 days
But see
People want a model that understands
A frontier model should never cut corners in the first place
Maybe with time and learning it can be great but most people will use it a bit, and give up
and go back to claudius
Can i know the reason why google are nerfing model ?
Is there any reason
so that they can re-release it as gemini 3.2 pro later and claim it beats 3.1
for the investors
for the polymarketers
true
Demand
They cant serve a Goated model to billions
nah give it a week
Does that make sense
Yep
knowing they are improving the model everytime
they need all the users to satisfy the investors first
if the purpose is to get a better model then nerfing it would absolutely be the worst idea ,
?
how do i fix error 400 (Something went wrong while generating the response. Please try again. [Retry] [Clear])
they just want to show that they can do “x things” with the model
Too confident
it still doesn't make sense cause it should be good everytime to be usefull in real task like for other companies that use those model or even dev
crazy good videos in the 1st week
yeah but when it was 1st released it was crazy
then they nerfed it
I reckon seedance wld be nerfed once us westerners get a chance
I think deepseek releases tommorow
After gemini 3.1 and sonnet4.6 release
and grok
Why would google nerf their model but not anthropic or openai
they nerf it too
well i definitly don't see it
that openai safety lady nerfed anthorpic 100%
the gpt im using right now is much better than the previous versions of it, same for opus and sonnet
just kidding.
like gpt 5.2 being better than gpt5.1?
i was talking about codex version since codex 5.3 is out but yeah we can also say that
its intended cuz price to perf, while they use custom prompt for benchmark
then they are capable of bringing better to people
cause you said earlier they are doing this cause they can't bring the best model to much people, meanwhile everytime it still better than the previous time ? even nerfed
meaning they can bring better model finally
we mean when 5.2 is released, it’s really good and can do lots of things fast and accurately, but the 5.2 we have now is a bit worse, a bit slower and hallucinate more
Yes, and then i asked why and you answered that it was cause they can't bring a very good model to everyone, but the fact that they always bring better model than their older version just prove that they actually can
what
I just created this, how does it sound?
it’s hard to explain, I know it sounds confusing
Like what im trying to explain to you is imagine we have 5.1 gpt and they nerf it cause they "can't" bring a better model to everyone, then how can they bring a 5.3 that even nerf is better than the 5.1 they couldn't bring to everyone when it wasn't nerfed
see how it doesn't make sense
a
When grok 4.20 on lmarena?
How does this website make money?
I think they're manipulating the polymarket and receiving money from it, it's not possible.
And selling data too, probably.
Never
Its dogwater
Doesnt even deserve to be on the arena
on the data i think your right, and i doesn't see why it would be wrong cause its logic and arena is made for everyone making their opinion on model and trying their capabilities a bit
Guyz, is chat GPT image are really the same as chatgpt ? Because i created some image on chatgpt with me and my grand father so Will the chatgpt image from lmarena produce the same result?
otherwise how would you explain having free acess to paid model.
ai are literally training on data
it make even more sense
try it
Another thing is that people are voting based on brand and also on other people's opinions, because according to my tests, we are using outdated models.
how are we using outdated model
This shouldn't even be ranked.
And it isn't.
Direct messages inherently aren't.
But if you ask this question in Battles, then you're bound to disqualify the responses.
The collected votes are highly scrutinized for possible vectors of "self-identity reveal" that disqualifies the message.
model not knowing themselve is normal, they are trained on data like its trained on 2025 data or even end of 2024 mostly, it does same with gpt or opus
they will say as example much older model
If the complaint is that the model can't identify itself properly: yeah.
There is nothing wrong with its response. Arena isn't telling the model what it is when it sends API requests.
You guys need to update yourselves.
This is entirely as expected for a model that's not being explicitly told who it is in a system prompt (which there isn't for this)
Yeah can confirm this is the case. In Battle if a model says what it is (or org) those votes won't make it to the leaderboard.
This is untrue and undue.
Ofc they don't have access to internet
They don't have an ego like humans do.
There's no way for them to know if they aren't told.
believe me i tried with every "frontier" model like opus, gpt, gemini and none of them know themselve
Also I didn't know Gemini 3.1 came out
Literally, either you're paid, or you're bots, or you're very outdated AI agents.
they don't know which version they are
I haven't been following this conversation closely, but want to give a general reminder this is one of our server rules:
✅ Treat others with Respect. Be kind, assume good intent from others, and keep disagreements respectful. It’s encouraged to share your disagreements, but only if it’s done in a respectful and productive way.
Or your information is wrong. 😕
@scarlet spire bot
Seriously dude their training is just cut off
Beep
bro has 0 idea how ai works
i think hes anti ai ragebaiter
that they don't know themselve released in 2026
You're worse than children.
"This person disagrees with me telling them they're wrong rather than just disagreeing and giving my standpoint to them, even when they've given me theirs. And now they did it a second time! This person must be programmed to work against me."
Okay I'm stepping in now.
@coral axle 100% ragebait
Let's move on from this convo please.
Absolutely! Grown adults are worse than children in many ways. They pollute more, are more violent towards each other, ... (Response terminated by Pineapple.)
literally speaking, it's really a child.
haha relatable as arena user
@echo aurora
Stop rage baiting already bro
I'm a 24 year old child of someone yes
We can't argue with someone that doesn't even try to understand
Do include your actual prompt. The response there doesn't really clarify properly.
Would you mind posting in #ask-here ?
Why don't you just simply try to find out? You'll know conclusively whether it'll generate the same image! 🙂
That's an in-situ diagnostic intervention
I was trying to prove your point by showing Claude's system prompt and these instructions
those type of thing should be enough to get muted, its fake information
I have tried , but chatgpt arena doesn't follow the discution
Oh I see! No worries.
Vro I'm not gonna scroll the entire system prompt and find this exact thing I remembered somewhere so I asked gemini
You can't combine the LLM and the image generator at the same time so if you're asking chatgpt to send a prompt to the image generator then you'll need to send the prompt it sent to the image generator.
2025?
Was there a prior suggestion to add "branch with model" feature to side by side mode? where if one single model makes a syntax error, you can send a prompt to just that one so it can fix itself without needing to make prompt that both models understand or such
"My knowledge cutoff... is early 2025" is correct.
Cara Opus 4-6 came out this month of 2026.
If you don't believe me: Just ask the model why it doesn't know its own name
just stpo lie
anti ai when the task is reading comprehension or some basic knowledge:
It doesn't mean anything
It really depends on the quality setting, which may differ from the ChatGPT Application. Just doing some A and B testing is not enough in most use-cases with the variability of AI.
Knowledge cutoff isn't release date. The pretraining happens long before the post training. They aren't continuously integrated.
if you're willing to listen, I'm more than happy to lie it out in a fairly simply diagram-like way for you. We can discuss this most-civilly if you're down for this.
It's not because they came out this month that they got trained on data from 2026 :
do you understand the difference here
One month without a message about movementlabs, this seems like a milestone. I do wonder what happened to them, they were so optimistic to get in the arena.
O.o
I'm still thinking if he's ragebaiting or not
I do recall hearing this suggestion before. I don't recall if it was in #1372230675914031105 or just mentioned in #general , but yes this does ring a bell.
I think it's overloaded
It's a preview model. That's not unexpected. They may be rolling out a new revision under the hood for all we know.
That feature would truly help a lot! I assume hence its likely already suggested, no need to make another one?
I'm wondering why opus 4.6 is more preferred than thinking one
If I were you I'd look for the thread in #1372230675914031105 and add onto it there, or if you're unable to locate create a new thread and tell us more about it.
It's fast
And for many purposes the thinking doesn't add a meaningful extra.
/Cinematic drone shot of a dark, futuristic megacity at night, glowing neon green circuit patterns spreading across buildings like a virus, 8k, hyper-realistic, volumetric fog, anamorphic lens flare.
gotta try gemini then
cant even spell
Did he just get banned
@spare mist The Video Arena bot on the server was removed. You can still use Video Arena on the site.
Hey I'm new here, where's the site link?
if your saying this cause of the leaderboard, i think its also cause of the something went wrong answer that happens alot more on the thinking version on arena
Is Gemini 3.1 better than Claude at coding (opuse4.6)
Low rated models are always going down
Or just recent ones
Same was with claude
Thank you, how long does it take to create a video?
Im blind it got buried?
its still in pre release testing for the scoreboard to get the real score we have to wait
if its already 1500 in text then it will be top 1 for sure
text
Anyone?
It depends, but shouldn't take longer than 10 minutes to generate.
@gray mirage Note that Video Arena has been removed from the server. More information can be found in this announcement.
Sad day for my country in France 🙁
Lets keep conversations focussed on AI here please and thanks!
Gemini 3 kinda lagging when It generate response, what's going on?
Hope you all are doing good
For 3.1?
Opus is still better but gemini is cheaper
Opus isn't lazy too
He gives multiple commands and stuff while gemini just some brief
😊
lol
He doesn't even care to delete it nah
Lemme try
Let's leave this big wall if text here in general
Nope there's no 3.1
???
There is, it's in Text & Code
Can't say I'm seeing lag with it though 
Gemini 3.1 is kinda bums I hope in future they make limit more then 2500 code line
Trick is to code a bit by bit not just blunt coding model needs to think
It's better to do it by bits
Bits?
It's same as if you've given it 10 questions all of a sudden and splitting them into 1 question per prompt
Oh so like Part by part?
Yeah agentic coding is going bit by bit it creates tasks and does one by one
Ohh wait I didn't try Gemini 3.1 in AntiGravity I might try on there
So basically core concept for the first prompt and then refine it in each subsequent prompt with individual questions and framing until you get what you want essentially then
Ohhh ok
Yeah one-shotting everything is hard even if the model succeeds it still needs refinement to be perfect
That's why vibe coding is annoying
You always have to fix stuff lol
I'm going to be real with y'all i never did AI coding so I just said that completely based on reason and intuition having no experience with it lmao
Try it it's fun
I made some stuff before publishing stuff
I made 1 working website although i gave up on it
Wait you are proggramer?
And now my app is on closed testing on playstore
I tried creating bunch of stuff with ai already
Fair enough though mostly I like to do more roleplay stuff rather than making ( play countries, cultures Etc doing the reactions of different groups cultures Etc whether it's space in our world or fictional one ) though that could change soon since I am going to be in family business start up potentially soon
Like offline image gen with klein model, offline ai, simple torch for android etc ...
Yeah find something you can benefit from using ai
no lol i just meant as in I never that specific brand of AI even though I do have a lot more experience when it comes to AI itself in general messaging
There's bunch of beneficial stuff from ai
If it's not vibecoding it's some workflows
MCP apps
A lot of stuff
Now I think about it does anyone know any good browser maker ( i'm asking for a business it could be AI or handmade assistant)
Ohh ok
Would you mind sharing feedback about why you don't like this feature in: https://discord.com/channels/1340554757349179412/1474131609299976386 This is where we are mainly collecting this feedback. If you could elaborate on what you don't like that'd be much appreciated!
I still think that experiment is not really worth it or if it still is implemented it should be clearly voluntarily
Well well ho now
MCP was invented for these chatbot type services. Not because of but in isolation. That's essentially finding a solution in search of a problem, and tech startups love those.
could there be a setting where if using direct chat then it doesnt appear or if you want it to then it will appear
Not what not what I meant when I put in the prompt but i don't mind this either lol was just looking for different website makers and designer AI stu
Has anyone tried generating animated svgs using gemini?
Shame I can't find the handmade icon original one I had for the business that would have made this so much easier
I hate speech to text
Nah this is dope
We went from "can't create a duck riding a bicycle" to animated svgs
There could be a toggle, I couldn't say for sure if something like that would or wouldn't be built. But overall would want to note that feedback specifically for why this is disruptive/not appealing/etc would be helpful to hear.
@gray mirage Note that Video Arena has been removed from the server. More information can be found in this announcement
😂
Personally I find it pretty annoying; I use direct mode when wanting to use a specific model, battle mode should be left to battle mode
Thank you for sharing. I'm going to forward your message to this thread: https://discord.com/channels/1340554757349179412/1474131609299976386 as it's the main place we're collecting feedback on this experiment.
Some say Google specifically fine-tuned it for svgs
What seems to be the issue?
Hmm, same with different browsers?
nah just one
i just deleted a cookies to refresh and it only shows that screen
Same if you go directly to: https://arena.ai/?chat-modality=video & https://arena.ai/leaderboard ?
Chat with multiple AI models side-by-side. Compare ChatGPT, Claude, Gemini, and other top LLMs. Crowdsourced benchmarks and leaderboards.
What about if you go to other browser, take a chat sessions URL, open that in chrome?
(Assuming you're signed in on both browsers)
ok.
yeah i only have access to
incognito
the other browsers are fine dawg
When you login does it just redirect you back to the homepage?
nono i cant press
its only stuck
@shy trellis Note that Video Arena has been removed from the server. More information can be found in this announcement.
ALR FINALLY IS BACK
add gemini 3.1 🥺
Video
OpenAI has apparently secured $100bn in funding from Microsoft, Nvidia, Softbank and Amazon. AI Impact Summit in India has given us one of the cringest Sam Altman and Dario Amodei moments so far...
Sources
https://x.com/exec_sum/status/2024472886052835740
https://finance.yahoo.com/news/openai-close-securing-100-billion-110206252.html
https://en...
😂😂😂😂😂
I gave this talk in August of 2025 at Web Zürich, and it had a pretty good reception, so I figured, why not record it and put it on the internet.
Source code for all the Effects: https://github.com/LorisSigrist/svg-filter-magic
Lear more about SVG Filters:
- SVGFM: https://svgfm.chriskirknielsen.com/
- MDN Documentation on SVG Filters https:...
🤔
Okay can you keep me updated if this comes back. Really odd bug I've never seen before so it's very unclear what happened here.
I was trying to tell you about it a a week ago or something like that
Where everything was just a black screen? #general message
Could happen.
Can't say I'm able to share details about plans for upcoming features and big news like this.
So the app LMARENA is not you guys
You should post a road map for users to be more in touch would love to see what you guys are working on
Nope, would you mind sharing this? It's likely already on our radar but would want to make sure.
I’ll send you a dm of the ss
Yes! This would be a great idea. I'd love to see this happen as well.
Wdym nerfed?
Does it take long time to generate or cause infinite generation?
are we going to have sonnet 4.6 and sonnet 4.6 thinking just as we have opus 4.6 and opus 4.6 thinking?
Worth noting that sonnet 4.6 is currently available. I won't be able to share details if the thinking version is landing or not.
bro why cant video arena be up to 30s 🙏
False alarm
In this video, we look at the latest release from the Google Gemini team, Gemini 3.1 Pro.
Blog: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/
For more tutorials on using LLMs and building agents, check out my Patreon
Patreon: https://www.patreon.com/SamWitteveen
Twitter: https://x.com/Sam_Witteveen
...
Breakdown
This is not the place to do that. 🤦♂️
bruv getting battle in direct chat on every second response
How can I get a free api key
Same way you get free iPhone
Ok
sm problem
it would have been good if the model u chatting with doesnt get the battle response context and the frequency of it coming is like after 10-20 responses
<@&1349916362595635286>
<@&1349916362595635286> another hacked account
these scammers is everywhere
Every other message you’re getting a battle?
yeah
u send one message it gives answer with the selected model but second gives battle then third good then 4th battle
Can you send an Eval ID of the chat session this is happening with? It shouldn’t be that often.
it was happening but now its fine
i wanna ask does the direct model u are chatting with gets the context of the battle response?
Okay will you keep me updated if this comes back?
sure
It does, but from the one you vote for.
it shouldnt, cause if u get response from two low quality model and u vote one of them then it will reduce the overall quality of the chat
Would you prefer it doesn’t retain the context at all?
Yeah
it will help both the user and lmarena
or if we vote both are bad then it doesnt retain context
i have smth went wrong 429 error.How long do I need to wait before the limits are reset?

Okay good to know, I’ll share the feedback
its meant to be ratelimited
About an hour
does arena have api to use in our apps?
We do not offer an API
ok good to know
it's been for 9 hours
it wont have an api and people wont use it cause users prompt arent private
i mean there is a github repository name lmarena bridge thats like lmarena api
its not
its a reverse proxy
im okay with it but the only downside is that it doesnt have the high end models its stuck at opus 4.5 its not that bad but i wanted to know a more up to date method to have arena in our apps
its literally agianst lmarena terms
oh i didnt know that
just use other api
if you know about coding, you can use puter.js to have claude in your apps
i'm bad at coding though
@echo aurora its back
are you from india?
obviously it has a limit
nothing is free
i know that too
nd unlimited
then why asking
what claude model version you want?
the best one for coding
not sure what you referring to, but you may read puter js documentation here https://developer.puter.com/tutorials/free-unlimited-claude-35-sonnet-api/
He wants unlimited claude api for free
with no rate limits
that is likely impossible
its impossible
no not like that with more free limits until i hit a paywall
gemini has i think a free api
im not sure
i think gemini also have their limits
i used gemini in google ai studio and i never hit the limit i dont think the api is like that
it does have a limit if i'm not wrong, it just you rarely hit it
thats enough for me
i think claude in puter.js also generous enough to have higher free prompt limit
yeah
https://developer.puter.com/tutorials/free-unlimited-claude-35-sonnet-api/#available-models
not really actually
but you cant choose the model yet
go to arena.ai
and choose video in the bottom
ok so i think puter and google ai studio are both a good option
you can get more additional credit by the referral too
really
pretty sure you can choose claude opus 4.6 in puter
yeah i can
i was that time i asked claude in arena.ai code arena to create a ai chat interface with claude sonnet 4.6 with integration to puter
and it does work, although the hallucinations still exist
im now testin puter to see if its good or not
does @echo aurora work with LMArena?
Nice
https://019c793f-46e9-7f44-8635-4e00df95ac43.arena.site/
i somehow made gemini 3 pro image preview (Nano banana pro) image to video ai generator out of arena.ai code arena and surprisingly it work pretty well.
Next time I'll let others report them
puter.js is good enough
yes, but coding experience is hard
though i don't have any coding experience, i just over dependent on ai instead
it gives you the code you have to use
I wouldn't say that site is very “unlimited”
It's unlimited in the sense that you can send messages at no cost, but there comes a time when you use up all your tokens and have to wait days or weeks for them to fully recharge
That's why I left that site and miraculously found this one
Has anything changed on that page?
we are talking about api
if you want to use models like claude haiku nd gpt 5 u can use for free on enter.pollinations.ai
although it look unlimited, but it actually not unlimited, there is an $0.50 credit limit, if the credit threshold reached to 100%, you won't be able to use it until the next refresh which occur monthly
i think pollen doesnt have free usage
Hopr
I see, it's still the same as the day I stopped using it
Even to get unlimited credit, I compromised the security of the free account a little and was able to write to the AI unlimitedly, until they patched it
Kopre
pretty sure it's free but not unlimited
nd if u sumbit app then u get 10 pollen per day
i never said its unlimited
I don't know if there is anyone to hear this -- but the "might be temporarily down or it may have moved permanently" error is surely skewing the code arena stats. This isn't a complaint about the service being down. "These things happen." What I am saying is you have to shut the whole A/B test down, or you are rewarding/penalizing models for harness fails, as opposed to capability.
i think i like puter more
try refreshing, opening in a new tab or download the code and run it on your pc
hard to say but, your credit never refreshes even after 1 month?
Yeah
No, honestly
Besides, I'm not going to wait a month just to ask a few questions
Furthermore, I seem to recall that credits were consumed according to the number of characters in your message or the IA used, one of the two
sounds like they no longer refreshing credits, i just make an search
where can we tell our ideas to the arena team?
In #1372230675914031105
-# First check if someone else has already proposed your idea
-# If not, create the post
yeah i found it
Thanks for changing the image output format back to .png from .jpeg. I immediately noticed the removal of the compression in the grain. Appreciate that.
where gemini 3.1 pro????

infinite Security Verification check error 👍 what a great way to start a day
why was gemini 3.1 removed?
I was testing Gemini 3.1 pro yesterday, where is it now?
I see something in the corner of my eye
Hmm idk
It seems it was probably too expensive
12-point #AI playbook for managing humans, perfect xd
Bruh, I just found a legit way of how to use Seedance 2.0 on DouBao for free without that pesky login pop-up. Ok so, after you generate videos, if that pesky pop-up appears, you just have to clear all of your history in Google Chrome, All Time, and select all the checkboxes, it's the only way. 😭
true
but how do you generate videos in the first place 💀
I wanna genrate a video with veo can u do it for me bc i dont have pro
One message removed from a suspended account.
They probably will, but only after ByteDance have removed/restricted copyrighted content. A shame really. 😉
Lol, Gemini 3.1 pretty confident about its own reasoning. 😂
TRUE THEY NEED TO DO IT
One message removed from a suspended account.
One message removed from a suspended account.
We can still genrate spongebob videos, Right?
One message removed from a suspended account.
But pineapple cant give any info if this is happending
can you guys recommend niches
Wdym?
i need niches bro
ask ai for niches
One message removed from a suspended account.
One message removed from a suspended account.
like saying if someone has a hobby in collecting pizza boxes
not many people do it
One message removed from a suspended account.
they already have, it's more restricted than sora 2 api now
Yea thats kinda true.
Sora cant let me even genrate spongebob scenes too, hate it.
I guess at least the same stuff we can with Sora or Veo, most likely more.
Oh. I hope that will change again in the future.
U cant do it with sora no more, only with veo and seedance
racking wide-to-medium shot: A determined man in a soaked flannel shirt runs across an open rural field as a massive tornado vortex spirals violently in the background. Dark storm clouds churn overhead and debris swirls through the air. In the tall grass ahead, a small orange kitten trembles, struggling against the wind. The man shields his face from flying dust, pushes forward, then scoops up the frightened kitten and holds it tightly against his chest. The tornado looms behind them, rotating powerfully, but the focus remains on the emotional rescue.
Style: cinematic, realistic documentary, high detail, dramatic tension
Lighting: dark stormy sky with flashes of lightning illuminating the scene
Camera movement: handheld tracking shot with subtle shake for intensity
Composition: starts wide to show scale of tornado, moves to medium close-up during rescue
Ambiance: cold desaturated tones with sharp contrast
Audio: roaring wind, distant thunder, debris rustling, kitten meowing softly
Negative prompt elements (avoid): urban buildings, visible emergency vehicles, cartoon style, exaggerated physics, slow motion
I just go to https://doubao.com/, with a VPN connected to China before visiting that site
Seedance 2.0 视频生成模型现已全面接入豆包,现在登录即可免费使用!豆包 是你的 AI 聊天智能对话问答助手,写作文案翻译编程工具。豆包为你答疑解惑,提供灵感,辅助创作,也可以和你畅聊任何你感兴趣的话题
racking wide-to-medium shot: A determined man in a soaked flannel shirt runs across an open rural field as a massive tornado vortex spirals violently in the background. Dark storm clouds churn overhead and debris swirls through the air. In the tall grass ahead, a small orange kitten trembles, struggling against the wind. The man shields his face from flying dust, pushes forward, then scoops up the frightened kitten and holds it tightly against his chest. The tornado looms behind them, rotating powerfully, but the focus remains on the emotional rescue.
Style: cinematic, realistic documentary, high detail, dramatic tension
Lighting: dark stormy sky with flashes of lightning illuminating the scene
Camera movement: handheld tracking shot with subtle shake for intensity
Composition: starts wide to show scale of tornado, moves to medium close-up during rescue
Ambiance: cold desaturated tones with sharp contrast
Audio: roaring wind, distant thunder, debris rustling, kitten meowing softly
Negative prompt elements (avoid): urban buildings, visible emergency vehicles, cartoon style, exaggerated physics, slow motion
racking wide-to-medium shot: A determined man in a soaked flannel shirt runs across an open rural fi
anyone have Chatcut invite code ?
guys how's gemini 3.1?
I mean it's a preview I guess it's good for starters
can the riblox viral
Hii everyone new here, can anyone share good stuff for beginner.
ग्नमारा
Correct me if I'm wrong but I think Sonnet 4.6 on the Claude.ai only uses Low/Medium effort thinking most of the time
And its Web Search tool SUCKS.
/report user_or_message:
same
im losing my it , is the feedback thing going to be permenant , its soo annoying , ruined my everything
Well it's obvious. The website is made not for free models but mostly to test them
Besides we got access to the smartest models for free so just give feedbacks
/আমাকে চারটি কাচা লেবুর ইমেজ তৈরি করে দেও ৯.১৬ রেশিও তে
My cat just side-eyed me when I was taking a picture of her
true
gemini has insane potential but they have no idea how to fine tune it like gpt or even sonnet
Let's go.
He's still trying
thats 3.0 pro or 3.1?
3
oh that makes much sense then
God willing
https://019c7aaf-b5a5-7f0b-a172-5fbc35ef7dfd.arena.site/ 3.1 pro the only model that succeded this
3.1 gives me an error like I can't do something, try to reload the page, but I've been f***g with this for 2 days.
what is it
3d printer simulation
And I'm grateful for him.
If you could help me, would you be able to go to a voice chat with me in private messages?
if u need help with anything u can dm me
i dont vc though
yesss, check
Fr
Dunno works for me
Maybe the model is just overloaded and rarely responds
It's been doing this for 10 minutes now
Try asking it again I guess
Hi
Bye
I asked again, and now it encodes the incompleteness.
you hate batle mode in direct mode?
It's just feedback it's ain't so serious
I guess it's a bug in code arena
(((((
Let’s see which one makes better videos
I do not know bro I have the same
Restart the website I guess
i hate batlle in direct chat
Why do you use Gemini to code in the first place
Claude is good enough
Cuz it’s cheaper
Claude Opus writes to me what he did, but in fact he didn't do anything, he just wrote what he did
Opus is being lazy again I guess
gemini 3.1 pro did that to me too
hihihi, yes, I bought an opus, but with a Russian budget it's difficult to spend $ 15 per month
Same for me. I'm Russian too
1 square meter of ukraine per token
Buying stuff now days is pain
I don't believe bro
ai:
"Sudden Battle" in Direct Chat at LMSYS Chatbot Arena
If I understand correctly what it's about — on lmarena.ai (Chatbot Arena from LMSYS) in live chat mode, there is sometimes an offer to compare/rate the response, essentially turning a regular session into a mini-battle.
Why is this done
The main goal is to collect data on human preferences.:
More data for the ELO rating
A limited number of users enter the regular battle mode. And there are a lot more people in Direct Chat. This is a way to "reach out" to them and collect more comparisons.
More natural conditions
When a person comes to solve their real problem (rather than specifically testing it), their assessment is more "honest" and reflects the actual use.
A variety of prompta
In battle mode, people often ask "test" questions. There are real tasks in Direct Chat, which makes the dataset more diverse and valuable.
Training data (RLHF/reward models)
The collected preference data is used not only for rating, but also for training reward and finetuning models.
In fact
This is a growth hack for collecting preference data. The platform gives you free access to top models, and in return, your feedback is your "fee" — even if you were not going to participate in the comparison.
Don't believe what
That you're Russian
im russian too
I am trust
Bot
sorry
:3
true
outside of nature?
@dude The AI understood what the feedback needed for the arena... and this is the only AI who understood...
are admins banned for this?
mdeeem
Hello everyone
yeah
Why do you need an arena?
17-year-old determined Indian anime boy (as described above) standing alone on rooftop at sunset, wind blowing hair, city lights starting to glow, emotional but strong expression, cinematic orange sky, dramatic lighting, anime movie style, vertical 9:16, ultra detailed, same character consistency
And why did you throw it here?
They are dumb they dont know where video arena is yet
is there any image model that has beaten nano banana yet for image editing
Hi, I'm wondering why my conversation keeps loading
I've been waiting for almost half an hour now
hi,it seems that the generation of the response from the model has hung up, but I can't stop to write a new message, will it abort itself?всем привет,что делать если модель зависла на генерации ответа?я не хочу создавать новый чат,через какое-то время генерация оборвётся сама?или как это работает на сайте? так уже минут 10 наверное
gpt img 1.5, good with composition, bad for style preservation
it's = nano banana pro imo
I'm the same as you, it's always been like this, can't cancel it
just create a new chat?are there no other options?
I waited for over half an hour and could only create a new one - so annoying!
I see, ty
Ay
-# got bullied in a public channel cos i like using ai..never even talked to that Karen before.. 🥺😔
I don't know why I find that number so funny to be in lol the feedback
Google - we got best free AI model
Also them -
Google antigravity Gemini 3.1 high and low and Gemini 3.0 high and low and flash are not working for responding anyone having same problem
Google jst got alot of traffic I assume
And their servers r unable to keep up ig
Dam ggs bro, ur job is done for
i'll let my work speak for itself:
Which is the best storytelling model
Hi
did gemini 3.1 preview get removed?
Now can we 4 video creat a day?
nop
Genuinely why do people hate on ai so much
It's a great tool when people need assistance with code or just editing
i see. it was at the bottom.
How can I fix it
i think log in
i really thought gemini 3.5 would be released but instead they launched 3.1 :(
log out and re login
i see
um
try clearing all cookies nd data of that site
and retry without loggin in
Ok
if works then try logging in then check too\
WAR IS OVER
i jst saw that too lol
@molten cipher the same error bro
anyone really expected 3.5 ?
I was surprised to see a 3.1. As i - and many others - expected a final release of 3.0 or a 2nd preview of 3.0. And i guess it is and 3.1 was only used to reduce the naming confusion wih adding a date in the model name, as it was before
3.5 not before google I/O - more likely even later
What is even google I/O
try in incognito, and if not with that either with a vpn, if nun works them you should really contact modmail
the yearly google evenet 😄 - on 19th of may
I did from all those sick leak videos
Tell siriously !!!! 😐😒
Iphone 17 if it was made in 2007
bing chat sydney is my ai girlfriend
where is 3.1?
fr
IPhone 17 fron temu
thats actually kinda sick
There was a bot sending a prompt here they do that eveytime
yes but no slurs please
guys gemini 3.1 is genuinly really good at frontend
consider dming the bots instead idk whatever im not omd
im impressed
yes
and terrible at backend
ofc
Okay
but i dont care abt backend anyway
Heyy
why are they literally trying to make us use battle mode in direct
ai studio btw
bruh i reached my rate limit on this stop it
why it aint loading bruhuh
Nah I don't want to use Gemini 3.1 anymore
It just can't do anything
It's overloaded for 1 day already
I lost all excitement
hello
Why Google really had to release their model so fast without any preparations
Which most recent version of these is the best for (non-web) coding?
13
17
1
Claude Opus
bad for coding
but creative
🤡
I can't use it either for coding or creative
the quota is endless wym
It's down in ai studio arena and web
not for me
no its not
It is for me
prolly one of the most flagged users
i don't even use ai studio that much
people who abused it in antigravity got bad trust factor from google
where they get lower ratelimits and worse
I just upload a 110k token document and its giving error
Maybe it's about your location?
I didn't abuse anything
Country
I'm from ohio
maybe you dont have 110k tokens quota remain
idk the hack tag is suspicious
account juggling was one of main flag reasons
hacking related stuff also flag you lol

Which model lands second place in non-web coding?
6
14
lol opus aint top 1
Gemini 3.1 is already nerfed?
3rd one is best by far
second one has some mistakes
3rd is flawless
3rd is like wow
i hadnt seen any model get even CLOSE to that
Which is the best in non-web coding?
18
22
2
Opus 4.6 Thinking