#general
1 messages · Page 311 of 1
but why would they do that if the answer is actually good?
Because they get another answer
You don't lose them because they disappear when you open this model
And how much longer do we have to wait for at least GPT 5.4 to return?
ban
yes, but they also lose the good answer model
is gpt-5.4 better than sonnet-4.6?
?? Wdym
Oh well yea but then it moves to another model that could be better
It depends on the person.
or worse
so you have an incentive to vote for the better answer. there's no easy way to abuse it
when I do a side by side/battle its usually both are good. there's very few cases where one is clearly better. i dont know if it actually helps arena if you do that
sure, you can say 'both suck' and keep cycling through hoping to get a better model, but it may never com e and youi're wasting a lot of time
i try to evaluate the answer and pick one is better but there are many times I dont know enough to say that
and you won 't know if you cycled away opus 4.6 thinking, until you say both suck, but then you already lost it
yeah i dont do the cycling with both are bad. it seems like a waste of their compute resources
i wish that was my result, i find 60% of the answers are bad 😄
this is already a free service thats hurting and losing money
what do you use it for?
different tasks
Sorry to say we don't have an ETA on this yet
research questions, generate a webpage, generate some code, what does this error code mean, etc.
also a suggestion for people - grok isnt that bad at a lot of research/code tasks, i find it produces concise answers that work
I wouldn't describe it like we're "hurting", we're planning for the long-term
Sorry for the ping, but I still want to ask:
And approximate dates can be given?
For example... The end of this week?
it has a very different way of thinking and sees less over engineered
Yeah sorry I just don't have an estimated answer I can give you
what are you using it for? what type of questions?
Well, we'll wait until they tell us at least approximately about the timing.
eg researching a new topic and getting suggestios. writing quick apps to do stuff
app - web/python. eg i am working on a catalog app in which i can store stuff i own, files on disk etc
and it's one-shotting with success? interesting
im not a vibecoder so i know the tech involved, so i can ask it detailed prompts
which grok model are you using?
im bad at ux and design, so i will probably end up using some of the coding tools. arena is just for quick design research
in that case, that explains it, python is probably the easiest model for LLMs but it still makes mistakes in architecture, but since you're asking it very specific architecture and using python, basically any model with sufficient training data will work for you
i also want to use ai to ask about my taxes but i dont think its a good idea because of the privacy issues, but local llm arent possible for me to run
The problem is, by the time you fix the problems, people will already be gone...
Please include at least 2/3 important models, not that you remove them all and leave the worst ones to the Users...
but they have such different answers when I ask for a design
It's our hopes and intent to bring the models back to direct and side by side
i mean design spec/architecture. eg with using python as language and sql backend
i am using grok-4.2, either thinking or normal
Imagine using ai when a Socrates comes up to you and asks "If Ai is your power what are you without it?"
i mean the same can be said about most modern living
if your job is social media, what are you without the internet?
Hi everyone, I’d really like to hear your thoughts on this. I’ve done quite a deep analysis myself, and I have a lot to say and ask.
Given the recent changes and restrictions on the Arena, I’ve had to switch to other services over the past few days because using the Arena has become almost unbearable due to all the limits and cuts.
However, I’ve noticed something interesting. When I use the same Nano Banana 2K on other platforms, even on the original and most official Gemini, the image quality and accuracy are noticeably worse compared to the Arena.
This makes me wonder: is it possible that models on LM Arena are somehow enhanced or fine-tuned to perform better? Because I genuinely don’t understand how the same model, used in the same way across three different sites, produces consistently poor results everywhere else, but significantly better results on the Arena.
So my question is: do LM Arenas apply any additional modifications or optimizations to the models? And if so, why does this difference in quality happen?
arena.ai uses the api directly, gemini app nerfs it some
the gemini app harness is sort of terrible
you'd be likely better off using one of the third parties that offer it
cause they have to use it direct api, most likely
But which ones exactly? I'm ready to pay, as long as the result is identical to what I get on LM Arena. So far, I haven't found a single service that matches it. Take something as simple as text within an image: on LM Arena, it comes out without distortions or defects, but on other sites, everything looks 'dumb' and 'plasticky' somehow, full of artifacts and 'AI-slop'
what else have you tried besides gemini.google.com?
Flow
Google AI Studio
Poe
Telegram bots
poe i'm a bit suprised at, flow is basically google harness
have you tried from aistudio.google.com?
Which telegram bots
Yes , But why does it work without any distortions there [on LM Arena]? To be honest, I haven't tried the paid version in AI Studio yet, but I have a feeling the result will be exactly the same
yo um wheres the channel where i can turn my photo into a video clip im not seeing it can anyone help me
I feel strange among so many people who speak English.
The Video Arena Discord bot has been removed from the server. You can find Video Arena on our site here: https://arena.ai/video More information can be found in this announcement about the removal of the bot.
Why?!
Imagine using ai when a Socrates comes up to you and asks "If Ai is your power what are you without it?"
They got rid of the models people use the most it seem
At least image look fine
dude
i think that arena isn't gonna last until 2027 rn
depends what youi mean by last
Ah, Claude Opus? That was the one I used the most
be still used
It seems that's exactly right. Strange.
Yeah
i'd im agine it won't have half as many users unless they bring back some of the top models or better ones

but, i don't know about dead
cause they're cheap/nearly free to run 😛
and it helps them with their stats on their leaderboard
I don't think Grok is as good as the ones that were removed.
Grok sucks wdym
But that's what we have.
they can be like "we've tested over 500+ models! and we host 100+ models on our site!"
For clarity the recent removal of some models was only from Direct and Side by Side, this change doesn't have an effect on Battle and the leaderboards
👀
What are the chances of me getting the opus in the battle? 😥
I was making up a story with Claude Opus in direct mode, well, one is an understatement, 300 minimum XD attention deficit
ty, but can we have more than just 2 models in battle mode? that would be helpful, at least to me, when both answers are bad a lot
1/86 400 000
lol
and we are back to 2023 - 2024
Changes to Battle could happen. There are some interesting ideas for how this can be done. Expanding the number of responses, or some kind of "king of the hill" mode. There are some creative ideas we're kicking around
Direct mode has removed a lot of top-tier models.

Only with the landfill
guys remember this is a free service that gives you access to a ton of premium models. there is really no reason to complain about removal of models. try to complain to openai or anthropic about how they keep reducing limits
I hope Opus returns before stage 2 sbr
what's stage 2 sbr?
can max route to opus?
Steel Ball Run dude
sir i dont know
bruh
God
You know, friend
I've just started it; I recently finished Stone Ocean, and someone gave me the SBR manga.
Best AI/platform overall for free users rn?
12
13
2
Gemini
definitely gemini
Anthropic should make 500$ subscription Max+ with access to mythos
With the pro model you can only send two messages.
hell nah
Hell yeah
I can't use Gemini XD, damn country!
ban
What place do you live in that doesn't have Gemini available? Excuse my language.
The only country I know is Russia
Cuba
Okay...
XD
Oh
:-|
?
i call my dad to ban you
Oh, sorry man.
I need a VPN for almost everything, even Lmarena needs a VPN.
you hurt my feeling
Anthropic should make Max ultra subscription 1000$ WITH 2X ACCESS TO MYTHOS
you over
My apologies.
leha i am going to crack your back
What...
over
hello guys .. where can i chat a support team ?
Nah, don't worry, I'm used to it.
Chat BM
He will help you
If I were you, I'd be careful XD
why is arena removing models like cluade 4.6 and gemini
you can also ping the moderator role for support but only if it is important
That's what we all want to know, my friend.
Money?
Cluade?
yh they removed cluade opus 4.6
It's just a mistake, poor thing.
i got my own clouds
i tried , My chat hit the max length and I can’t continue it. It’s important for my project 😅
Any way to keep using the same chat or fix this? delete the first half of it or something ?
I was in a chat room of mine that only used Opus, so this happened. Maybe I'll be away from the chat for a long time.
at least yet if i am correct
I'd start with using #ask-here as it'll be able to answer most questions, if the bot isn't helpful, ping me in the same thread
Bro what are you talking about
pineapple fare needs your help
Do you need any help?
Calm down, Leha.
that funny i already got clouds that help me
cloud
this might be the worse day of Arena 💔
i like being near with my beloved clouds


this is kinda funny
yes thats exactly where they are
i tried 😅
Yeah, Pochita
new arena model!! ernie-image
the model no one knows
Still without an opus...
seriously?
well its https://ernie.baidu.com
ERNIE is a conversational AI developed by Baidu, global technology leader from China. It's designed to understand complex questions, provide clear answers, and assist with learning, problem-solving, and communication.
Is it any good
yes
O
Huh
ai week 🥀
I'm trying
So you're basically in the sky right now?
DDDDDDDDDDD
uh no
Ernie
no i got two robots near me they are robots that i call clouds 🤖 🤖
Oh
deep seek v4 early acess
Ernie...
Who even uses this?
are there alternative to arena ai
When you can use Gemini flash image 2.0
sorry guys its my first time using it .. so there is no way to continue in the same chat ? should i give up ?
Hmmmmmm
bro
are there
Yet not
it just released
are there alternative to arena ai
It's a SOTA image generator
Can you ping me in that same thread?
hi
Opus again gone 😕
it was better if i didnt talk about clouds
are there alternative to arena ai
i asked ernie-image for a rubiks cube with a mirror reflecting it 
too expensive man
what a nonsense i made up lol
whats up kenny
Gratis no creo
What????
i will test deep seek v4 right here
Cursed
ok lets see it
but its really fast, instant it seems like
Cool if it's really deepseek v4!
I kinda like it tbh
are there alternative to arena ai
yea, the provider's websites
it's an early acess so i don't expect too much tbh
Don't delay, pochita
what are they, u know
please dont get sad but i have to say it
I prompted "roblox" not really roblox but looks cool i guess lol
anthropic, gemini, grok, chatgpt, qwen
are the top 5
i think
that are free
well
im gonna use Roblox cuz it's the easiet option
-# too lazy to install godot rn
Tell me what it is, man.
smol
bro i am asking you'=
bruh
😥
text is good!
wish me luck
is ernie new?
crisp
eh, i could do that ||joke||
gemini can do something like that
I believe so, their last image model wasnt really announced and used ernie 4.5 as base
?
prompt was just "donald trump"
LoL
you should try pineapple
with glasses
good idea
ask it to make this
All that was missing was for them to put a demonic horn on it, since it's from China XD
dont ask where i found this
google definitely
iykyk
=)
we just asked for a pineapple with glasses
not a whole background
holy
has anyone tried realistic image edits?
oh hello
i want to edit a pic to make it look good for my linkedin lol
wondering which model to use
try this new model nick is showing to us and gemini
gemini > openai's image edit?
yeah
i will try
(for what i used definitly)
thank u frend
no problem, i hope i help you
I usually use image AI more for editing than creating images; I'm lazy.
Idk this model kind of sucks bro
which
The ernie one
i mean its their first image model
O that makes sense
for a first image model, it's kinda good
put me in a room with some paper and a pencil
ill do better
@echo aurora i tried 
put me
in a room
with nothing
and i will do nothing
- not me
is this ernie image?
gemini and openai
im making the second one my pfp
ok
ernie image only allows text input so i had to specify every detail
clear glass filled with yellow liquid, floating green pineapple stem emerging from top of glass. lime slide on the rim, yellow and white striped straw sticking out. two black paper circle eyes and a pink 90 degree rotated 'D'-shaped paper smile decal on the front. cartoony 3d render style.

ernie keeps giving me errors
we should have a shareable link for chat
most of them
^
too, my favorite is avocados

im allergic to most fruit and veg 😅
ernie better trust
yea 😔
for a first model it's actually good
why am i getting "Which is better" with max and it gives me deepseek v3.2 and 2.5-flash
😭
im using max for a reason bro
direct battle purpose is just to annoy ppl at this point bro
🥹
ok gemini actually cooked too?
why B&W
testing stuff with gemini
o
@echo aurora ^
but also im gettin errors with ernie often, can you ask team to look into it? thx :)
a95f44b5-3457
019d6fdd-17ed-790b-8f01-97bfa521b778
yall ernie image is gone

@pseudo hemlock @desert pendant
RIP ernie image
nvm its back
🎉 lololol
Anyone experiencing this also?
wait wdym
ai final boss
nanogpt, arena, claude, chatgpt
i asked ernie image for gta 6 gameplay footage
I don't understand, I'm just coming back after 3 days
dont worry, im just messing around
v4 is expected to be released around 2 weeks
-# expected not true tho
oh they added this 2 days ago i think
Thanks. I just tried the one I saw atop the list earlier, it's working now
Nns
THIS LOOK GOOD WHAT
caps lock mb
deep couldn't even do something like that in just 2 prompt
🥹 kinda liked the result
Made with which model
hard to believe that deepseek is totally free. been using it and i dont see any paid options
is it being ran for competition reasons or what?
free?
from what i can see
interesting
is that in html or through roblox studio
hm but arena's leaderboard shows that it does waste money
I just started a new chat and usage limit has been exceeded after a prompt now I'm being asked to start a new chat, wtf is going on with arena
Nailed it
Depending on the prompt it may already exceed the context limit
But it wasn't like this before, you guys keep reducing value
Limits can change over time, but I'd note that the context limits haven't been changed since ~2 weeks ago
Unclear what you mean by "before" though
finally realized making almost every model available to be accessed by everyone with almost no guards is unsustainable
Would not context limits have always been in place, but recently a unique error message was added so it's more clear when this happens
think they used to be substantially higher though
Because you're just a bot, you guys should just restore things as it used to be
tbh providing the newest models for free was never gonna be sustainable long term unless they get massive funding
i think most of us suspected it's not gonna last forever once the site grows
But they removed the newest ones already
YEah that's true
yes
but this site is like 3 years old
Not a bot
Anyhow, it's free to use.
I' m happy gemini 2.5 pro is still here
arena does not seem to be able to create AI videos that is vertical. anyone knows how to solve this? or any other software/programs that can do this?
this you ?
Lol, maybe worse
I really have projects to work on and I'm just sitting here seeing different problems
I'm sorry to hear that, we looking to make changes and improvements to give Arena users more control over what models they'd like to use.
are you a student? maybe your school/uni has free access to newest google models
Any way for Arena to create videos that are vertical instead of just horizontal?
I hope a solution comes soon
Yes, does my school even use AI? No
This was reported to the team, looks like a bug. Thanks for the flag 
There is not a settings sorry to say
will you guys look into doing this in the future?
What is april26-chatbot2?
Is it Claude Opus 5 Pro/Ultra or Claude Mythos Mini?
It outputs very slow, maybe it's a huge model
i believe it's nvidia, probably the 400B Nemotron 3 Ultra model
I want to buy "claude-opus-4-5-20251101-thinking-32k" from somewhere. The responses I received while using it on arena.ai were excellent. Is there an API service I can use similarly? I'm thinking of buying it with cryptocurrency. I couldn't find "claude-opus-4-5-20251101-thinking-32k" on OpenRouter. OpenRouter has "Opus 4.5" but no Thinking mode. Can you recommend a place where I can buy it?
as i remember
you can use open router or original claude
I was using "claude-opus-4-5-20251101-thinking-32k" in arena.ai. Would Claude Opus 4.5 or Claude Opus 4.6 in OpenRouter give similar output?
Have you ever bought anything from OpenRouter?
I really like the web interface on "arena.ai". Is there a place where I can use the OpenRouter API in a similar way? Does the API have a feature to remember previous messages?
thinking mode is something you select in the chat interface with opus 4.5, it's not a separate model, it will still have thinking
it will give similar output because arena uses the same apis openrouter does
openrouter has its own chat interface that you can use your credits in
I usually send 100-150k characters of C++ code and expect a response.
Opus 4.6 and "opus-4-5-20251101-thinking-32k" on arena.ai did this very well. I hope the OpenRouter web interface supports this.
it should, i don't see any reason why it wouldn't
Sometimes change is good. You never know, you might find yourself even stronger than before.
Opus 4.6 on arena.ai was legendary and great.
Opus 4.6 on antigravity is terrible.
Could you please check if there's a reasoning feature for Opus 4.5 in OpenRouter?
I considered buying it from Claude's own website, but people have left very negative reviews. Even for $25, they only allow you to ask 3 questions and get answers within 5 hours. They say it's bad.
I think what they're offering isn't the API, but the Pro membership.
I think the API is better than the membership.
I don't know much about it either, I'm new and don't understand much about these things.
Thank you all for giving me the opportunity to speak on this platform.
did you use opus 4.6 or opus 4.6 thinking
yeah anthropic doesnt have a good reputation on rate limits xd
i don't use claude myself, but i imagine you'll get similar value in the api as you would if you bought claude pro, with the extra value that you're not hooked to a sub and there's no rate limit, just be careful not to spend all of your credits
Well This issue is getting very complicated, if the Arena team makes a wrong move in this situation, I don't know what will happen next... I hope the Arena development team will soon find a mutually beneficial solution with users that use Arena.ai xd thanks to @echo aurora and other's developer for the hard work onto Arena.ai
Yes, we think alike. Thank you for your answer.
I would like to thank Arena.ai for everything.
It provides a great testing environment for users.
And a powerful and user-friendly platform.
I hope they develop their own AI someday.
they kinda already did make their own ai a very long time ago (when it was still part of lmsys), but it's so obsolete now nobody really cares about it
https://huggingface.co/lmsys/vicuna-13b-v1.3
idk if this counts much, but it's the closest arena ever had to its own ai model
Wasn't Vicuna an already existing model?
also the "demo" link i think just takes you to arena now
this was back when arena was part of lmsys in its early days, it was as much of a pet project as this was
it's based on llama 2, but it's a finetune
yeah that's what I meant
it's not a 100% fully custom model
true (i think arena itself made some customized models once, let me find those)
I tried to do something like that once, I ended up re-creating nanogpt.. it's very hard to make one from scratch that's actually good
also a finetune (of qwen2 or qwen2.5 i think), but actually made by arena
https://huggingface.co/lmarena-ai/p2l-7b-grk-01112025
I'm going to build a Transformer, embedding project from scratch. I'll partially code it to learn neural networks using artificial intelligence, without using libraries like Torch, TensorFlow, or Pandas.
With 50GB of book data. It should be enough if it can learn grammar rules and respond to everyday sentences.
best AI for editing and writing draft assignments?
probably claude or gemini
any specific models
cant use opus or 3.1 anymore so just use claude sonnet 4-6 or gemini 3 flash
I guess the issue is not where to find the training data, but rather the computational power needed to train the model on it
I'm not rich, I can't afford to buy an H100
maybe I could rent it, but still..
is there a paid version of this platform?
No
damn
but hey, you can donate to this platform if you want it to
Muse Spark has this huge DNA from ChatGPT. It replicates even the habitual fallacy (no scotsman) and misreading. Almost like a copy, tbh 🤣
Check it > https://meta.ai/share/Kl4GsILvxAc
Woaaa what site is this
Isn't this the ai from Facebook
The new model from Meta, yes. It's available in their site and app, but probably not directly in their social media yet
what website is this
Hi
doing hard prompts again and again 😭
/image to vide o
oh,claude opus models
gpt 5.4, gpt-5.4-high
gemini-3.1-pro-preview
oh,claude opus models
gpt 5.4, gpt-5.4-high
gemini-3.1-pro-preview,
I'll wait for you to come back!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
The heck why im got empty response?
it's kinda glitchy recently
did they say a reason why they removed all the best models?
最近是不是把gemini 3 pro和opu 4.6都移除了
I think the cost for them. And to make them better to use without so many errors as they had
Find a way to make them more sustainable
Yeah there are going to be limitations and restrictions in place to ensure we’re well setup for reliability and availability sake. It’s our intent to bring those models back to direct and side by side in a sustainable way.
i wish that there would be an eta though because it just seems like it wont be available forever without one.
i still expect it to be here again by the second half of the year
Yeah sorry to say we don’t have one to share
even if it doesnt come this month i still feel like if it was to be announced it would be announced here and probably next month or june
Overall we’re going to avoid giving expected dates unless we’re very confident in that date. Shifting priorities, unexpected variables, etc just make it too difficult to say at this point
true. the credit system though needs some adjustments definitely, but with it i think frontier models will come back faster. again people were saying that ads could help in monetization of this website
Be sure to let us know your thoughts in the thread (if you haven’t already)
whoever is this Assistant B. it's good
somehow the model just didn't appear, but that's fine to me 😊
We’re looking into the bug where it’s not displaying properly. Can I have the Eval ID for this? It’s the random set of numbers and letters in the url
this thing?
Yeah
here's the code hehe: "019d70ce-38c8-7a73-a249-843246e5e476"
it's just one request for my favourite prompts of all time (which is asking for LaTex) or creating PolSci Scenarios (What If)
@echo aurora when is the arena gonna have an ai music arena session?
Not sure, couldn’t say
are the models back or are they gone for good?
It is said that the GLM-5.1 AI model has surpassed Sonnet 4.6 — is that true? Has anyone used it? Please share your thoughts.
they're still in Battle
but not in Direct or neither Side by Side
BENCHMAXXING
basically every chinese llm is benchmaxxed
no exceptions
uh oh
I've been out for days
back in arena and found claude opus 4.6 and gemini 3 are both gone
but they are still on the leaderboards
is there an document arena update, so they removed them temporarily
or just permanently
perm
they're gonna return it once it's ok
after testing that thing, it doesn't even produce a good latex at all'
like its treatment of LaTex is non existent. it just blank
Had it with Gemini previously and then it was removed
..what the i used max 1 time but the
{"error":"Chosen Model(s) are no longer available"}
WHAT THE HECK???
Wheres is opus and opus thinkking and gpt 5.4 with files?
theres no more opus lmao
why?
Hi, I’m experiencing an issue with a long conversation on Google Gemini. Today, when I send a message, it keeps loading for more than 30 minutes without any response. I’ve tried refreshing the page, logging out and back in, using a VPN, and even switching to another device, but the problem still persists. Could you please check and help resolve this issue? Thank you.
Yeah abd fix rate limit and the empty response in searching model
i hope they do come back
Is make me so mad 🙂
we dont even have gemini 3 pro
didn't Google themselves removed it?
PLEASE tell us what website this is
thnx for the clarification
so i need to start new chat ?
i think they did but lmarena had it for a while
Yeah
If you refersh the model back to issue infnite genration
ok np thnx
Arena had it before Google deprecating Gemini 3 Pro, not the other way around
i could have sworn the arena had gemini 3 pro even after google ai studio ditched it, but alright
I mean, Google removing it from AI Studio and Google disabling the API are different things after all
any got imagine v2 acces in arena?
CAN SOMEONE RECOMMEND ME AN FREE AI WEB THAT HAS A COMMON SENSE, GOOD MEMORIES AND CREATIVE WRITING AND RESEARCH 😭😭😭😭
it's openrouter chat, which is sadly paid
5.4 have codex? Wtf
What was that wbsite or interface? are you use local? base on that photo?
That's normal 5.4...
whoever is that Assistant B. i would love to know it and say thank you
reklam
Guys where are the opus gemini gpt 5.4 models? Why they got deleted
Yes
Sorry bro
Is this gonna stay? No plan to bring them back?
Cause the model can't pay anymore
They finding way read in #announcements
O
Tnx man that was great while it lasted!
Thanks everybody who contributed to this
sadly opuds also
they were too expensive, so they took em away
you can use gemini 3 flash if you want
flash for what purpose? it can't even do my request...
i use llms for making up stories for me, that are enterntaining, so flash works well for me
well it's not fault of arena, these models are expensive to provide. Moreover the purpose of arena was to test these models, and provide accurate benchmarks without any bias.
But people started abusing it by creating new accounts and using same model in direct chat over and over again
it was eventually gonna happen
technically my use is not for that...
there's seven ais (reduced to two right now) who can do my request (a very long hard prompts that i do a lot)
Well
Can flash translate in perfect Albanian for an Srt file I don't think so
yes it absolutely can
honestly none of the current models can perfectly translate a language to another, imo, it's very close tho, if we remove the gaurdrails it's gonna get almost perfect. There's a reason why experts say gaurdrails are holding a model back.
If the language you are translating consists some swearing words or words that might get flagged by gaurdrails then Ai is not good at translating them hence resulting in a not perfect translation.
I'd advice if you know the material you are translating contains some words that might get flagged then its better train an agent yourself or eh grok works better than most models at this sadly
4.5 flash good enough to translate
5.5 flash is better
I edit the srt afterwards but i just want less work you know ?
pineapple got 4.5 million dollars by Gemini 4.5 cuz he is now a pineapple juice
3 days left till pineapple juice goes
then it's good
js use mythos Thinking dude
Claude mythos Thinking
This one
or Claude Pineapple Ultra could work too.
is that real
In parallel universe, yes.
wow!
where to get it
I am actually starting to get so annoyed why is verification happening every time I tried to do a retry😭
there must be a special place in hell ...
using uncensoreddns to lmarena don't give me verification (i first time see him but returning... he disappeared)
..WTH???
we all know this is f12 edit
Even without vision, too weak 🤧
how do you do that
press f12
wait....arena can edit with CHROME DEV TOOLS?
magic isn't real
everything is editble
what about ai system?
what about limit token?
wait what is F12 ?
developer tools
i can do like
what about removing recaptcha (worst captcha system)?
How to unlimited video create
some stuff you obviously can't
No need to thank me, guys
f12 changes aren't that hard tho, here is an example
-# he didn't said this ANY time
stuck on this horrible system i guess
recaptcha is sinful
....wait that's not a arena site...
i haev a doutbt where i can see archieve chats
you can't
pressed f12 to see sources
and one more thing is claude opaus is removed ?
was too expensive
yes
wait what does tools uses for ai models (lmarena)
But still it's dumb!
Just say you're broke and follow openrouter 50rpd.
what
It says snapshot from 1970
here's recaptcha one:
they collecting:
IP addresses - tracks your location
Mouse movements and clicks - behavioral profiling
Device information - OS, browser type, screen resolution
Cookies - persistent tracking across sessions
Time spent on pages
Browser fingerprints - screen size, resolution, language, plugins, JavaScript objects
Keystroke dynamics - timing and patterns of keyboard input
Screenshots of browser windows - Google literally photographs your browser
All data transferred to U.S. servers
and reCAPTCHA Violates GDPR
that's why you guys getting asked with
That's the deal for arena
I don't see anything bad in it
Well at least someone will watch me
i wonder who is watching you
No one rn
Except cucumber on the table
It watches me
Straight in my head
lol
seems like what Russia/US/China would do
reminds me of backstate apps (like MAX)
RuNet and China ROM for example. does the same stuff like you said
although i use Yandex if Google Images is not working
ah yes recaptcha collecting same stuffs as runet and china rom
how awful
Max is not a virus
It's just a contact with the government
what?
it's released publicly
I know
yandex is bloated so that's why i use only yandex alisa
You can contact russian government with it
my colleagues doesn't even know Yandex, like i'm probably the only one in my class (that is dumb for the most part (Mathematics/Physics) that knows random stuff like that
pineapple answered me with "This area isn't my expertise so I'm unable to provide specifics, but overall will say our team is looking closely at this system and are actively making adjustments overtime to this system."
Yandex is very popular in Russia
It's basically everywhere
not in PH
when teams need to take actions about this tbh
not PH the site, the country
polyhood?
some Country with the highest Tax in SEA
treating itself like a douchebag
Philippines
yep, it's a douchebag country after all
if S.D wins, well that's good for the dumb ones
when teams picked google login where recaptcha: problems
It's me or Flux 2 never actually work XD
GLM doesn't work too
it just gave up on me, because it can't do my hard prompt at all
GLM was pretty "decent" for the text to wait for claude to come back
if somebody says that it's better than sonnet 4.6, then think about it, why it can't PRODUCE a good LaTex
I wasn't able to resist and took a month subscription on Anthropic
Arena might lose their large community if they don't bring top models back(at least Gemini 3.1 pro or gpt 5.4)
Nah, id learn code than burning my brain with ai enforced code 💀
it can't even do Text Coding.
I don't need that..., i need it for Text Coding, specifically for LaTex
id prefer not to learn code than burning my brain with learning code for years
like i already want to be composer
it's okay ! I was just saying that I'm using it for stories. You can't use it for code on the website of anthropic ?
it becomes limited sadly
how is my composing?
I was surprised by the limitations on Anthropic too
I mean, I'm paying and I can't use it as much as I was using it on Lmarena when it was there
oh hell nah, not my stool erupting again
if only Gemini 3.1 pro preview have File and Video File Upload in Arena, i would have been transcribing for a long time
sadly it's only file upload
for it's existence
except for AIstudio
when
turns into 
i recommend using deep seek
of you want an good AI for coding
i asked ai to suggest an idea to keep arena.ai sustainable, lol
lol
not sure is it a good idea, or a dumb idea
it was provided by claude :)
are there any good alternatives to arena? where opus 4.6 is available with the same limits as here 😖
sadly no
actually yupp
yupp is stopping
oh
probably because they were offering free opus 4.6 lmao
still wondering how they can offering all the (top) models for free
data to train the AI
I guess
they want us to evaluate which model deserve to be better
Yupp is already stopped.
how does lmaerena benefit from direct chat
they may contact me too haha
are people suppose to use the 👍 and 👎 buttons for each message
does anyone even do that
yes
I did
i occasionally do
ppl also provide the original data to train
Whats the reason opus is gone?
is jt gone again
opus is toooo expensive
too expensive and many people using it
get it
they have to cope the cost
Yh thats why they doing a credits system
did it get returned yet
is it effective now?
it was a monster model
Its gonna happen tho
i thought
In like the future
now it gone for me again
im using deep seek rn
im so scared,,,
he is saving the day
cool
I felt like Cooper from Interstellar watching you guys argue yesterday about AI, and I couldn’t say anything because I got a 24-hour punishment for speaking Portuguese.
then we need to work together to figure it out what to do to ease it
deepseek is also a monster
they added this new option and im using it
my own advice is to impose stricter limit, stopping bots.
What option
this instant and expert feature
can't u just intercept lmarena api request and replace the modelAId with claude opus's id?
is expert paid?
if anyone saved claude opus model id on lmarena
more free than opus
free
💔
wait
I asked deepseek to splits a CSS 4000+ lines file into 10 seperate css files for each function, did it perfectly
that's why arena.ai art and coding contest/event is gone, because they are dealing with budget
gotta ask him for speed up this thing
The best way to make this viable is to assign a point cost to every model. These points would be earned by using battle mode.
Instead of just choosing which model is better, users would also need to provide a reason for their choice. That reason could then be verified by a smaller model.
For example, something like a 7B Mistral could check whether the reasoning is actually valid and reflects what the user experienced with the model’s response. This would make the system more reliable and reduce meaningless votes just to farm points.
i think battle and liking/disliking for earn credits is fair
battle should be a bit more tho
The problem the admin mentioned earlier was how to validate the votes. By requiring a reason, and having that reason verified by a model, this issue would already be addressed.
you could earn daily points also like 100 points to use
yeh
lemme try deepseek
i gave him a 120p pdf, 5MB to analyze
still obessive with opus hmm guy
how much time would it cost with expert mode
pretty quick, almost instantly reply
well it's done in 1 min
this is a thesis i've been reviewed for days
compared to my paid claude opus 4.6, deepseek expert is more merciful
Has anyone managed to use the Meta Spark model? I tried it yesterday, but it would start and then crash midway with an error.
basically yupp. ai
The difference is that lmarena completely dominated yupp.ai and is now the only one in the market.
🤓
@abstract hinge signed up for yup then they closed
😨
Guys, GPT IMAGE 2.0 will be released tonight after GPT image 1.5
Release gpt image 1.0
opus 4.6 understandable, then gpt 5.4 high, and now Gemini 3.1 pro all top models disappearing.. how much message per day is their limit before they disappeared guys?
why models are disappearing guys ..
Most likely price. And to find out how to make them work better
I would love to thank assistant B, whoever the llm/ai is
when will top model(gpt 5.4, gemini 3.1 and opus 4.6) comeback again?
although it looks like GPT 5.4 High
since 5.4 High gave me that kind before in the past
please we need them
well it's not fault of arena, these models are expensive to provide. Moreover the purpose of arena was to test these models, and provide accurate benchmarks without any bias.
But people started abusing it by creating new accounts and using same model in direct chat over and over again
It was eventually gonna happen
Hi
source? Pretty hype if true ngl
but yeah gpt image 1.5 is already out
and is kinda ass ngl
whats the best cheap or free model? i feel like that is the meta now, all paid models are pretty much the same now
Your reviewer looks so cool, do you upload your notes first, or do you let the ai do everything? :000 I wanna have ai reviewers for my cets TT
it's good at some and ass at some
when it comes to being realistic af it's ahh
it's so obvious to spot a gpt generated image
yeah its realism is bad but also when it does digital art style it has this wonky weirdly vibrant art style that I don't like
Many users on X are trying it, just do a search
thanks for the source :D
I know some people have access to it in chatgpt right now but I don't
and I don't use twitter, not even before elon musk bought it, and now I'm even less likely to
onb
can't do pixelated or anime properly
let's goooo
can any model do good pixel art yet?
gemini mad good at it icl
Here are some images leaked on Twitter of GPT IMAGE 2
not good enough but really good
mfs alr on that good shi
#ai-creations message I posted a few of my own gpt image 2 generations starting around here, during the brief few hours it was available on arena
I hope it will be available on arena
i thought we solved image gen, why should we care about a new image model lol?
AI is scary, now I don't know what's real
its not solved, every time you think "oh it can't get any better" it gets better
but by a small amount
hell even gpt image 2 screws up background details a bit
Omg
How can you generate videos now ?
It only generates Battle mode for me
basically what i transcribe, i pass it to the AI (to Opus itself, when it existed)
and then i just ask, a very hard prompt, and what's the specific structure of the thing
because detect ip. ykwim
do you do quizzes too? coz I often study with asking for quizzes to da ai
damn why disable multiaccount usage?
Claude 4.6 models? Aren't there only one (4.6) and then 4.5 of sonnet
nopee
only haiku is on 4.5
sonnet and opus are 4.6
basically it's LaTex (although i know how to do LaTex, i'm just so lazy to code for 7 hours then having no results
if you're gonna do a LaTex reviewer, don't do it under LuaLatex or Xelatex, because it's gonna crash
you're gonna do PDFLaTEX
if a credit system should be hourly. (in the new usage system feedback eme eme) i would disagree, but let's say 12 hours.
or 16 hours lol
https://gemini.google.com/share/23f28095f68b play my funny game, flip and do 360s to put your combo meter up, and land straight to dont trip
that's like tomorrow already too
created with gemini canvas
Why don't I have claude-opus-4-6-thinking...do you know other alternatives where there is claude-opus-4-6-thinking?
This what I generated with image v2 before it got took off arena, my prompt was simple
Hey why cant y'all add a donation system?
cant believe image v2 recommended dr disrespect
who makes image v2? I wanna play with this
Chatgpt and it’s only available to select users
But they saying it’s coming out today at 1 pm
Really??? Share prompt
What is image V2 is it available at arena??
comeback of the century if true, google's nanobannana has been the best by far for so long
A twitch screenshot of a semi popular dude, he just got a trickshot in modded cod server
Be super detailed and very specific and very realistic
god damn
1pm what time zone?
in 2 hours?
Eastern, 10 am PT
???
nice
not yet. it was briefly as a codenamed model on the weekend
can't wait for it to get lobotomised 2hr after launch
I mean hopefully not
Yeah but this is openAI we're talking about
They always just let it fly with no guardrails for a while to build hype then restrict it so much it's nearly unusable
Basically what they did with sora
That’s why I’m testing it strictly on arena
Also here’s what I got with nb2
lemme try this with 3pro rq
they didn't do that with gpt image 1.5 I don't think. (That one was just bad on release)
also they actually dialed back the copyright restricions on gpt image 1 after a while
good point
Yup multiple sources saying this
Yaah i am doing the same
So is it coming out tonight?
If you're doing webdev, sonnet is pretty damn good
Nah i do like desktop apps and stuff
language?
Could be, I don’t know anything about other image releases and if they had a rollout or a surprise launch
Bro what you guys think is gpt iv2 will be lauch on arena ??
3 pro output
C++, Rust, C so like low level languages
better than nb2 but the text isn't perfect and it put the cam in the wrong spot
I know that qwen 3.6 plus is reallyy gosh darn good
It’s decent, but there’s inconsistencies like the webcam and the website in general
Bro 3 pro just did a mistake 😭😭😭
beautiful
Qwen & Gemini 3 Flash are available on Arena and seem to do alr for that
TOABS
It's having detail
are we not going to talk about the fact that he's literally the pog emote
This image v2 nd I tried to create a aesthetic photo
🤣
reference
Let me give one more try
all we need now is this kinda performance paired with the new Google research to get a local version of this that could run on ~20gb vram
yep i think NB has officially been dethroned
perfection
No it's absurd GPT image 2
I have to make a fake girlfriend and make people believe I have one hahahah
Bro no one can beat it
can't handle this more bro