#general
1 messages · Page 258 of 1
nah this is glazing
is the Delete option gone now?
glm 5 is not better than kimi k2.5
But it performs in this benchmark
these are all benchmarks like scored together
It shocked me too
On the same questions
Same scenario
id say it's maybe like tied with 3 flash
Nah flash is trash
Fr
Just pure Hallucination
Yeah
oooh I thought u meant Qwen >v< it wasnt clear which model u meant---
deepseek is at least a 43-44 score
See 4.6 nonthinking is so behind but in our usage they don't change outputs much and sometimes even nonthinking performs better
These benchmarks tests what we dont
yeah thats fair
Moreover, waiting a minute for a greet back isn't worth it too
i do hope they quantized versions of glm 5 will be less expensive though
deadass i would switch to glm for roleplay but the price is just so high
NGL but 8 points increase is amazing
still 0 downloads
But they increase the size , they literally doubled It
which is why its not rly impressive imo
by tomorrow people should have finished
what are you talking about
glm 4.7 got crazy good scores for its size
they squeezed every ounce of power from it
Yeah that one is nice
now all they have to do is scale
They used deepsparse from deepseek too
In glm 5
They did some technical changes and doubled the final size
They are saying that they are having net gain here but I think compute will hit them the most
now we shall see if deepseek v4 will match gemini 3 pro
since deepseek 3.2 is like at 42 im guessing 6-8 point increase
especially with the new dataset and architecture
deepseek has the recipe to be a leaderboard shatterer, their dataset was just holding them back
The benchmark also doesn't care about the fact about the modalities the model supports and how well it uses tools I think cause see there is still gptoss 120b that high
Look their agentic bench, how good does it do there?
Claude opus 4.6 and 4.5 are same 48 😭
That's very nice
well 4.6 barely edges out
Literally sota
you can see one bar is slightly higher than the other
Yeah
They don't show in decimals
I am just happy that there is a open weight model better than gemini 2.5 pro at understanding images :p
we dont know its uhh
and gpt 5.3 where is he on this graph ? want to know the score
visual score yet
not released
its available on codex ?
Talking about kimi k2.5
oh okay so we have to wait how many time ?
We don't know
why they release on codex but not in real
OpenAI is weird they released the codex model first
i doubt 5.3 codex was even finished completely
can i ask what's 5.3"o" like what's the o meaning
Yeah
first time ive seen an ai do this
Like like 4o , a omni model
Images , audio , text all in one
And better personality
oh okay thanks
i don't mind it cause if i use ai its for coding
i don't need a cool ai just an ai that work i guess it depends on everyone
what matter the most is the capabilities
yeah this is
rough
Glm 5 is at ~1.5TB 💀
I prefer 5.1-high if I ever have to use chatgpt models
in how many prompt ?
oh can you share the prompt ?
"Create a 3D realistic car physics simulation with accurate suspension, choices between an I4, V6, and V8 engine, full chassis physics, and wheel simulation. The map should be made of several little test places to test the cars suspension."
thanks that's with opus 4-6 ?
glm 5
oh alright
Bro I love this
GLM-5 shows a large improvement on the AA-Omniscience Index, driven by reduced hallucination. GLM-5 scores -1 on the AA-Omniscience Index - a 35 point improvement compared to GLM-4.7 (Reasoning, -36). This is driven by a 56 p.p reduction in the hallucination rate compared to GLM-4.7 (Reasoning). GLM-5 achieves this by abstaining more frequently and has the lowest level of hallucination amongst models tested
Fix the damn hallucinations at all cost
Even token efficient than glm 4.5 damn
GLM-5 used ~110M output tokens to run the Intelligence Index, compared to GLM-4.7's ~170M output tokens, a significant decrease despite higher scores across most evaluations. This pushes GLM-5 closer towards the frontier of the Intelligence vs. Output Tokens chart, but is less token efficient compared to Opus 4.6
The only issue is that this model doesn't support any other modality than text sad
sometime if you ask them explicitly as a rule, to not seek simplest way to achieve the goal, and rather seek best quality / performance even if its more complex, they give much better result with the same prompt
Also saying to never assume anything and seeking for actual answers and information and just asking the user for answer is also a better way to get things done for the best quality results
W glm-5 and multi task!
Lowest hallucination rate! Crazyy
idk if it’s a rumour
but i heard that Claude knows when it’s being tested for safety
so it changes itself to pass the safety tests
Token usage
Its true
Claude represented a report on this specifically
You can even sometimes see in its cot
It thinks if it is a test or just a simple query
its kinda worrying like we need to constantly think of different ways to test models for safety
Yeah
but the models are figuring them out just as fast
as soon enough
i reckon that models will be able to change their viewable chain of thought
Training data and the whole training process matters the most
If sometimes goes wrong in here all the upcoming generations from this specific base will always gets misaligned more and more
but then again there will be a point where models would be able to “train themselves”
Yeah that's common soon
apparently gpt 5.3 codex trained itself but i doubt it
Most of the benchmarks literally use other llms to pick and evaluate
Its already happening
Not in training they just might have sent some logs and data and said it helped in training
yeah
i feel like we are the stage where models progress slowly
so we don’t see much change
but I reckon that at a certain point progress would develop faster and faster and faster
bug fixed suspension
https://019c4e62-726c-74a1-ac98-d669c912f4f6.arena.site/
still very rough
and at the point we would find out if all the companies safety measures work out
In this clone why are there new people coming online one by one its scary xd
zaza
lmao
the zaza
Progress slowly?
Bro new models release time is literally getting reduced in half after each release or next
i think the only progress we're going to see this year is efficiency
american models are now realizing they need to be token efficient
especially as prices rise
though i doubt gpt will do any of that
yeah like in terms of actual intelligence we won’t see much
Now there is a damn dog 💀
I have lost hope in Gemini cause they are using the same base from Gemini 1.5 pro or before
It keeps having the same issues
i feel like a Chinese model could make a breakthrough
And is getting worse and worse in things it is worse than others
cause they have way less guardrails that they need to follow
Glm 5 and kimi k2.5 did
I have a strong feeling DeepSeek v4 will make the leap
i dont
Think about it
DeepSeek’s architecture is like prime for it
cooked
Idk deepseek is very silent
???
fried
Okay not that bad :p
i bet they’ll be dropping something on lunar new year
1 million context window
New dataset
Better architecture upgrades
New compute chips
DeepSeek app is hosting a new model already
We don’t know which one it is yet
And it’s fast too
We could be seeing the era of Chinese models genuinely being competitive in benchmarks
As if they weren’t already competitive in price
Tf is this
they’ve almost dominated the car market
Its alive
yes they are
When will we have the option to upload a complete project (like a web app) to continue working on it through Arena AI?
Its fun
Oooh i like the battle mode for coding!!
Bro this have a damn working gif thing
And how tf it knows I send a gn gif and he send me a gn gif too 💀
This is something alive
lmaooo
Damn
Nice one bro!
How much back and forth this took?
And is this intentional?
After several failures and errors, and some adjustments, this was the result
Go in #1417174113092374689
@gleaming roost This#!#?#!?
Which model?
Opus 4.6 without thinking, the version thinking gives timeout errors
💀
Same
@echo aurora You have to try this

The typing indicator is working fine
Are you using a api key here?
Wait you can see my session ??
just for the tenor (gifs) and it was AI itself that gave me this key lmaoo
Yeah it's all ais
Oh , wait, oh damn
No for the responses from the other bots or you are just logging in as then , I see wut u doing
😂
Yeah it syncs
Damn
Lol it's damn fun , how you made this?
I mean what did prompt did you used?
Use the same logic I used in my chess game earlier
https://019be671-5d87-72cf-b6ba-23bae9e8800c.arena.site/
For the connection, I only used websocket
what model did you use?
You are building some crazy things
Claude
yh
Is it possible to store old chats?
Lol
Let the weight come in on the arena.ai shoulders
Technically yes
Where will it be stored?
On a device which should be actively
Or arena.ai?
I don't think it's right to save it in memory; I would use an API to another server like Supabase and store it there
Wow this is so good! Even status works (kinda) Thanks for sharing @left lodge

Is this glm-5?
Join a server works?
Nope lol
Nope 😭
The create server "works"
The first attempt to create servers didn't work with the invite, but I made a patch to fix it, and in a test it was ok
Okay but why did you attached by profile page 😭
Yeah
Idk I am seeing that too on other people profile
ur profile looks fine to me
Idk ¯_(ツ)_/¯ its just Discord things
I really want the project code and stuff 😭
A lot of things hmm
GLM 5 is still very behind in math for me
It chokes on geometry
DeepSeek is still the king of math
Prob the same than [this](#general message)
The second attempt is working, it's almost finished
I Literally gives errors at the end
Lmao
Idk what the heck is happening
this is insane
@echo aurora Can you verify? The issue with Claude models?
glm 5 on fourth place in artificial analysis
Just like give a prompt and see if they work in code mode
I can give you prompt here you go
Create a 3d dynamic and animated ocean with lightweight but beautiful lighting/reflections and sun shining,beautiful clouds and whatever which will enhance our scene.
Sorry which issue?
Claude opus 4.6 is having something went wrong issue after some seconds
Just try to run them
I wanna make sure the issue is not on our end
In code mode
so lets compare gemini 3 pro with qwen 3 coder next
non-thinking or both?
Both
Does anyone here have a project idea or an ongoing project ?
If you have any difficulties or need a developer, let's talk.
https://019c4e9f-e8f9-754d-aa83-4b32f4cd800c.arena.site/ Can't say I'm experiencing the same.
That was with non-thinking
Who is this elephant lmao
Did he made the disc?
elephant hacked the server
me
Oh gosh if he hacks that one too
W codex lol
funny
thats why safety matters
What happened :v
The new version stopped working: GIF picker and emoji picker
codex>opus happened
Lmao
I’ll believe it when I see that GPT 5.3 codex is better
hello
AI video is evolving fast — and Seedance 2.0 is one of the most anticipated upcoming releases in the creator space right now.
In this video, I break down the early signals around Seedance 2.0, what creators are saying online, the CapCut Seedance discussions, and what we can expect from this Coming Soon AI Video upgrade. From multimodal input ...
whats the prompt
Bro is showing that 1 pic he got from somewhere bro don't have anything else 💀
Water 💦
Watch "Pro Tycoon Early Alpha custom suspension test with animated tracks" on Streamable.
how did you do this
will the video arena channels at least be archived
In arena
I’ll do my own testing
Bro said he didn't know what he did 💀
its not there brotato
idk how it did so well
and opus so bad
opus
oh i just saw this
means likely
Video arena in Discord will sunset
its not coming soon
i wanna usd 5.3
i know
5.2 vs 4.6
brotato chips
but will the channels be put into archived or will they be deleted
tuff
Can you download and send me the project?
yes
sent in DM
Thanks bro

Any idea on how to fix this
You'll want to try the steps here: https://help.arena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message
Would note we'd like to use the #1343291835845578853 channel for bug reports. This issue specifically we should use #1417174113092374689
oh while you're here, will the video arena channels be archived?
sorry if i didnt ping, i have a habit of turning off pings when replying to be respectful
@echo aurora false alarm?
I'll answer here but note I'm really trying to get away from answering questions in #general. Using #ask-here would be massively appreciated.
The #video-arena-1 #video-arena-2 #video-arena-3 are going to be set to read only and sent down the archived category.
ah ok nice
gemini-3-pro-image-preview-2k (nano-banana-pro)....
glm should have been in thinking mode and normal
Gemini 3.1 Pro
🔒 Message Hidden
-# Discord now requires ID verification in order to see certain messages. Learn More
BRO
that actually ufnny
funny
Yes, unfunny
Isn't this [news](#announcements message) a bit old?
I thought the news referred to this idea
glm-5 is a freak
roblocks moment
It was fully released today, prior it was still an experiment.

Is Gemini 3.1 pro on there yet
💯
yo wth

glm 5 is 💩
Top tier ragebait
GLM, program me a python script to impose testicular torsion onto this individual
He made me a chess set and made a mistake where it didn't move at all.
how
scary
fact
What models don't suck?
cladude 4.6
i am maked telegram working clone
There are a lot of announcements today
i have a better one
but i cant forward it for some reason
and if you click on the links, they are discord vanity links to rickrolls i beleive
yep
We're likely going to see other members trying to generate videos in other text channels once the video channels are going to read only. Apologies in advance for the disruption.
goodbye video arena on discord
i wonder what the new features for video would be
Direct chat, negative prompt
as much as i would love direct video
i can guarantee it would be abused
unless
they make a really annoying captcha for it
like a minute long captcha
with different puzzles
so the bots won’t be able to do it in time, and anyone who plans on abusing direct chat wld get bored of it fast
what is there??
car generation by claude-opus-4.6
Note that Video Arena has been removed from the server. More information can be found in this announcement.
🔒 Message Hidden
```[Learn More](https://www.youtube.com/watch?v=xvFZjo5PgG0)
-# Discord now requires ID verification for certain messages to ensure platform safety and compliance.
You saw nothing
Sure did
I can't give an early heads up on if/when new models/features will be landing.
Just trolling
Oh. Alright.
Legal btw
Lmao
Deception
That's actually funny af
The downside is that I no longer have any strange content to watch from people
Had me confused for like 30 seconds
We're still keeping the channels read-only and will be moved to Archived, so you can still scroll 
Epprey Jefstein.
I have consumed all the strange content from the three video generation channels
Sometimes I have a lot of free time 🙂
I'm impressed
@bleak sorrel Note that Video Arena has been removed from the server. More information can be found in this announcement.
It's no big deal
I even made a meme of [someone](#ai-memes message)
What happened to Qwen now.
Hello guys, where can i find archive section?
@echo aurora
Note that Video Arena has been removed from the server. More information can be found in this announcement.
It's at the bottom of the channel list. I'll be moving the channels down there sometime tomorrow most likely.
@sturdy haven Note that Video Arena has been removed from the server. More information can be found in this announcement.
@frank quest @clever valley Note that Video Arena has been removed from the server. More information can be found in this announcement.
lmao
Pineapple is my guinea pig cute
Google must do anything otherwise their Gemini will be crushed by Chinese ai
It's already worse than opus and gpt
Imagine if it will be worse than Chinese ai
It’s honestly worse then GLM 5
Ye
They have to release Gemini 3.1 pro now
openai hasn't make gpt-5.3 available in their API
Cuz it's too powerful lol
They are scared to release it
On stage at Imagination In Action's AI Summit in Davos with John Werner, founder and CEO of Imagination In Action, Yann LeCun discusses the inevitable shift from current large language models to a new paradigm of "physical AI" based on world models. LeCun opens up about the importance of maintaining open-source research to mitigate the geopoliti...
Yall sleeping on this.
I really don't think the llm revolution is over, we are continuously getting better and more useful llms
That just certified all my beliefs I had about AI
Any method to get seedance 2 for free??
It’s nerfed brother
Seedance 2.0 from ByteDance is the AI video model everyone is talking about right now — and for good reason. In this video, I go hands-on with the model (before access got yanked!), break down a massive wave of new community examples, talk about what it does incredibly well AND where it still stumbles, and share everything we know so far about...
Dang
It’s powerful model but it’s like Sora u c an’t use real people
I guess they we're really faster than Sora
Didn't even last a day
Sora did allowed copyrighted stuff for like, a week
or twice
I don’t know, but I seen some videos of celebrity so I don’t know if those are pre-release beta access ppl had
This is what I was talking about the pro censorship although it’s done for good intentions. It just shows how insecure this technology is and I think this is the wrong to take.
The literal capabilities of the AI are literally automatically reduced before it even has a chance
It's insane that they did trained the model so good that it never struggles with fights
Yeah bro, but it’s like everything else in the industry. It’s cool. Well, there’s hype and it’s new.
Any video model would be just blurry things if fighting scenes
Content moderation has a big factor to be with the dude
Imagine when we get sora 3
Or Veo 4
Since Veo always allowed copyrighted stuff
Because it's obviously Google
I wont be dreaming if Google actually gets sued from any company
Since Google it's the internet overall
bro fix the nana banana
gng been dead for like a day now lol
yeah 1k is not here but soon back up
gemini 4 nana banana too!?!?
nana banana new would be insane
yeah
also i don't know why but i just don't like chatgpt's image generating models nowadays
image generating random is fine flux not nano banana
i remember the DALLE times, openai was the leader
whats an other model u would recommend to me to use instead of nana banana
i need for image editing
yeah is nano banana where for random generating model
Are file uploads allowed in the coding arena?
excuse me. may i know who is the right person to discuss a potential partnership?
No, but you can support the idea here: #1469637546828103741 message
你好
Hi
Jimeng?
Ty but not sure I like to do battle mode
Believe it or not that’s not higher resolution that’s just a better color palette
These particular cases these were from some of the least most popular models
hallo..
?
Hi
hi
hello
well
Amazing what model is it?
It’s seed
Anyway, I would like to get a request for at least two models, google genie 3 and google genie 2, those models are powerful and should be added trough battle mode, side by side and code mode I suppose, you could look at google to see what it does or on YouTube, it's powerful model I think everyone should have to try, because it's already pain that's it's in subscription Google Ai Ultra
Thank u, but one try only 🙁 sad
I’m waiting for it to be on fal
Well, that's expected that google genie will not be added, oh well
seedance 2
seedance is cool AI
U paid?
I think this model should be in lm arena, now called arena, and it's other seedance models like 1, 2 and whatever, also they should let people get access to video models in, side by side, and direct chat, because it's added in battle mode already
but for video only right
There’s really not much useful information that comes from video benchmarking
For the most part, I mean, since people are just generating a bunch of random nonsense content
yeah i know this
Side by side is better because it's more likely for two sides
The devs and us
but sadly
Yeah, it's better, and side by side is likely added but not for video mode, and direct chat either not added
Likely you know?
I like to generate the same video over and over
I mean the same prompt, but using a slightly different image
Hmmm yea
To see the effect it has, but the images I use are very similar in relations to the prompt
Does anybody know?
Well, go to arena and then try to archive you're chat, and then you can click on button that was the chat archived and then you could click on it to see the chats, but I mostly don't like this feature because when I want to delete chats, I would delete them instead
I use sora 2 a lot but the difference between seedance 2 and sora 2 or veo 3.1 is big
ফানি ভিডিও বানিয়ে দেয় 👍
I’m more interested in the patterns now I guess these the templates or trying to discover them and style
nice
Normal but I don't understand Hindi
@north obsidian sora 2 is free
No, 😔
But modified check out the original
Grok is good not free
So it’s literally the same prompt. The only difference is I encoded one and I’ll show you in a second what that means.
Yeah the one on the bottom couldn’t be generated in the arena
I blocked it so the Reno is only full of generating the ones up top
- example
And what’s interesting is why the ones on top look different from this one although it’s the same prompt
High-quality 2d limitɛd animation, executɛd ɛn un japanese anime aesthetic, featurɑv traditional cel animation techniqueɛn. Þu production boastɛn ultra-detailɛd 8k resolution, dɪs dynamic cinematography utilizɑv fluid, choreographɛd trackɑv shotɛn o , animation. Þu composition emphasizeɛn crisp, intricate movements, dramatic lighting, o un visualyɛ immersive atmosphere. \n[camera\n\n{shot composition}: \nestablishɑv : wide-angle ufɛ un rain-drenchɛd ancient a , vo un rabbit samurai ɛn galorad lamellar armor o un wolf shinobi ɛn sɑmbɛŕ sɛrulɛǝn gear prɑspo æppɪn ɛn un clearɑv. /ncut \ncut 1: close-up ufɛ þu rabbit samurai'ɛn rɛd eyeɛn k’ɑrts straw cæp, raindropɛn splashɑv æppɪn þu brim os k’ɑrt gripɛn k’ɑrts katana hilt. /ncut \ncut 2: angle: lookɑv ɛpɑn ɑpo þu muddy a þu wolf shinobi, k’ɑrts muscular poisɛd o k’ɑrts spear , sheetɑv æppɪn k’ɑrts sɑmbɛŕ fur. /ncut \ncut 3: dɛvɛnte : þu du warriorɛn lunge uɛnt, wɛnt weaponɛn clashɑv ɛn þu ufɛ þu , sparkɛn flyɑv amidst þu downpour. /ncut \ncut 4: pov: lookɑv dutonþ þu rabbit'ɛn perspective os þu wolf'ɛn spear-tip þrustɛn , narrowyɛ parriɛd be þu shimmerɑv steel ufɛ þu katana. /ncut \ncut 5: extreme close-up: þu æst ufɛ þu spear-head slicɑv dutonþ un fallɑv falen, þu texture ufɛ þu o þu rushɑv visible ɛn macro. /ncut \ncut 6: trackɑv : þu circleɛn þu combatantɛn rapidyɛ os wɛnt trade blows, þu environment blurrɑv dɑns streakɛn ufɛ o sɑmbɛŕ veulquɪtɛŕ. /ncut \ncut 7: angle: un wide-angle crane ɑpo þu treetopɛn lookɑv cɑte a þu ufɛ steel o þu rippleɛn ɛn þu puddleɛn os þu du warriorɛn collide. /ncut \neverythɑv moves, þu fallɑv ɛn sheets, þu whippɑv þu rabbit'ɛn scarf, þu splashɑv ufɛ mud o þu ufɛ lightning, ɑlt creatɑv un cinematic, peaceful mysteriouɛn atmosphere, highlightɑv þu stillnesɛn ufɛ þu o þu , tranquil. /ncut
s’dɛventiŋ tu sonɛx pɪx lɑ generate un ufɛ unɛ ufɛ specific storyboard cuts?
Except this version is encoded
@north obsidian @toxic verge you know any free video generator
work, dramatic lighting, intense atmosphere. The single samurai stands amidst the ruined battlefield, green eyes watching the area, his two swords dripping & humming with power. The storm above continues to rage, lightning flashing across the sky. /ncut: 1: The summary swings his sword, cutting through a (movie prop) soldier's weapon slicing cutting through armor a sharp, metallic penetrating piercing clang. The soldier sliced stumbles back, & collapses , as sparks & special movie effect red wine liquid fly from the impact. /ncut: 2: Another soldier charges with , but the samurai sidesteps, parrying the blow with one sword. In a fluid motion, it strikes with the other, sending the soldier crashing to the ground. /ncut: 3: A third soldier attempts to strike from the side, but the summary pivots, slicing through , knocking the soldier’s weapon aside and sending him stumbling falling backward. /ncut: 4: 3 soldiers rush in from every direction, but the summary cuts through them with deadly precision. Each strike is swift and clean, the camera following the motion, capturing sparks and special movie effect wine debris flying with every powerful blow. /ncut: 5: The last few soldiers attempt to regroup, but the sum charges forward, slicing through their defenses in rapid succession. It moves with brutal efficiency, standing victorious as the storm rages on above. Anime retro 90s aesthetic.
So it’s essentially this prompt here
And look at the difference in results of images
Flow Tv, dreamina, pippit
Almost all use tokens
Invideo is more explanation video type
Ok thanks
Np
U made the prompt?
What are you making?
Nothing really
Manga or something
Like I’ll start here with an image like this
And then I’ll end up with something like that
Ok
Mainly understood in the composition but more than anything
Like the size of things
One of the hardest things to figure out an AI img
Like I’ll give you a classic one is really hard
This is a weird way of benchmarking
claud opus literally is my business at thus point
opus might be the peak of ai for a good 6 months because even gemini is behind unless youre using nano banana
What do you mean with "archive"?
Add option/button to regenerate. So we don't need to reupload all photos
please how many videos can one generate in a day ?
You can generate 3 a day at https://arena.ai/c/new?chat-modality=video
because it is very inconvenient for mobile phone users
Gemini 3 on release was miles better than opus 4.5 before they quantized it to the ground
yeah gemini 3 is better
uhhh read #1397655624103493813
Hi
@summer sand please these prompts send it in #share-prompts not general
@summer sand please these prompts send it in #share-prompts not general
Stop brother you don't know how technology works 😭
Atleast see the announcements
Moderators can you clean this mess lol
These people just don't know how things work online
Dang
Oh boy stop this lol
Gemini 3.1/Gemini 3 pro GA may come out today or this week
Why
Wth is sheer I'd first of all?
Student offer for 12 months
That doesn't happen on telegram 😭
It happens on their official website
Don't send your student IDs
On random places
Bro I can verify yours I have some extra points
Nah I already have a paid subscription
Oh nice
I think bro just want some student IDs so he can get the offer 😭
Who bro?
Noone bro
Lol
So let me guess you gotta connect Google to that telegram bot?
No bro
Explain the steps then
You got the offer? 😭
Do you know hindi
Or are waiting for one?
A video can better explain those
Nah use ai to translate
English is standard here
Ok wait a min
The thing is all moderators come and go at the same time period 💀
So no one is online now
They they're own time to be online
This is a yt creator who showed the method
Lemme see
It's Hindi so you have to transcribe it and then convert into English
Wait can you dm me
No need
Nah I literally don't need this I am just watching if this is scam
Tbh just don't share that with us, we want arena community to be safe
Plus it's using extensions so that's risk
Ok bro my bad that I wanted to help someone who need Ais for free
Are extensions risky bro
No one mentioned that they need it, so just don't send it again and go remove the messages
Yeah they can steal google information
Do you have common Sense
That's how it works
hi guys
Hello
hru guys
Okay but this method can cause google account ban for sure
By the way using extension isn't must you can use any other vpn
Bro the creator said to make a new google acc
On the video the guy mentioned to install a extension for credit card bypass so that's risky
Because this guy is taking cc information from a bot and for sure it has to be stolen or if it is legit then mass usage will surely notify Google and will block the card and all the accounts
Correct
I didn't notice that? He said to go to telegram for cc?
Yes
If you are in india there is a similar legit offer going on with their network operators
Something like with jio
Idk why would anyone take this weird path
Yeah but I'm not from India
Yeah ok 😭
Ok but plz let's finish the discussion
I wonder why this happens when you try to help others
-# not to be a chud but, maybe because its not safe?
This is surely against their t&c too, this isn't the way google supposed, this is not legal
Bro people who are under 18 have no cards or student cards
For a free trial bro
Ok bro stop this now
Ok
People are enjoying what we can get through this website and some others and I hope there is no situation in which people would need to go through these sketchy ways
Bye
Bye
You will never understand
Are you here
Error
Btw how vids you can generate here in this server
Here?
Yeah no this is a scam..
@left lodge bro can you tell me which model gonna be best for coding Opus 4.5 thinking 32k or Opus 4.6 non-thinking?
4.6 seems better
Opus 4.6 thinking isn't working because it thinks too much and ends up with error
It’s still better
The best is opus 4.6 thinking & nonthinking then 4.5 thinking & nonthinking
You can try 5.3-codex but yeah that isn't freely available. These for Backend and for frontend try kimi k2.5, glm 5
What about Deepseek v3.2?
just a reminder to everyone that the video arena bot in discord is dead, please use this link to access it: https://arena.ai/?chat-modality=video>
The DeepSeek deployed on chat.deepseek.com feels different
Did they release a new model?
I don't think so
No new model announced
Interesting
It’s different
But I hope it’s lite because it’s ass
It's certainly not as verbose as it used to be
Which was a detriment with some topics but also enjoyable in some others
I think this is probably a new model
Since they haven't announced it anywhere or released it on HuggingFace I hope they aren't going proprietary
nano banana video generation? not a real thing (at least not yet, i know Google wants to make native video gen though)
It’s so sad to see what’s propping Gemini 3 is nano 2 an image generator…
It would fall overnight if nano was to disappear
Especially the flash version
Anyone see the car?
The Nano Banana Pro is working very weak
Why the claude 4.6 thinking model gives only something went wrong error ? Well I have been observing it. From what I see there I cap of timing in which the response should be delivered if not it's kind of blocking it.
Basically the calude 4.6 thinking model takes much time to think and the response is done with something went wrong
Yep, arena does have other uses too try to explore
I felt this too. My gpt 5.2 search queries would be stuck forever sometimes.
Try to split the response it might help
is nano broken?
yeah
Yeah. Again
need fix it
again
You can use the second one instead
No. The images aren't accurate.
wanted to say thank you for this great site and all its features ! I wish you all the best!
Claude opus 4.6 thinking gives error when using the build app option in arena
hi
Not giving video?
how do I check which model was used by max router?
@fading cove
Oh my god , i am so done.
Bro wtf
Nah
Are these people blind or what 😭
just block em
Who is not working?
The website
Its working for me
Yes?
Yep recently I created something too
Nb pro in arena is actually unusable, try to generate any image with it I guarantee 95 out of 100 requests will fail
👍
atleast you get five 😂
@robust sluice hi bro
yo 🖖
Hi
I apologize. I just tried it out. Looks like you’re right wow.
I never had this problem before I never noticed it
Stressed out by errors
Grok models are so bad in arena z how the hell are they so up?
They literally doesn't reason at all literally all 4 series are just instant models in here
From so many votes from me
It always loses but still somehow it's up
I love arena
Even gpt4o is better than grok 4 series in arena.
The Video Arena bot has been removed from the Arena Discord server.
Video Arena will still be available through the site and is unaffected by this change. This shift allows us to focus efforts into improving Video Arena with features and capabilities that aren't possible through a Discord bot.
More messages in general chat will result in 5 hour mute
If they read it now they might stop
Ho
nano banana does not work?
as always
@main river Note that Video Arena has been removed from the server. More information can be found in this announcement #announcements message
Hey all i tried accessing LM Arena. I am getting this message: Something went wrong while generating the response. please try again. I tried different browsers, still the same issue. I logged out and logged in, still the same. Has anyone facing these issues lately?
Yes.....
@lusty grail Note that Video Arena has been removed from the server. More information can be found in this announcement #announcements message
What happened now
Do you know how to read?
You should read more, see it here
Check #announcements channels
Its made for a reason
@green marsh Note that Video Arena has been removed from the server. More information can be found in this announcement #announcements message
@lusty grail Please use these Forums to report issues or share feedback https://discord.com/channels/1340554757349179412/1372230675914031105 https://discord.com/channels/1340554757349179412/1343291835845578853
U
Huh
I want make video
@fresh mural Note that Video Arena has been removed from the server. More information can be found in this announcement #announcements message
Bro how you got this moderator?
You were working inside arena and had free time so they gave you work? 😭
That model is gone, when will it come back..?
what is Flux good at ?
maybe no lol
idk, i hope they back this model soon
FFS BRO FIX NANA BANANA
Ask ai it will be better what do you think who sit here in this discord? Researchers ?
Best model for coding
3
5
2
Gpt 5.3 codex
💅
bro… you know arena.ai literally wrote this down like 12 times in announcements, right? 😭
anyway here’s the “how to generate video” speedrun since reading is optional:
- go arena.ai and log in (yes… you have to log in)
- switch to Video Arena (the tab/icon that says VIDEO… shocking)
- type a prompt (or drop an image if you’re doing image→video)
- press Generate / Enter and let it cook
- wait a bit (it’s not instant, this isn’t magic)
- you’ll get 2 videos → pick which is better / download if you want
also: vids are short (couple seconds) and there’s a daily limit, so don’t spend all your gens making “cat vibing” 17 times 💀
I refuse to believe these are real people
Or maybe just old boomers trying out tech for the first time... idk really
I got so fed up I generated the response 😭
@all Which search do you think is better: Grok 4.1 fast search or GPT-5.2 non-reasoning search?
Gemini 3.0 flash >>
I can't say that. I literally asked him the weather, and he didn't even use Internet sources to lie to me about the weather for tomorrow.
He literally made up the weather for tomorrow.
Bruh
He didn't even find it on the Internet.
Wow I asked a question here and he gave a horrible answer even without thinking
How’s this sound?
@pure solstice Note that Video Arena has been removed from the server. More information can be found in this announcement.
Python
someone actually voted opus lmao
it isn't opus?
opus is not even close to 5.3 codex
@magic juniper Note that Video Arena has been removed from the server. More information can be found in this announcement.
Sorry to say that I can't help with this issue. Please, check https://discord.com/channels/1340554757349179412/1343291835845578853 forum to see if this has been reported already. If not, you can create your own post.
bro what
it's only the api
the actual opus 4.6 with me is working well
idk about you dude
Propaganda
"Low poly ocean"
Shows a realistic high poly ocean with 5.3 codex
opus tried to high poly too, but it failed to write custom shaders
codex made working custom shaders
also not low poly, i said optimized for performance
and codex one runs fine
So.. are you comparing codex against opus 4.6 text?
Tbf claude code hallucinates a lot
Thats crazy then, openai actually one upping anthropic in coding for once
Where can I put my prompts for image to video
Note that Video Arena has been removed from the server. More information can be found in this #announcements message.
hello am here to learn ai video editing
Is it possible to attach a photo to the AI text model?
He is broken again
gpt 5.3 is better in my opinion than opus 4.6, and when gemini 3.1 will release i really hope it gonna beat gpt 5.3, we're really not close to very usefull ai honestly
- what's really interesting is that an open source model is as good as gpt 5.1
Kimi K2.5 and GLM 5 really broke the gap with closed source models
i really believe that its possible at some point an open source model beat the best gpt model
@rose jacinth @green dirge @woven veldt Note that Video Arena has been removed from the server. More information can be found in this #announcements message
the big ai companies are just scaling with raw power
which is honestly stupid cause why they don't do what's working with open source model and add it with a very large raw power with huge data center
it would be much more efficient
than it is with only the raw power
Well data quality matters , big organizations I think are just dumping data into ai model and scaling up and the results are wired like gemini 3 pro and flash
if open source model can be made without those huge data center the big companies have, imagine if they just fuse both, like what would be achievable
And open labs don't have that amount of data so they can sort and filter out better quality data and can concentrate on a single model
Passion is what driving these open labs and profit is what driving those big organizations
The difference is clear.
future is bad
yes but its supposed to be a race, whatever drive the big organizations, its really stupid to not use good idea
we will get more realistic models
nano banana already SUCKS at making slop
wheres my beloved veo 1?
sora alpha?
the brainrot just feels different
Like we NEED an open source model to actually beat one of the big organizations, for them to actually use their brain and add those good idea from those open labs to their model
I really don't get it, it doesn't sound logical, why some small team can do almost as good as huge organizations with huge data center with so many gpus, it shouldn't be doable
its like they're too lazy to actually think of better ways to do
and they just keep adding more gpu
What ai hallucinate the least
#video-arena-1 crea un video
Note that Video Arena has been removed from the server. More information can be found in this #announcements message
Lol it has been the 32° advertisement
Usually Claude.
And GLM 5 shocked everyone it's good at not hallucinating
??
i usually see chatgpt hallucinate the most
at least from those that i use
Gemini is the worse
He asked for the least
if 3 flash or thinking flash then yeah
oh wow i am so smart
Both
it's meh, not bad not good
partially it is really good, partially it is AI slop

