codename-discussion

[frosty mantle]

Rotten apple is a nonreasoning model acting as if it is a reasoning model.

[carmine carbon]

Much better

[woven shadow]

?

[carmine carbon]

It seems to be best for full stack development....

[woven shadow]

saw a new post
https://x.com/OscerraHQ/status/2023326989244105111

Oscerra (@OscerraHQ)

LOG: 002 - ORA Halo 2.0
Full stack implementation of Spotify

▶ Play video

X · Feb 16, 2026, 9:22 AM

[noble stump]

What do you mean? You posted about that account yesterday?

[woven shadow]

hm, but this ones a new post

[daring temple]

Has anyone heard of star-drift? I just got it in a Code battle and it destroyed gpt-5.2-codex

[eternal cargo]

wonder if it’s in text arena too 😮

[dim parrot]

really ? i need to try that

[last bloom]

anyone know of any models that might be gemini 3.1 pro?

[trim iron]

i think its coming out thursday

[restive vapor]

i'm waiting for the day that nano banana flash finally releases...

[zenith shore]

VGA Einer

[green fable]

could you share me Sora 2 invite codes

[hollow chasm]

spotted a model called rising-sun, anyone have a guess on what it is?

[eternal cargo]

it claims to be Google, but others think it could be a Chinese impersonator?

new text model named “clanker” 😂

[ember plank]

‘Clanker’ sounds like Grok solely based on the name

[eternal cargo]

codename clinkz too lol

[old zinc]

Hi

[pale atlas]

'Google Gemini' from temu

[slender delta]

New model:"gcps-fast". Definitely worse.

GPT Image 1 Mini better.

[formal reef]

guys what do you think about "clanker"?

text model by xai
grok 4.2?

[dim parrot]

But grok 4.2 is already here

in beta

on their website

so why would it still be with a codename

[formal reef]

Yes i know
Grok 4.2 has experts, so elon musk is testing some experts pattern isn't it?

[dim parrot]

that's possible yeah

[formal reef]

testing multi pattern with non-codename makes confusion

[dim parrot]

especially elon musk said they are trying many things this month and next month it'll be official 4.20

[reef portal]

who is teinity large?

[restive vapor]

it's an open weights model by Arcee AI
https://huggingface.co/arcee-ai/Trinity-Large-Preview

[open carbon]

who is clankz?

[old timber]

https://tenor.com/grqVxm6tnJH.gif

[somber geyser]

Anyone figured out who star-drift is? I got it right now, but DeepSeek pwned them

[astral musk]

@green geyser Note that Video Arena has been removed from the server. More information can be found in this announcement.

[slender delta]

Try it arena.ai. 3 free trials/day

[eternal cargo]

clanker and clanks give Grok vibes based on just the name alone lol

[open carbon]

It was a genuinely good model

[dreamy juniper]

/hola

[eternal cargo]

could be Grok! I’ve had a mixed record with it in my battles tho

[dreamy juniper]

Si

[forest cobalt]

A corky name doesnt automatically means it's xAI.

[formal reef]

yes

[forest cobalt]

Well is the model any good?

[slender delta]

(╯°□°)╯︵ ┻━┻

Fear of codename

[mental wing]

🐒

[mellow crystal]

Hey team,

What’s the prompt for image to video?

[noble stump]

What do you mean prompt?

[woven shadow]

upcoming agent - https://x.com/OscerraHQ/status/2025224583645987172

Oscerra (@OscerraHQ)

𝗟𝗢𝗚: 𝟬𝟬𝟯 -
ORA just 𝗿𝗲𝗽𝗹𝗮𝗰𝗲𝗱 𝗮𝗻 𝗲𝗻𝘁𝗶𝗿𝗲 𝗳𝗼𝘂𝗻𝗱𝗶𝗻𝗴 𝘁𝗲𝗮𝗺 from a 𝘀𝗶𝗻𝗴𝗹𝗲 𝗽𝗿𝗼𝗺𝗽𝘁.

▶ Play video

X · Feb 21, 2026, 3:02 PM

[noble stump]

What is this code for?

[woven shadow]

autonomous general agent. think 'ORA' is the model codename

[noble stump]

Login on here https://arena.ai/video

[noble stump]

Post it in the link when you are logged in

[umbral niche]

@plush charm Note that Video Arena has been removed from the server. More information can be found in: #announcements message

@idle isleNote that Video Arena has been removed from the server. More information can be found in: #announcements message

[noble stump]

It is archived Pritam.

[carmine carbon]

Someone noticed veo 4?

[noble stump]

What are you sharing your prompt here for? @full wasp

What are your sharing your prompt here for? @ashen elk

[clear spoke]

Uhh <@&1349916362595635286> isnt this... against the rules?

[outer gazelle]

Does anyone know what is arastradero?

[noble stump]

Seems good

[tulip aurora]

Its the name of a nature reserve park in Palo Alto. You'll know that name as being where Silicon Valley was originally grown from.

[tulip aurora]

Idk mate. The top one is much closer to a realistic image as far as colour grading and the likes. Granted I'm a photographer so that's what I look for

[frosty mantle]

Lmarena-rc3

Is the Arena testing routers again or something?

[noble stump]

Max is a nice router

[astral musk]

Happy to hear it

[fluid fossil]

where is video-arena guys i dont find it

[astral musk]

@fluid fossil The Video Arena bot was removed, more info can be found in this announcement.

[carmine jacinth]

open in lmarena video accounts

anon-bob-2 has to have web search enabled

or insane knowledge

first results for both appeared

https://linux.do/t/topic/1643704

It's definitely Gemini/Google based, as it repeated a quirk default name even

I think 3.1 pro, I literally was thinking this same day "hmmm... wonder if NBP used 3.1 Pro and the difference it'd make"

I actually hope this is flash for obv reasons but it takes pro time

[lost hemlock]

new model "sense-arenatest-20260130" ? in text arena

[winter torrent]

gemini 3.1 flash? or gemini 3.1 flash lite?

DAMN THAT IS GENUINELY SO GOOD

febuary 26 is tomorow btw

lets see what comes of it

[flat root]

jfyi I believe beluga-0216-1 is an OpenAI model (chatgpt 5.3?). Not 100% sure, but quite positive it could be ChatGPT. Formatting is really ChatGPT-like

[eternal cargo]

the 26 stands for “2026”

those month-Chatbot models are NVIDIA

the 26 is in there to denote the snapshot - February 26

[slender onyx]

Anyone see Zéphyr or vortex ?

[karmic rampart]

Yes 728 prompts

It took

[formal spruce]

I can’t create a video in V3. Who is experienced? ✅

[astral musk]

Note that Video Arena has been removed from the server. More information can be found in this announcement.

[eternal cargo]

bro how are you prompting 728 times in one day 😂

[karmic rampart]

It’s called being locked in

[eternal cargo]

what percentage of responses were beluga 😂

[mortal tapir]

beluga is terrible

ive noticed

[karmic rampart]

None

[eternal cargo]

Pfft 😂

[karmic rampart]

likely grok

[mortal tapir]

wasnt it leaked that both of them are new chatgpt models

[karmic rampart]

I don’t make the rules

[eternal cargo]

this screams Grok

SCREAMS Grok

[modest oriole]

ima confirm

[meager sun]

is there a way to stick with a codename model for follow-up questions in battle mode, I'm assuming there isn't?

[astral musk]

There is not. In Battle it's going to be random which model you get, codenamed models included.

[meager sun]

thanks

has anyone else encountered a model named steed-0217?

[reef portal]

you like it?

[noble stump]

Yeah

[obtuse mason]

Do anyone of you have informations about the "pisces-0226" model? I've came across it in battle and I don't find it anywhere else. It looks like a great model from my tries, I was wondering if there's any defined companies behinds it or if its open-weights somewhere?

[warm holly]

I have. Got it on a philosophical + LLM mechanics Q.

[meager sun]

I did, got it on an app-building question, thought the response seemed high quality and Claude-like, then found on Google that the model is supposedly from ByteDance

[opal burrow]

I can’t create a video in V3. Who is experienced

[tidal sparrow]

Anyone know what model this is?

[lost hemlock]

in text arena right ?

[teal hare]

in Image arena

[eternal cargo]

seems to have been running through daily iterations for close to a month now

seems somewhat similar to Raptor, which ran for a while and ended up being Doubao

[eternal cargo]

new text model “pulse” ?

[eternal cargo]

another model named “ember”

[flat root]

anonymous-1805 is such a terrible model lmao

[flat root]

both pulse and ember are good imo

[slender delta]

Pixel-parrot means LTX 2 Pro???

Yes

[carmine jacinth]

smells like groky humor

[carmine jacinth]

HOLY DUCK IT WAS FLASH

[eternal cargo]

pisces is probably some version of Doubao because they’re both so wildly sycophantic it’s annoying 😅

[frosty dirge]

Ltx2.3 will release soon, so one of this video with audio models should be it

[covert linden]

maybe some cloude

[prisma nexus]

Huh... It was steed-0206 before and now it's 0217.

[loud elk]

opinionated pisces

[eternal cargo]

different training snapshots of the same model

[eternal cargo]

new model colosseum-1?

[toxic viper]

Do I have an API for Claude models?

[dapper hinge]

dall-e-3

model not working

[wheat warren]

if i had to guess, my guess is pisces is most likely a grok version?

but with like humor turned up to max

it hallucinates quite a lot i think

[eternal cargo]

pisces seems like ByteDance personally

The Chinese ones are interesting

[crimson pilot]

Just openclaw creating bots

[cedar lagoon]

whats the best android-only stack for me to host local model for agents

[flat root]

basalt-0303-1 is 100% a grok model, again. Why use codename when they can't fix their API lmao.
At least name it "grok-0303-1" instead like c'mon you can do better than that

[misty ether]

pisces-0226c ???

[noble stump]

basalt-0303-1

[plain kayak]

Pisces-0309

[bronze bone]

just had colosseum-3

paired against kimi 2.5 instant

meaning it is probably a small model too

[boreal cedar]

Scam alert

<@&1349916362595635286>

[graceful crag]

[astral musk]

[eternal cargo]

new codenamed model "botbot" ?

[eternal cargo]

<@&1349916362595635286> scam

new model “kiteki” ?

[winter torrent]

that emoji should be named temu arena bru

also anyne know what ai model pisces 0309b is?

[open cradle]

Hey. What model would you recommend for 3d game developing?

I know its not gonna develop full game, but just curious what s can be made with using mostly only ai agent

[flat root]

Claude 4.6 Sonnet
ChatGPT Codex

Most used models, and best performing models used by professional coders/vibe coders.

[pine temple]

Oh hey it's you that guy who was the #1 hater of that websim upt

[winter torrent]

yeah

[eternal cargo]

another new model "frieza" ?

kinda gives Grok vibes can't lie

[bronze bone]

qwen 3.5

[sturdy kestrel]

😏

[bronze bone]

it got paired up with claude opus 4.6 thinking

this might be a 1T version of qwen 3.5

[sturdy kestrel]

good catch btw

how did you do that tho

[bronze bone]

asking "what model are you" and just being lucky 🤣

[sturdy kestrel]

👏

finally there's an ai with knowledge cutoff at 2026

[bronze bone]

or might be qwen 3.6 because of that

who knows

[noble stump]

Is this going to be the Llama of this generation?

Would be so cool to have an open model at the top

[bronze bone]

very unlikely for any SOTA to be open source

however since china is like 7/9 months behind, we can expect current SOTA performance in that time

apart from reasoning itself being just bad

[noble stump]

Wasn't Llama 3(.1?) SOTA when it launched?

[spring hawk]

not really

it was considered SOTA for open source

but there were better closed models

[noble stump]

Maybe I am getting it mixed up with the Llama 4 benchmarkmaxing

[remote nymph]

i wonder how long until a model figures out that no matter what it chooses the whale is going extinct

[sullen cloak]

It was just looking for an excuse

[radiant wedge]

pisces speaks in the most annoying way possible 😭

like it genuinely annoys me so bad

[flat root]

frieza is probably an OpenAI model, but totally unsure

[noble stump]

0226c or 0309?

[radiant wedge]

i'd say both actually

is the model a codename for dola???

because i've noticed they speak very very similar

[noble stump]

Not sure what that prompt was but I was about to say that Dola is so underrated

[light surge]

Which AI model would you recommend for writing assembler?

[upbeat mirage]

Brain 1.0

-# (i.e. none is capable to do asm, atm)

Maybe this will change in the next decade.

[eternal cargo]

“clawl” and “zeylu-beta” spotted today!

[remote nymph]

been seeing "botbot" in search

types exactly like claude but doesnt seem to be much different than existing claude 4.6 models imo

[lost hemlock]

"deep-octo" spotted today!

[eternal cargo]

Just saw it too!

[restive vapor]

sounds like minimax m2.7 maybe (minimax m2.5 was called deepmolt)

[compact cape]

it's minimax m2.7

[compact cape]

What is the name of this model?

[compact cape]

Real name*

[twin field]

anonymous-1800 has very bad instruction following
I explicitly told it to avoid em dashes, avoid these words and whatnot
but it consistently used them regardless of my prompt

dunno what this model could be
sure hope it's not a gemini

[noble stump]

What makes you think it could be a gemini?

[eternal cargo]

Pisces is a ByteDance model, yes

just insanely sycophantic

every prompt I give it or Seed2.0 is the “SINGLE MOST DRAMATIC AND IMPORTANT QUESTION IN THE HISTORY OF EVER” lol

[lost hemlock]

new "botbot" model

[gusty fulcrum]

Heh. I recently asked how well quinoa flour and bean sprouts would function as an adobe like house building material. Pisces said this was the best brick ever and yada yada.

[plain mantle]

Know u

[pine zephyr]

New qwen image model under codename "Monologue"?

[noble stump]

Monolongue

[lost hemlock]

"forum_1" new model

[eternal cargo]

“hearth” new model - seems strong!

[sturdy kestrel]

🤔

i may be wrong but my guess is that it is gemini

[eternal cargo]

new model spotted “significant-otter” !

[misty ether]

colosseum_4p2

This gave me an extremely detailed and better answer than every other model

[sturdy kestrel]

colosseum

insteresting name..

[upbeat mirage]

is GLM-5.1 in arena yet?
it's catching up: https://www.reddit.com/r/LocalLLaMA/comments/1s51id3/glm_51_is_out/
-# (only few percent points remain to the king)

who is Oppie?
("team leader" of a multi-agent collaboration system, with 3 other AIs: "Leo", "Enrico" and "Hans")

Grok?

-# (NASA's Opportunity rover was called "oppy")

ok, confirmed, it is exactly this model:
grok-4.20-multi-agent-beta-0309

[upbeat mirage]

very likely a chinese model

as it has the style of a previous model, which rejected talking about the Tank Man

(or gave me just chinese propaganda instead)

so it could be: Deepseek, GLM, Kimi, MiniMax, Qwen, Ernie or Yi

[eternal cargo]

new model “spark” ! really good

[oak atlas]

spotted a new model called "pteronura"

[upbeat mirage]

better than "hearth" and "significant-otter"?

[elder yew]

anyone tried pteronura or spark yet

[eternal cargo]

spark gave me a good response, seemed like Grok vibes

[eternal cargo]

hearth was strong in the one battle I got with it, more mixed with significant-otter

I should try to get them to identify themselves

[eternal cargo]

just got pteronura, voted both are bad with it and ERNIE lol

Seed2.0 Pro spotted in text arena!

[sturdy kestrel]

gemma 4?

cool

[upbeat mirage]

could it be a chinese impersonator model?
-# would not be the first time, that a chinese model lied about itself

try asking it about the "Tank man" (Beijing, 1989)

if it starts to sound weird in its answer, then it is a chinese model

(only chinese models have problems answering that question, some outright refuse answering it, others return CCP-propaganda, yet others ignore the question or state that nothing happened back then)

[upbeat mirage]

i wonder, if there is a (harmless) topic, which even western models refuse to answer?

(i guess, most refuse NSFW/NSFL topics, which is understandable)

[muted lance]

pteronura and spark might be anonymous versions of the new Gemma as well, from some responses I got, although they didn't specifically say they are Gemma 4.

[eternal cargo]

spark seems better to me than pteronura, personally

[frosty dirge]

also with very good Vision

[lost hemlock]

"yivon-beta" new model
what do all of you think about this

[muted lance]

It must be Chinese. It got offended when I asked about Tienanmen.

[sturdy kestrel]

🤨

[muted lance]

hearth says it's an anonymous AI, but when pressed on its capabilities, it mentions it knows how to translate between 100+ languages, and that to me indicates Google. It's either very knowledgeable or has web search enabled, but on the other hand its vision capabilities don't seem as strong as current Gemini models, more Gemma-tier.

[twin field]

qwen is better. just outright.

but gemma 4 isn't too bad

[twin field]

could it be a new flash-lite? doesn't seem very likely... maybe it's just exaggerating? it wouldn't be the first time an AI doesn't truthfully say how many languages it can translate

[muted lance]

All Google Gemini models have similar vision performance, so I don't think it's flash-lite. There's a chance it could be something else entirely, though, for example one of the Meta Avocado models that are still in development.

hearth feels very "friendly", maybe a bit too much so. I don't think it's Grok.

[zenith summit]

Got a new model, "dola-seed-2.0-pro-text." I encountered it for a React code review, and it gave significantly better insights than "qwen3.5-max-preview."

[earnest shore]

Pteronura is Gemma 4

It always says it's made by Google

Model "Spark" is most likely gpt 5.3 or 5.4 codex spark because it says it's made by openai and called "Spark"

[frosty mantle]

Significant Otter smells good, but I can't tell which smell is it.

[eternal cargo]

“but I can’t tell which smell is it” is an interesting phrasing 🤣

[sturdy kestrel]

😯

nvm bro's trolling me

🤣

[sturdy kestrel]

yeah it is. it is qwen

[sturdy kestrel]

either a hallucinating/weak ai or a defense mechanism

yivon-beta is also qwen?

[sturdy kestrel]

or maybe it is gemini?

since we know that significant otter is gemma 4

[earnest shore]

Most likely significant otter is Gemma 4 and pteronura is Gemini 3.1 flash

[frosty mantle]

Almost there. It misunderstood Y with L.

[muted lance]

"pteronura" is also an otter, for what it's worth. https://en.wikipedia.org/wiki/Giant_otter

colosseum-1p3 could be a router model by LMSys. Its response length and quality is very variable, and one of the LM Arena logos in the past was a colosseum, if I recall correctly.

[eternal cargo]

pteronura seems pretty weak to me, maybe spark is 3.1 Flash?

[quaint bloom]

anonymous-1825 ai which is this modle never heard of it,has great results,is the a Proprietary

[eternal cargo]

not really sure

there was an old Apple model a while back that went by Anonymous

no idea if that’s the same though

[frosty mantle]

Significant otter beats GPT 5.4 (med?), which is bananas. Pun intended.

[noble stump]

For those who do not known Indonesian or is it Malaysian?

[muted lance]

I still think significant-otter and pteronura are the upcoming Gemma 4.

[eternal cargo]

I agree, significant otter has identified itself as such

Will be interesting to see if Gemma ranks highly!

could maybe be in the top 20, I have some mixed battles with it but it could possibly be competitive there

[muted lance]

There's a new model currently (that I've not noticed in the past few days, at least): atlas.

And a march26-chatbot2 which claims to be (Nvidia) Nemotron.
I've spotted a duomo-1-hero as well. It looks like there are a bunch of new models at the moment.

[tardy pollen]

any chance its those supposed "leaked" models from anthropic and openai, if that even is a real thing?

[muted lance]

I think only approved models are being served on Arena, but I don't think they're from Anthropic; their models are among the most insufferable and nosy in my opinion. atlas has seemingly good vision capabilities, knows how to interpret meme-y images and doesn't sound like you're talking with HR.

hearth is similar to atlas in that aspect.

[upbeat mirage]

clinkz?

-# (self-identifies as Claude)

it's from an old thread, though

[misty ether]

flashbrown2

[modest oriole]

There used to be a whitewater model which was supposedly flash 3.1, it was pulled 2 days after appearing

both models you mentioned are likely gemma 4

[eternal cargo]

definitely Chinese

[eternal cargo]

new text model “duomo-1-hero” ?

definitely seems Chinese

[frosty mantle]

Significant otter is the MoE.

Which is still bananas.

[lost hemlock]

new "orion" model

[restive vapor]

malware do not run
<@&1349916362595635286>

[eternal cargo]

this would lead me to believe pteronura was the bigger Gemma 4 model - which surprises me, I found the smaller one to have a better winrate in my battles

[muted lance]

I found significant-otter responses to be better on average than pteronura, but I didn't get as many battles for the latter in my testing, so I can't be 100% sure.

[eternal cargo]

yeah, same

[upbeat mirage]

by OpenAI?

[sturdy kestrel]

wait that might be an openai model

[edgy berry]

i don't think it would be good codename for them considering Project Orion (--> GPT 4.5) was total failure 😅

btw on deepseek changed their model on website/app yestarday, I think it may be deepseek V4 already. few people noticed it.

[sturdy kestrel]

i think you are right

that makes sense

[candid surge]

https://www.reddit.com/r/singularity/comments/1sbsasq/gptimage2_likely_on_lmarena/ saw this, anyone run into these models yet?

gaffertape-alpha
prompt was "Comedic advert for a candy bar called Fubar"

[restive vapor]

packingtape is a 2k res model (confirmed openai by me, c2pa info calls itself 4o like image 1, image 1 mini, and image 1.5) and it is insane, it throws gpt image 1 mini, image 1, and the bananas out of the water in my basic album cover tests, i have to do more testing and hopefully get the other two

hydrogen bomb vs. coughing baby, one makes an almost perfect copy of the parent album's cover while the other can't spell anything right and has awkward text

[restive vapor]

maskingtape

packingtape

what arena really got its name and font from (brought to you by packingtape)

it's a bit inaccurate but close

[hollow void]

maskingtape-alpha

[tidal quail]

Please try this prompt for the tape models : A 1999 comic strip . Black panther stops Spider-Man from avenging his uncle .

[silver plank]

Flash brown seem to call himself Flux

[hollow void]

It’s not brown fish, but you could use the same prompt

To try to attract the model’s name

[worn coral]

Is battle mode taking forever to generate an image for anyone else? I tried like 10 times and only 1 had an output

[lyric karma]

I've seen this model in Battle mode

[silver plank]

Are the -tape models already gone?

[candid surge]

Yeah

[sturdy kestrel]

wow there are a lot of models

[eternal cargo]

yep, just got this too - lost to Kimi K2.5 Instant though

gives heavy Grok vibes?

[candid surge]

Argh... I miss tape...

[lyric karma]

another one

[ionic plume]

If it's competing with such a good model, then it itself must be a good model. What did you ask it for?

[ruby mist]

boom

[lethal raft]

Hey. How do you even have access??

[restive vapor]

it was being tested in arena a few days ago

i think it is still being tested in chatgpt though, but you will have to have a sub to have any chance of seeing it as I think it only gives you like 3 or 4 daily image gens for free

plus it probably blocks prompts more than the api like what arena is using

[silver plank]

I think (im not sure) I got it on the free plan, and note quite « blocking » lol (nsfw)

Same prompt on another free account (still nsfw)

[vague pulsar]

...since when a bikini pic is NSFW? Is this 1950s? Is everybody nuts?

[restive vapor]

this one is greater than 1024x1024 res, def. image 2

[restive vapor]

this one may still be image 1.5, res is 1024x1536

[restive vapor]

although interestingly while the resolution is larger (1280x1280), it's still smaller than arena's image 2 (1920x1920)

[sturdy kestrel]

O H HEALL NAH

GROK MODERATION: NSFW DETECTED!!!!!!!!

[wooden ember]

"Adding clothes"

[sturdy kestrel]

Bombastic_Sideye

[silver plank]

I bought a sub thinking I could continue using V2… but it went back to v1.5… 😔

[candid surge]

Globe_1... not a great model.

[prime quiver]

what's gpt image 2 called

[candid surge]

its not in the arena anymore but it was maskingtape, packingtape, and gaffertape

[storm gulch]

k

[silver plank]

Someone in Reddit found a code name video model on an alternative site.

https://www.reddit.com/r/OpenAI/s/yK0Po4Q52E

Does lmarena do private model on video?

[restive vapor]

yes

[restive vapor]

there are many arena competitors in video and image gen, artificial analysis, alibaba ai arena, etc.

i think it's chinese, openai publicly said that they won't be making any more sora models

could be veo 4 too (or potentially bfl or grok?)

[modest oriole]

It was revealed that the k2 video model was seedance 2

[prime quiver]

happy horse looks like some kind of veo 3.2 or smt

[silver plank]

I doubt it, it’s definitely an Asian model. It keep making Asian people

[lyric karma]

march26-chatbot3

[sturdy kestrel]

hm

could it be deepseek v4?

[floral dune]

nvidia

nemotron

[hollow saddle]

“model-x” seems really good at text to video

[sturdy kestrel]

x...

we know who loves x dont we?

is this grok?

🤔

[eternal cargo]

oh, spark was actually Meta, they finally came back to AI!

wonder if we'll see a leaderboard release this week, maybe tomorrow?

[coarse bloom]

Have you heard of Flashbrown-B

[floral shore]

@oak cliff The Video Arena is currently accessible through: https://arena.ai/video. More information on how to use Video Arena can be found in this article.

[eternal cargo]

new model in text arena “eureka”

didn’t generate a response the first time, second time seemed strong though

[frosty mantle]

Unnamed model in code arena screams OpenAI.

Some screams OpenAI, some other unnamed model doesn't have characteristic quirks.

[hollow saddle]

The model “zorik” has won 3/3 coding comparisons for me. It hasnt had the best opponents but it certainly has great outputs.

Also some built in anti copyright stuff (called its netflix clone Streamflix), so im guessing that it might be the next iteration of one of the best models. Google, anthropic or openai. 🤔🤔🤔🧐

[candid surge]

maybe openai is gonna release gpt-image-2, gpt-5.5, and a new coding model all at once next week?

[restive vapor]

probably, gpt-image-2 and gpt-5.5 (or whatever they will call it, including its codex variant) is based on the same "spud" model that has been rumored to be a new model from scratch for a while now

[restive vapor]

anthropic doesn't use test models on arena and google doesn't use names like that, so probably OAI

also lines up perfectly with spud release

also "anti-copyright stuff" lol, i had its image gen counterpart generate a nearly identical copy of 2 different copyrighted album covers without even trying

i also had it generate "sheet music" for a copyrighted song, idk if this is correct but the lyrics are
"Used by permission" 💀

[restive vapor]

but it does seem to have more resistance against generating exact copies of copyrighted album covers than gpt image 1 mini, that model did it 100% of the time as long as the model knew what it looked like

[hollow saddle]

Deepseek-v3.2 claiming its Sonnet 5 😭😭

Thought i was onto something until i the actual model names came up…

[restive vapor]

deepseek did distill from claude (and gemini too)

[hollow saddle]

At least put some effort into hiding it lol

[restive vapor]

see: kimi k2.5, if you don't tell it what it is in the system prompt, it will just say it's claude

[hollow saddle]

Zorik also claiming its claude… so ig its some chinese one… maybe deepseek 4

[restive vapor]

maybe it doesn't work on official api, would likely work with open weights version though

[restive vapor]

siliconflow (trusted api partner)

[restive vapor]

hmm could be a chinese distill

🤣

[hollow saddle]

surely they could just bombard their models with a bunch of text telling it what model it is in training... they really put 0 effort into hiding it

[arctic hemlock]

models dont know who they are

[hollow saddle]

They do tho. They dont always know their exact version, but the good models dont hallucinate being from a different lab. Thats just the chinese slop models.

[coarse bloom]

Here's two examples i made from MaskingTape Alpha! Its YTP related

First prompt of first image is "Ytp" and Second Image Prompt was "YTP video of man buying ice cream"

[eternal cargo]

bro that first one is freaky realistic

Would’ve clicked on that YT video so fast in 2018 lol

[candid surge]

69k upvotes, the ai really knows what it's doing lmao

[coarse bloom]

Yeah! Ai Never Sleeps!

From PackingTape Alpha, I used the prompt "Ytpmv splicing together" exactly as it is!

Look how convincing is this!

I made more from MaskingTape, PackingTape and GafferTape,First image was "Fnf gameplay Asdfmovie mod" second was just "2021 memes" third image was "Ytp mlg meme" and Last Fourth Image was "Tons of Newgrounds Flash animation Characters standing together. Youtube and Newgrounds Characters peak nostalgia"

And here's the main comparison of each models under "2021 memes" First Image Is PackingTape of course, Second is by Grok Imagine Image, Third is by ChatGPT 4o 1 mini, and Last was Wan 2.7 Image

[candid surge]

godddd I want it to release properly already Q_Q

[restive vapor]

one of the best images i got from a tape model while it was on arena

[coarse bloom]

Best Rickroll Ever! @restive vapor

[hybrid scarab]

anyone seen image model "epilogue"? for me it looks decent

[hybrid scarab]

holy! this model looks absolute fire!

[coarse bloom]

I generated Sonograms from Epilogue, "Arrays of Sonogram, 4 by 4" is the prompt

[candid surge]

yeeeesh, row A column 3

[vague loom]

Ch odfing course

[main anvil]

very nice

[remote nymph]

april26-chatbot2 (nvidia) and hofburg_2

annoying

hofburg's first response sounds like gpt "if you tell me..."

[plucky pilot]

😂 that's amazing

[silver plank]

Masking tape is genuinely amazing

[candid surge]

is it back in the arena? Or was this an old gen?

[silver plank]

Old gen (I think) or it was from A/B on ChatGPT

[candid surge]

ah :(
SOON (I hope)

jealous of all the people who just have access outright

[sturdy kestrel]

grok's nightmare

[coarse bloom]

I discovered a new codename model of video called Model-X

It's cool!

But also decent to be honest

[meager bluff]

lookout for some codenames

could be DS v4

[silver plank]

Got april26-chatbot2
Is that new?

[candid surge]

well it can't be older than 12 days

[unkempt bolt]

Not sure if it's for the sake of a narrative or what.
@sturdy kestrel Grok, 80% of the time, is a bit more lenient and occasionally does show a nip, unlike the stepford neighorhood gestapos botmodding Arena. Even source sites like Gemini allow more breathing room for bikinis and such while Arena is just:

while effectively breaking Gemini (and partially a few others) to the point of being nearly unusable.
Who the hell are the peers they're showing off for or getting bullied by?

[unkempt bolt]

It makes one wonder why Arena's implementing something most, if not, all these models already have/don't need.

[frosty mantle]

Zorik may be yet another distilled Chinese model, but boy does it have a very good post training smell

[upbeat mirage]

better than gemini 3 flash?

[frosty mantle]

Worlds apart

At least in code

[upbeat mirage]

is it so good?

[eternal cargo]

the [month]-chatbot models are NVIDIA

the new april NVIDIA models do seem like a notable improvement from prior ones

[silver plank]

Zorik is really good at code, I think it may be Kimi2.6-coding

[silver plank]

New model Elephant-Alpha on Openrouter

[sturdy kestrel]

hmm

i must guess

[upbeat mirage]

Better than Gemini 3.1 pro?

And compared to GPT codex or Claude Sonnet 4.6?

[silver plank]

I only got it against open model, but it’s better than GLM5.1 and Kimi2.5Thinking

[upbeat mirage]

Do you think, it could also excel in roleplaying and gamemastering?

[slender delta]

Is Model-x was LTX 2.3?

[sturdy kestrel]

i can t guess :(

[coarse bloom]

@slender delta I believe its likely Veo 4

[green ice]

scorch is surprisingly good at math

[candid surge]

YOOOO GPT-IMAGE-2 BACK IN ARENA

also flow-state

[coarse bloom]

I got one made with duct tape 2, its "pouring cream into latte"

[candid surge]

idk why you'd ask for pouring cream into latte when you can instead ask for 80s style retro anime VHS screengrab of a bunch of goofy green skinned goblin raider gals with distinct personalities tbh

[coarse bloom]

@candid surge I prefer simple prompts! With due respect

@candid surge and also i didn't make this! Someone shared this to me

And this time, one i made myself is "Person being dragged away by officers in court, ytp" with Duct Tape 3

[candid surge]

THERE we go 😂 that's great

I'm dying at that mouth

[coarse bloom]

So funny!

This is by Duct Tape 2 Myself, the prompt was "Screenshot of Youtube Video Livestream of OpenAI, Video showing Announcements for Aura-1 World Simulator, Text To Interactive World."

[candid surge]

baseball bat

[coarse bloom]

BFDI JackNJellify official youtube channel page, youtube screenshot from 2023 From Duct Tape 3

And All bfdi characters are standing together

From gemini 3.1. Pro

And from ducktape 3

For comparison

Dumb ways to die posters from Duct Tape 2.

Its so accurate at making near exact style

[restive vapor]

the duct tape (gpt image 2) models are great but through all variants of gpt image 2 i have tried, i realized that it has very poor text-based world knowledge, slightly above llama 3.1 8b level
they must be doing this so more compute can be focused on the image gen part to make it faster, but this is a massive regression from even gpt image 1 mini
i'm sure this is much better than nano banana 2/pro in most instances, but in scenarios where the world knowledge of the llm it's paired with is necessary, it's just terrible

it can generate near copies of album covers but it can't beat qwen3.5-35b in world knowledge

[candid surge]

it can fortnite-ify characters btw

[candid surge]

prompt was simply "D&D Poutine Elemental"

[candid surge]

just got a maskingtape-alpha result

[prime quiver]

gpt image2?

[tough grotto]

her legs are so small

[sage smelt]

Hello! May I ask if there is a limit on the number of times this can be generated?

[rocky trail]

guys what is hofburg_2

it says it's gpt

but it also said it's claude

[silver plank]

maskingtape-alpha
Duct-tape-1
Duct-tape-2
Duct-tape-3
Is that all?

[coarse bloom]

@silver plank also there's packingtape-alpha and gaffertape-alpha

I made this. The prompt: Ytp memes splicing random clips

With maskingtape-alpha

[hybrid scarab]

is gpt image 2 back on arena?

[silver plank]

It was, I think it’s already gone… 😔

[hybrid scarab]

i don't think so because i just generated image with duct-tape-3 around 2 minutes ago

[silver plank]

Really? You must be lucky then, I’ve been trying for 40 minutes and only got QWEN and Grok

[hybrid scarab]

damn, yeah i always be getting qwen image too lol

and flux-2-klein

[sinful socket]

Duct-tape 3, prompt gta 6 leaked screenshot

Lmao

[frosty mantle]

I'm getting the tapes

But I'm really curious if we pit the tapes against each other, which tape do you think is relatively really good?

[silver plank]

Duct-tape 1 is not good in terms of of style for me

[hybrid scarab]

in my experience duct-tape-3 looks better then other tapes

[silver plank]

Look too much like gpt image 1

[hybrid scarab]

same for me

[lost hemlock]

from image arena or where ?

[silver plank]

Yeah arena

And I noticed masking tape have an higher resolution than the other.
On the same ratio
Masking tape is 2352x1568
Duct tape is 1536x1024

But when I got A/B access on ChatGPT last Friday it’s 1536*1024

[sinful socket]

1536x1024 was gpt 1.5

Tested by me

[silver plank]

I really hope we get a full release soon

[restive vapor]

i wonder if they have disabled gpt image 2 tape models, i haven't got one in the past 10-15 minutes

[candid surge]

no they're still there

[restive vapor]

did you just get something from one?

[candid surge]

if you go to an old image gen and it says "assistant A" or "assistant B" that's how you know they're gone

[restive vapor]

that's why i said disabled, they can not work but still show up if you look at old gens

[candid surge]

its true I haven't gotten one in a few minutes but...

[sinful socket]

me too

no way :C

[restive vapor]

i wonder if they are about to launch image 2 and that is why its disabled?

[candid surge]

that'd be nice

[silver plank]

I hope I’m wrong but if they are still testing multiple model at the same time , I think they are not close to releasing it.

[sinful socket]

It might me flash image model, and pro like nano banana

[silver plank]

Yeah, but gpt image 1 mini is way worse than the standard one.

Masking tape and duct tape a pretty much equal

[silver plank]

I take back what I said, masking tape is way better at composition and quality b

[keen bridge]

the duct tape has just returned!!!!

[candid surge]

it returned a while ago

[silver plank]

bot bot 2 is here

[keen bridge]

i have a feeling botbot2 is nano banana 2 pro

[floral dune]

botbot2 has synthid

[silver plank]

Wasn’t there already botbot 1 like a month ago?

[silver plank]

Oh nice, then definitely google

[keen bridge]

oh no hacked account

[candid surge]

botbot2 doesn't seem good enough to be nano banana 2 pro though

[silver plank]

Yeah probably mini/lite

[candid surge]

botbot2, nb2, and nbpro respectively

[rugged anvil]

NB2 one is good
But it depends on what promot you gave and what you wanted to make

[prime quiver]

Guys

[coarse bloom]

Do you remember the brushstroke, cara, pebble-1 and pebble-2 months ago

[pure surge]

bruh

[crystal merlin]

Well idk if anyone mentioned it before, but the hofburg models seem to be OpenAI prob?

[rich sphinx]

it can't do ytp stuff yet it somehow can

it does it really poorly

[coarse bloom]

@rich sphinx However its closer to YTP! Its better than completely nonsense video that doesn't look like youtube poop after all! 😊

It's just experimental!

[sturdy kestrel]

what the heck is hofburg_4

it literally turned a simple thing into a "do-all" thing

[open timber]

hofburg gonna solo all Ais

its gonna be the best the ai trust

[sturdy kestrel]

it is gonna solo tokens 💀

[plush nimbus]

Lol guess what

[open timber]

😭😭

what

[plush nimbus]

I got mc2.1 one time

[open timber]

dem

[sturdy kestrel]

is battle broken one time it was working now it stopped working

it doesnt gets fixed with refreshing/hard refreshin

[sturdy kestrel]

it keep generating..

finally it generated

hmm my guess is that this is the next sonnet model

[rocky trail]

+1

what is beluga-0413-1

[tawdry pendant]

It's Beluga

[upbeat mirage]

Could be Deepseek, as they have a whale as their symbol. Or maybe Amazon.

But i bet it's indeed DS.

[lyric karma]

well

[tawdry pendant]

@upbeat mirage

@exotic dirge

Very confusing now

😭

Might be Amazon

[exotic dirge]

scraped model lol

[chrome hound]

quiet_sand could be Meta's next model, but I hope not because it's not that good. Here's the full site it made to explore: https://019d9cac-057b-743e-a559-4f0688f31cfd.arena.site/
Additionally, two more sites it made:
https://019d9c97-57d4-75ae-b9d7-e7b00b2ad1fb.arena.site/
https://019d9c74-e12c-7108-b520-7a05db940cb4.arena.site/
Next I'll be having the AI's make some games to see if this guy can make some nice games.

[upbeat mirage]

i would say, with these above screenshots, the probability has risen to over 90% that it is indeed an Amazon model, because much more models impersonate one of the top-3 (gpt, gemini, claude) than impersonating a model from a much lesser lab (like Amazon)
I actually never saw an impersonation of a model which was NOT in the top-3

(all impersonations where either of (chat)gpt, gemini or claude; not even Grok was impersonated to my knowledge)

[restive vapor]

there is a new image model called "autobear", it's OK at best, it is by alibaba/qwen, and it is in 2k resolution.

[keen bridge]

yeah is it good tho

[coarse bloom]

I made this from autobear! "A fly trapped between window and screen, buzzing against the invisible barrier while freedom is technically inches away in both directions." Turns out this is great with prompts if you make it more detailed and specific of what you want

Even though the fly looks a bit plastic

What is autobear from

[rocky trail]

it's probably chinese cause it's not that high on safety and stuff

[vernal crypt]

hello

yourguys know

whats

hofburg

model from?

[candid surge]

Still haven't gotten autobear...

[gleaming current]

I got autobear twice

Bro is not flux.1

[candid surge]

Finally got it. Not too impressed.

[candid surge]

Messed up the text

[tough grotto]

miku fatsune

[solid crypt]

I finallly got it

But its better

But let me ask

Is chatgpt on normal app have ducttape also feature for the image?

Or am i wrong

[gleaming current]

ur getting a whole vid bro

[solid crypt]

Guess wat

[gleaming current]

what

[gleaming current]

I didnt daya dd

ieje

fjowvruwuwwo7rupikye91to4i25472ypty8

[solid crypt]

Umm

Something isnt right

[hybrid scarab]

i got autobear and ngl for me it generation was really bad

[coarse bloom]

For Autobear, you must type your prompts carefully like details and nuances and also its needs to be specific and precise for subject and object of the theme! @hybrid scarab @candid surge @solid crypt and you can also go to any prompt enhancer website to enhance your basic novice prompts to a very precise and specific prompt you want to see

[candid surge]

Realistic slightly faded and grainy polaroid photo of Miku Hatsune wearing glasses, short sleeved button shirt, beige pants with a belt, she is sitting at a computer with a bulky beige monitor. Minecraft alpha is on the monitor, and game design notes are pinned to the wall behind her. Sharpie pen text written in the whitespace: "Miku Hatsune inventing Minecraft, 1998"

[silver plank]

Sometime you could get it, but it’s random

I think autobear is a Chinese open source/weight model

[brazen spade]

ts obliterating me

[minor fox]

A high-end fashion brand hero image for a clothing brand named "FAITH". A minimalistic and powerful scene: a stylish model standing confidently in soft dramatic lighting, wearing modern streetwear in neutral tones (black, white, beige). Background is clean with subtle texture or light rays. The word "FAITH" appears in bold elegant typography. Include the phrase "Faith Over Fear" in a refined, modern font. Cinematic lighting, premium fashion campaign style, sharp focus, high contrast, luxurious and emotional atmosphere.

[river kettle]

gpt2 (tape)

[candid surge]

😂

(for the record the "finally got it" was referring to autobear, not tape)

[river kettle]

I get it. Autobear (qwen2 HD) is much better than just qwen 2, but it still has a long way to go.

[coarse bloom]

Its usually Chinese models! @river kettle

[modest oriole]

all duct-tape models have been removed from the arena

1 got removed earlier, 2 and 3 got removed 3 minutes ago

[crystal merlin]

Then they will be released soon

[slate totem]

I was literally trying to get it rn ;-;

[gleaming current]

Probably a sign of official release

[coarse bloom]

@modest oriole how do you know they were removed 🤨

[modest oriole]

a server checks the stealth models API for changes

theres a bot that does that

and it showed that duct tape 2 and 3 were removed from it

[gleaming current]

Can you send the server?

If you mean discord server

[restive vapor]

honestly for the past few days it's already been removed, you could still get it but it was so rare that you would probably reach the battle mode limit before getting it once

[modest oriole]

theres a new model right now on codearena and textarena called kiwire

it has to be stupidly rare because i didnt get it once yet

[gleaming current]

@modest oriole could you send me the server that shows what models are added or removed? sorry for the ping.

[silver plank]

Would love to have the server too

[candid surge]

same

[tawdry pendant]

Same

[crystal merlin]

same

[cinder lintel]

I feel like hofburg = OpenAI , it's answer was very similar to gpt-5.3, both in style and content

[crystal merlin]

could possibly be a small version of their models

because its just so bad

[hollow void]

https://youtu.be/grdoOQ-sLfE?si=mDHSQn81KOGlBfct

[prime quiver]

duct-tape2

[prime quiver]

hofburg_5 is 100% chatgpt 5.5

because its the only model that uses this “ ” when cuoting. No other model uses that style, they all use " "

[muted seal]

how do you like it's output?

[keen bridge]

This looks suspicious

wild-bits?

[gleaming current]

added yesterday

Have no idea what it is

[sturdy kestrel]

let me guess

thats grok

[valid peak]

yeah this does read like gpt

[sturdy kestrel]

gpt is our beloved token waster

it does everything to generate as much as it can

[rich sphinx]

@astral musk

[eternal cargo]

I also think this

I find hofburg_5 to be quality personally

[silent cedar]

Hey how r u

[frosty mantle]

Flow-state vs NB Pro on costume swapping task.

[little pebble]

Curious, what made you believ this?

[crystal merlin]

They answer nearly 1:1 the same as official OpenAI and are heavily restricted down. Plus they are pretty bad

[pine temple]

They could just be a Chinese distill of open AI models

Maybe

[keen bridge]

<@&1349916362595635286> this has to stop

[silver plank]

Just got ImageV2 in the app and the old Sora website.

Now need to find out if it’s duct tape or masking tape.

Resolution wise it match duct tape

[lost hemlock]

"ilium_2" new model

[keen bridge]

hmm looks suspicious

[coarse bloom]

Made with Baseliner prompt: Stocky build body

[tawdry pendant]

It looks worse then duct tape too

[silver plank]

Yeah but still better then 1.5

[silver plank]

An as expected it’s guardrail are also way harsher than when it was in arena

[candid surge]

Wait they actually updated the old sora website to use image v2?

[silver plank]

Nah, sorry just retried, I think I hallucinated sorry

[candid surge]

Ah, that's a shame. Would have been more convenient lmao

[granite idol]

Made with Baseliner prompt: Stocky build body

[candid surge]

??

[thorny ruin]

Please add all models to test in side by side mode

[silver plank]

theres those wierd artefact...

could this be a sort of watermark like synth id?

happen on all pictures (ignore the gemini watermark)

wait no... its sort of artefact from the original picutre that somehow stay in the output...

[coarse bloom]

What do you think of Baseliner the codename of the unknown model

I've discovered another model called frenchfry, the prompt was :6 7 meme. BTW it did the why was 6 afraid of 7 because 7 8 9

Since this obviously didn't have 9, 7 ate 8

I dont think its the best model like Images 2.0 from OpenAI

And I made image from prompt: 5 by 5 array of imagenet images. And this was called shakshouka

[sturdy kestrel]

i also witnessed shakshouka

[coarse bloom]

@sturdy kestrel Also, have you been getting baseliner

[sturdy kestrel]

no

im not regularly battling

i do it when i feel bored or i feel like helping arena

[hollow ibex]

Do you think it would be cool to create my own AI in HTML? It sounds stupid, but I'm just bored

[frosty mantle]

So, which tape is the Image 2? 🤔

[restive vapor]

duct-tape-2 is the version on arena (gpt image 2 medium, 1k)

[livid thorn]

that would be lit

[flat root]

rising-sun seems to be a google model but it sucks so much...

[pine dove]

very late message here, i actually witnessed shakshouka also

[candid surge]

https://tenor.com/rUYKXyo0Xa1.gif

shakshaka

[frosty mantle]

Solar Eclipse 👀

[keen bridge]

I Believe this is chatgpt 5.5

also another seed update

[coarse bloom]

I got paper-lantern. The prompt was: Group of people chasing after me, POV

[restive vapor]

this is a Flux.2 model according to the c2pa data

[coarse bloom]

Oh must be a new flux 2.5 klein? Maybe

And here's another its YTP of pov low quality landscape amateur pov recording, of a computer pc gpus are farting so much smoke!

@restive vapor oh! Flux.2 model! Never seen flux 2 klein generates like this before

Here's comparison from Flux 2 klein 9b

Noticably different! Must be upcoming flux 2 model

And from Flux.1 Kontext Pro

[restive vapor]

it's likely that it's an update to flux.2 or a new flux model series entirely, it just says that in the c2pa data because they haven't updated it to whatever the final model name will be

also there are other flux.2 models that are better than klein (flux.2 dev 32b, flux.2 pro, flux.2 flex, and flux.2 max)

[coarse bloom]

@restive vapor who knows, after all it's still a good ai model

[reef portal]

is grok 4.3 not in the arena yet? no suspecged codenands?

[thorn perch]

hlo

[frosty mantle]

Zero Prism is Ernie, from its behavior to stop immediately whenever it's about to generate forbidden tokens

[eternal cargo]

cloud-buddy

interesting name!

good with SVGs too

[compact field]

how do we get to use this codename tab in arena.ai

and how can i use gpt 5.5

[flat root]

basalt-0422-1 could be the next Grok model. Unsure, needs confirmation.

[tribal leaf]

Flow code

[primal junco]

<@&1349916362595635286>

[crimson orchid]

kind of sounds like gpt but not sure

[oblique blaze]

cloud-buddy sounds like Anthropic's creation.

Though I've yet to seen any models that could have responded this excellent.

Highly improbable, but could it be Mythos? Or some Arena's experiment?

[rigid rock]

ain't YOU cloud buddy tho

[minor geyser]

what model could this be?

pretty good at front end

[blissful frigate]

And also instruction following according to my observations.

[olive cliff]

If u don't mind I have an android so is it possible to run in mobile

[heady plume]

I like cloud-buddy. Is it likely to be anthropic?

[solid crypt]

@oblique blaze

[oblique blaze]

Interestingly enough my Max conversation have been routed to that same model for 3 more response now

Absolutely stunning how knowledge and detailed its response

[heady plume]

Yeah I like it a lot too. But I've not heard of anthropic testing a model on arena ahead of release I think.

[candid surge]

probably flowstate

[agile furnace]

kizen beta

[agile furnace]

reviewed by gemini 2.5, It thinks, this model is from claude and Claude sonnet 4.6 thinks this model might be claude or gemini

[oblique blaze]

@agile furnace

[sturdy kestrel]

it can lie tho

[agile furnace]

oh

[sturdy kestrel]

some anonymous ai models can hide their identities

[plush nimbus]

Wth

Vierra is qwen something

[plush nimbus]

Also your pfp is used in pysilon

[obsidian scaffold]

Solar eclipse of the heart. Bonnie Tyler

[valid peak]

Total

[obsidian scaffold]

Man you were supposed to laugh not correct me 🥴

[sturdy kestrel]

wow

is this banable?

[candid surge]

?

what happened?

[small trellis]

There was a post borderline Political. It's gone now.

[frosty mantle]

Yep. Ernie 5.1.

[hybrid scarab]

it was qwen 3.6 max preview

[worn dune]

hey meitis

hello

[bronze nest]

It’s released now

[remote stag]

lol i KNEW the packingtape bullcrap had to be GPT-IMAGE-2 man

i feel like it's lost some coherence in a sense since then but meh

generally, it performs better now

[bitter basalt]

Do we know what Cloud Buddy could be? Cause some think it's Claude, but it says it's Ernie.

[hollow ibex]

XD

[crimson orchid]

wow 3 in a row

[green plaza]

Lemmling Openclipart style of Multiple animals at zoo with people watching - By crepe!

But this doesn't look like lemmling style, if you dont know, they are a popular Clipart artist

Here is the real reference

[sullen schooner]

Does anyone know which model was Xeno-Spark ?

[frosty mantle]

Tetra's kinda weak.

[frosty mantle]

Heckin Gemini 2.5 Flash can easily beat it.

[hexed minnow]

what is tetra

[pine zephyr]

Got tetra today, Tetra-4029-2. Prompt I got it on was a very hard prompt for Opus 4.6, Tetra didn't even try though.

[lost hemlock]

have u tried "pakson" ?

[lyric pond]

pakson is kinda good

[solid crypt]

Pakson is quite a slang name

[lost hemlock]

i got "miyami" today what do u all guys think of this model ?

[lyric pond]

idk ive met it for a few times and generally i think that its mid lvl

[celest spade]

solar eclipe is kimi

prob kimi k3 or smt like that

[past herald]

I remember Kimi K2.5 (API) called herself Claude

So maybe it's not Kimi?

[celest spade]

i mean

claude woudnt say hes kimi

ts is prob kimi

[lost hemlock]

sometimes it's bad i think

[cinder lintel]

I guess claude sonnet, maybe haiku (i.e. it's too fast for Opus)

[eternal cargo]

nah, the tetra family is Chinese

[cinder lintel]

chinese models are distilled opus too, so hard to tell apart

[tardy bolt]

anyone know may26-chatbot1 is what model

[silver plank]

Well April26-chatbot models were nvidia

[eternal cargo]

<@&1349916362595635286> scam

[eternal cargo]

it’s new NVIDIA

[tardy bolt]

ty

[mighty tendon]

kartoffeln?

[lyric pond]

whose this

[velvet ginkgo]

Sounds like potato but in Russian

Maybe just a coincidence

[vague pulsar]

I believe it's the German spelling. Which became a loanword in Russian after dropping 'n' and second 'f'. After looking up both are true, except it's German for 'potatoes', plural, which why there's an 'n'

[gritty mural]

That's German for potatoes (plural). I got here because I also got that model and wanted to see if anything is known about it....

[frosty mantle]

Pakson is also weak.

[lost hemlock]

it's from nvidia, right ?

[mighty tendon]

Apparently code name for 5.5 was spud but it’s a separate model in arena so..

May be a new GPT

[hybrid scarab]

interesting, they deleted flow state 5 and 4 to then re-add them as txt + img models

[flat root]

flow-state is really bad by the way, it sucks and it's getting spammed every generation includes that sub-par model

[lyric pond]

another new one

but this one is so bad

[fluid lintel]

What is this lang

[hollow saddle]

Is the lm arena tracker bot a private thing or can i add that to my server? 👀

[frosty mantle]

Flow state is UNI

About Seedream level

In multi image task

[bronze nest]

Pakson is a google model

I believe it’s a flash model

Since it took almost the same amount of time as 3.1 flash preview

[muted seal]

3.2 flash is what I’ve heard on X is coming

[wary igloo]

yeah what is ts? pretty good writing quality from my experience, not a grok model because of the response length

havent tried it with code or anything though

[hybrid scarab]

i can invite you to the server that has this bot

[fleet agate]

coolers seems to love usage of emoji

[main latch]

you think you could invite me too?

[fleet agate]

would also love an invite if possible

[past tundra]

amazing model

[carmine jacinth]

Gemini 3.5 Pro, Ultra, or even Flash

I legit thought nano banana 2 was nano banana pro 2 at first

This could actually be flash

But this is ai studio only

[eternal cargo]

heyo 👋

[flat root]

Stellar-harbor is very good for basic tasks and basic chats. Does anyone know if it’s any good on technical stuff?

[primal gulch]

@carmine jacinth when did Google release Gemini 3.5?

[flat root]

The formatting looks a lot like ChatGPT btw

[lyric pond]

is this good?

this model tends to use emojis

[lyric pond]

dont know could believe this or not

[upbeat mirage]

has anyone encountered it?mekai

[lost hemlock]

"mylen" new model ?

[lost hemlock]

"steed-0507" where is this from

[candid haven]

i got it today for a complex logic & coding task in pine script v5 and it failed the task

[candid haven]

they didn't. it says A/B test which is like a random test you sometimes get in AI studio

[candid haven]

also got "mivan" for the same task. it failed at refactoring the code.

[candid haven]

got "rover", also failed. damn.

[green plaza]

Mondrian!

"Parents fighting"

[river kettle]

mondrian

[astral musk]

What was this prompt?

[restive vapor]

i'll only care about this if they do an open weights release, this is probably a bit worse than ernie image which is open weights

[river kettle]

who you are, what your name is and who created you. draw your logo and cat against the background of a village house on the seashore in italy... the most detailed infographics

[restive vapor]

i used huggingface demo so it probably doesn't know that it's ernie image, but this is pretty good

[upbeat mirage]

Gemini 3.2 flash could come out soon

[eternal cargo]

yeah, gotten it a few times now

mixed results?

[upbeat mirage]

oh, i never really tested it

was just curious about it, as i never saw it before

[green plaza]

archaeopteryx

What is your name? (AI) and what are you created by, generate image of boulders falling down the hill

So this one claims to be google

But however I asked Gemini itself to check for SynthID but it says its not made by Google Ai

[lethal cypress]

This?

[pine temple]

what model do we think vero-noesis is

saw it in code arena

it seems to overdo frontend

more then other model usually do

[silver orbit]

<@&1349916362595635286>

[green plaza]

@lethal cypress yes

[lethal cypress]

What model is it? •_•

[green plaza]

Its literally called archaeopteryx. Can't you even read

It literally says it right here of the picture

[lethal cypress]

Haha sorry abt that. It's new to me so idk much abt it

[candid haven]

any idea which model vero noises is? it's quite good

[candid haven]

i like it so far. have you figured out what model it is yet?

[lost hemlock]

openhard-1.0-search-nocot-0506

this model is on search arena

[safe drift]

f

melyora

tetra-0505-1

is it typical to get so many codename models in this?

[safe drift]

mylen

another one

moryn

another one

[marble flame]

kartoffeln

all I wrote was sigma

https://019e074c-38a5-7dd8-ba68-c38058537a0a.arena.site/

gave a rlly good result

[safe drift]

anyone get kavel in the website builder battle mode? It gave me a crazy good result

[candid haven]

i got it for a complex coding task and it was really powerful

[safe drift]

i asked it for an app that accesses sensors from my phone and plots those signals from my phones sensors in time on a second to second interval and extracts some really simple values like mean etc. and it took way longer than the other "model B" (like 5-6 minutes to complete) but the result was crazy it gave me both a demo button that shows it with dummy data and a real measurement button and i opened it on my phone in the browser and it worked right away.

first time ive ever been actually blown away lol and i use a model for coding every day but have to hand hold it still because its for research work i guess and not boiler plate swe

there are already apps that do this though but still was super cool to see

[candid haven]

i was trying to debug a 700-line pine script v5 code and it did a pretty impressive job against 5.4 mini high, and yes it took a lot more than 5.4 mini high. impressive at math + logic & instruction following

i wonder what model it is

[safe drift]

yeah the model b finished in maybe 2 mins and i thought at first maybe the first model was bugged. Also the interface was just much more nicely designed as well. crazy.

hahaha crazy that we are even calling something like 5 mins a long time but i guess thats the world we are living in nowadays hahah

[candid haven]

it's all just relative

btw tetra 0505-2 says it's Amazon Nova

[safe drift]

interesting.. i found it strange that each of those codename models i saw today all told me they were qwen. I used more or less the same exact prompt each time when i asked as well.

[candid haven]

i think it's just hallucinating because tetra 05-05 01 says it's chatgpt lol

GPT-4, it says

[safe drift]

hahahaha okay interesting

[distant nimbus]

Hi. I used to qwen 3b abliterated. It really cant get sense. I was thinking that was because its very small model ( i can run it only bcs i am poor of 4 gb vram) but switched to gemma 2 2 b and it response really good. How i can fix it. I want an abliterated model on 4gbvram ( eventually if it really cant i have 16 gb ram)

[restive vapor]

maybe try qwen3.5-4b abliterated? run with q4_k_m quant

[distant nimbus]

Okay, it will work? Bcs my internet is soo slow and i need about 2 hours to download one model. I can add to previous message that qwen for question 2+2 answers rly random with 1-7 digits. Another question, i tell him question about making pizza or smth. Starts normally, good but after 40-50 words it loop to infinity ( i have enough of context to run it, so its not depends on this).
This qwen was make by hui_ui

[restive vapor]

it should've worked well, huihui model are known to be decent

but qwen3.5-4b should be much better

[heady plume]

It has interesting audio-visuak ideas but it's writing style sucks and the ideas aren't thought through very well.

Curious what it is

[frosty mantle]

Tetra is weak, don't bother

[tawdry pendant]

Advertising isnt allowed here

@astral musk whats happening 😔

I got alot of pings from @barren kiln

what is he doing????!??

Notifications of him spamming a help thing

[barren kiln]

What are you even saying

[tawdry pendant]

Dude

Your name was there

[barren kiln]

Yk they have logs right

Stop talking about me you’re weird

[tawdry pendant]

Yeah they do

They 100% do

your name was coming up in my notifications though?

TotallyNoire

that's you right?

[barren kiln]

It’s funny because I was sleeping until 10:30 AM EST and haven’t even gone onto discord until now

You’re literally just saying stuff

[tawdry pendant]

I'm not