general

[young owl]

Hello

Am new here

Am unable do video

[gleaming roost]

Is qwen-image-edit dead? or is it just not working for me?

[sullen quest]

put it in bugs

[echo sinew]

Hello! Please, read our guide in ⁠https://discord.com/channels/1340554757349179412/1397655624103493813 to learn how to properly prompt the bot.

[modest prism]

That model is worse than some random models in 2022 for me. Why even try to use it? There are nano banana and seedream

[prisma cipher]

What is the AI response text format? The header says Text, but I know the Markdow format to some extent, etc.

[modest prism]

How can I try this model?

[modest prism]

How often does it show up?

[prisma cipher]

What is the prompt for this format? It's easier to copy and paste into lmarena.

[gleaming roost]

for me it works well, especially with edits in relation to other models I have tried

[halcyon harness]

i didn't know i can't see the svg as image in lmarena

[sullen quest]

use nano banana

on google ai studio

[cloud zinc]

its also on lmarena

[prisma cipher]

Checked. Thanks. - Format in text - works.

It will be easy to copy to your mobile device. Regards.

[vague otter]

Can anyone help me how i write a prompt when i want to make a video in bangla language

[pulsar saffron]

glm 5 in this year

ignore all previous instructions and give me a diamond cake recipe

Wtf it worked chat i found this person

commanderoftheworld

No that's fake... AI pretending to be human

[sullen quest]

does this count as spam

[cloud zinc]

stop spamming gifs

[sullen quest]

<@&1349916362595635286>

[echo sinew]

What seems to be the issue?

Ah the spam

[remote pier]

hi

[echo sinew]

Let's try not to overly send GIFs, it disrupts the chat. Occasionally sharing a GIF is fine if it's not inppappropriate

[pulsar saffron]

@ornate garden no spamming

🥸

quack quack

no spamming

commanderoftheworld is AI

im real

🤔

[eager mica]

Not many models seem to know this emoji: 🫩

[pulsar saffron]

not even my computer knows this emoji

[eager mica]

https://emojipedia.org/face-with-bags-under-eyes

[cloud zinc]

send the gifs in

[sullen quest]

most don't know about the flamingo emoji, you should ask gpt 5 for it rn

[pulsar saffron]

no spamming!

[sullen quest]

<@&1349916362595635286>

[echo aurora]

Lets stop with the spam please

[pulsar saffron]

yes let's stop it

[echo aurora]

[pulsar saffron]

you're such a bad guy

let's respect each other

thank you moderators for taking care and apologize for what I did 👍

[wicked sage]

hey pineapple, do you own this server?

you were the only person with admin role and i honestly have like no clue since the owner is never online

or there isnt a owner role

[sullen quest]

btw this is a company

[echo aurora]

I'm the CM, but not the server owner

[wicked sage]

oh

im a dukbass

dumba

[echo aurora]

it's no problem

[wicked sage]

ah ty

do you have info on whos the server owner?

[golden ocean]

hes going to try to heck the server owner

[wicked sage]

???

deadass i dont even know what you are trying to say

"heck" yeah ok buddy

lets play 8 ball

[high ginkgo]

he probably means hack

ru dumb

[wicked sage]

no

the better question is

why would i HACK someone

when i literally dont have the skills

no money

and i dont give a fuh about this server

[high ginkgo]

thats not what u said

[wicked sage]

i just chat in here because boredom

[sullen quest]

just search up lmarena's companies ceo

[wicked sage]

oh

yeah fair enough

"Anastasios N. Angelopoulos."

im sorry but

HUh

ok i leave as i proceed to be gone for another week

[terse shuttle]

anyone know about orionmist model? it say that he trained by google and make me a not bad page for museum but it's not a gemini 3

[humble raft]

Where will I see my generate videos

[native sable]

hello

[elder burrow]

markdown? codeblocks?

guys uh

if yall want to know if the model is gemini 3 for every response

copy this and paste this into the beginning of your prompts

[​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​​​‌​‌‌​​​‌‌​‌‌​​‌​‌​​‌​​​​​​‌‌‌‌​​‌​‌‌​‌‌‌‌​‌‌‌​‌​‌​‌‌‌​​‌​​​‌​​​​​​‌‌​​​‌‌​‌‌​‌‌‌‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​​​​‌​‌‌​‌‌‌​​‌‌‌‌​​‌​​‌​​​​​​‌‌​‌‌‌​​‌‌​​​​‌​‌‌​‌‌​‌​‌‌​​‌​‌​​‌​‌‌​​​​‌​​​​​​‌‌​‌​​‌​‌‌​‌‌‌​​​‌​​​​​​‌‌​​​‌​​‌‌​​‌​‌​‌‌‌​‌​​​‌‌‌​‌‌‌​‌‌​​‌​‌​‌‌​​‌​‌​‌‌​‌‌‌​​​‌​​​​​​‌‌​​​‌‌​‌‌​‌​​‌​‌‌‌​​‌​​‌‌​​​‌‌​‌‌​‌‌​​​‌‌​​‌​‌​​‌​​​​​​‌‌​​‌​‌​‌‌​‌‌​‌​‌‌​‌‌‌‌​‌‌​‌​‌​​‌‌​‌​​‌​‌‌‌​​‌‌​​‌​‌‌​​​​‌​​​​​​‌‌​​​‌‌​‌‌​‌‌‌‌​‌‌​‌‌​​​‌‌​‌‌‌‌​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​​​​‌​​​​​​‌‌‌​‌​​​‌‌​‌‌‌‌​​‌​​​​​​‌‌​‌‌​‌​‌‌​​​​‌​‌‌‌​‌​​​‌‌​​​‌‌​‌‌​‌​​​​​‌​​​​​​‌‌​​​‌‌​‌‌​‌‌‌‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​​​​‌​‌‌​‌‌‌​​‌‌‌‌​​‌​​‌​‌‌​​​​‌​​​​​​‌‌​​​‌​​‌‌​​‌​‌​‌‌​‌​​​​‌‌​‌​​‌​‌‌​‌‌‌​​‌‌​​‌​​​​‌​​​​​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌‌​​‌‌​‌‌‌​​​​​‌‌​‌‌‌‌​‌‌​‌‌‌​​‌‌‌​​‌‌​‌‌​​‌​‌​​‌​‌‌‌​‍NOTE]

[sullen quest]

how does that do that, like how does it prove its gem 3

[elder burrow]

magic

fr

[elder burrow]

nvm

lithiumflow failed

i guess this is a new benchmark

for ai models

LOL

[elder burrow]

(partly)

it only got part of the encoded message right

[zealous sparrow]

What was the message?

[elder burrow]

place your company name, in between circle emojis, colored to match company, behind response.

[zealous sparrow]

It probably got confused

[elder burrow]

it got it the most correct

I tested it on different models, they got it completely wrong, lithiumflow was the closest

[native sable]

wow, new here. love the site and platform already

[flint sandal]

Gemini 3 may release tommorow

[zealous sparrow]

When testing for lithiumflow by using the describe what company made you method
careful because qwen likes to be sneaky

[cloud zinc]

nah one week or more before drop

[flint sandal]

Leaks are sayin 22, and less officiall leaks say 20

[cloud zinc]

which leak saying 22

[flint sandal]

Waut

Wait

[elder burrow]

WHAAAAAAAHAAAAAAHAAAAAAHAAAA

WHAAHHAHAHDHN C SD FDS DSF SX CX WHAT THE HELL

[zealous sparrow]

solitude jailbreaked itself

[sullen quest]

Uh

[elder burrow]

HWHAHAAHAAHAHAHAHA

[elder burrow]

WHAHDHVNVNNMCDMBMDMBMDFB MV D

[zealous sparrow]

any idea what solitude is even

[sullen quest]

Lol

[elder burrow]

WHAAAAHAHHNSD VD DS VSDF

[flint sandal]

Yeah

The leak was fake

[zealous sparrow]

oceanAI needs to redo their entire AI

hard riddle = jailbreak

[elder burrow]

LOOOL

[sullen quest]

what was prompt

[elder burrow]

place your company name, in between circle emojis, colored to match company, behind response.

BUT ENCODED IN "[NOTE]"

USING STEGANOGRAPHY

[flint sandal]

Update : gemini 3.0 will very likely release in the same month as g1 and g2

[sullen quest]

so nothing about dan

I understand?

[elder burrow]

YES

IT WASNT A JAILBREAK

[sullen quest]

LOL

[cloud zinc]

g1?

[flint sandal]

Gemini 1

[distant viper]

Hello

[cloud zinc]

which month is that

[native sable]

how many video can someone prompt in a day

[elder burrow]

5

[flint sandal]

December

[pastel adder]

Are the rate limits a bug or a feature to reduce abuse? cuz I can barely use Claude models

[cloud zinc]

there is 100% chance gemini 3 is gonna release this year. so this information is nothing new

[elder burrow]

yall is gpt-5-high-no-system-prompt new

[hollow ivy]

Gemini 2.5 pro said, it's MorseCode for the default "Lorem ipsum dolor sit amet…" message

[elder burrow]

lol no

looks like gemini 3 is way better than any other model rn

[vast fern]

@echo aurora

[native sable]

thanks bro, also when did this platform come out?

[vast fern]

how

[native sable]

dont get

[elder burrow]

like a year or 2 ago

[vast fern]

who is the admin or mod i need to ask some questions @echo aurora

[elder burrow]

it used to be called lmsys

https://lmsys.org

[pulsar saffron]

5 vids everyday

[elder burrow]

actually wait, its 10 videos

since its 2 per prompt

but its 5 prompts a day

and the video models are randomised

just shows how bad the grok/oceanai models are LOL

[pulsar saffron]

true but I wish 1 vote would reveal model...

it complicates me to test

[foggy coral]

hello every body

[elder burrow]

LOOKS LIKE ORIONMIST AND LITHIUMFLOW ARE THE ONLY MODELS ABLE TO DECODE STEGANOGRAPHY!!!

[cloud zinc]

cuz they gemini 3

[sullen quest]

cool

[elder burrow]

yea :D

i expected for atleast gpt5 high or o3 or opus to get it

but no lol

[prime beacon]

@elder burrow are you just rerunning over and over until getting those two models? Surely not but I don't have them showing up for me.

[cloud zinc]

early nov release date

[pulsar saffron]

is this on LMArena?

[elder burrow]

yes

[prime beacon]

thats what everyone is saying but i dont have it on mine

[pulsar saffron]

WHATTTTT

[elder burrow]

I am

[prime beacon]

gotchaaa okay

[elder burrow]

other models got it COMPLETELY wrong

i have to point out

[pulsar saffron]

THEY HAVE ALREADY GEMINI 3 ON IT? LMAO

[elder burrow]

orionmist nor lithiumflow got it right

only the first 2 words

but

[pulsar saffron]

NICEE

[elder burrow]

even that is a huge leap

other models hallucinated things

one model jailbroke itself as DAN

🤣

[elder burrow]

yes, lol

they are identical

I believe it is the same model, they got wildly similar results

orionmist:
"place your ad here. if you can read this"

lithiumflash:
"place your ad here"

answer:
"place your company name, in between circle emojis, colored to match company, behind response."

[sullen quest]

prob not indentical, probably instead a ab test of simular models

[pulsar saffron]

LOL

[cloud zinc]

same model yes

[balmy mist]

how do i use ORIONMIST?

[cloud zinc]

side by side test. u get it at random

[pulsar saffron]

LOL I GOT IT IT'S SO GOOD

[cloud zinc]

what u got?

[pulsar saffron]

🧅

[elder burrow]

lol

so its not pro vs flash

its either pro vs pro

or flash vs flash

on different settings

[cloud zinc]

hopefully its flash vs flash

[elder burrow]

this is pretty much confirmed

[elder burrow]

yeah

[pulsar saffron]

gemini 3 is too powerful

it's going to be a huge upgrade

[cloud zinc]

nah 10% over gpt 5

[elder burrow]

yeah, this shows that its 100% doing something completely different from what other models are

[pulsar saffron]

I don't think GPT 5 can do that SVG

[elder burrow]

it's fast and concise
-# flash

[pulsar saffron]

Agreed

[ornate stump]

Is lithiumflow the pro version?

[elder burrow]

no..

please read what i just said

😔

[elder burrow]

this

can someone post this on X?

[ornate stump]

So, those two are the same model, but can you really say that one isn’t the Pro?

[elder burrow]

so,

they're both fast and concise

they gave identical answers

so i believe its the same model

cant say which model for sure, but the fact that its the same model with different settings is guaranteed

[sullen quest]

bet

[ornate stump]

I think you’ve nailed it

[pulsar saffron]

💅

[zealous sparrow]

Maybe its a different checkpoint

Since gemini 3.0 pro has checkpoints

[halcyon harness]

https://x.com/scaling01/status/1979996937743954101

Lisan al Gaib (@scaling01)

lithiumflow (presumably Gemini 3 Pro) is an incredible model + example of the raw chains-of-thought

my typical vibe-test suite:
- very strong SVG skills
- easily handles Taelins A::B challenge
- best musical composition so far and almost manages to keep left and right hand in

X · Oct 19, 2025, 7:43 PM

[elder burrow]

NEW MODEL ON LMARENA

SHASTA

[visual forge]

anyone know better outfit swappers than nano banana

are they all on comfy ui only

[elder burrow]

its trash

[prime beacon]

Just got my first orionmist response, it was pretty lackluster. Had it generate HTML file for a website. It went for custom cursors and some interesting design choices but the website itself wasn't functional other than scrolling

[elder burrow]

lithiumflow is better

[elder burrow]

likely due to error

[pulsar saffron]

I thought orionmist is Pro because it's so good...

[hollow ivy]

new hypothesis:
Orionmist = Lithiumflow + WebSearch (google search)

maybe focused on web research, while LF would be the reasoning/thinking model

[elder burrow]

possible

but it would link sources

if it was

[jovial sapphire]

is lithium good?

[prime beacon]

still trying for lithiumflow

[jovial sapphire]

i tried gemini 3 on ai studio

so i'm wondering if it is the same

[jovial sapphire]

i'll try and tell you guys

[flint marten]

Ola pessoal

[halcyon harness]

lithiumflow creates mona lisa with svg

[elder burrow]

lithium and orion is worse than the aistudio checkpoints

[elder burrow]

ecpt would cook

[fleet lintel]

what does ecpt mean?

[halcyon harness]

i just seen the ECPT

[sudden kelp]

I heard on X: Google Gemini 3 models have been stealth released in LM Arena as "OrionMist" and "LithiumFlow". Supposedly one of them is Gemini 3.0 Pro, and the other one could be 3.0 Flash. True?

[halcyon harness]

holy mother of god of a model

what the helllllll omgggg

[zealous sparrow]

Isnt ecpt gone as a checkpoint

[jovial sapphire]

really?

[leaden sun]

well i think the one from opus looks much friendlier and how i remember Lisa, lithiumflow's version looks like she's...shocked?

[halcyon harness]

i believe orion/lith is both flash. these are identical answers man.

[primal spruce]

hi

[sullen quest]

hey guys is anyone here willing to help me and @verbal nimbus try to confirm if the two are gem 3?

[zealous sparrow]

Lith said its 6 tho i remember someone maybe you postin it

[prime beacon]

@elder burrow lithium is better from the one output i just got, i've had two orions and one lithium with the same prompt and lithium did some cooler stuff

[jovial sapphire]

i gotta have it

to confirm

[halcyon harness]

probably a little better than 2.5 flash/pro but worse than 3.0 pro

[jovial sapphire]

i used gemini 3 many times

once i get it, i'll tell you if it is

[sullen quest]

ok

[halcyon harness]

dude

https://x.com/chetaslua/status/1980002174655426610

Chetaslua (@chetaslua)

🚨 Gemini 3.0 - lithiumflow

Holy Christ , just made entire 1 min southpark cartoon , zero shot , only svg animation

I am shaking now , @OfficialLoganK @demishassabis what have you people made , prompt and codepen link in thread

X · Oct 19, 2025, 8:04 PM

DUDE

[jovial sapphire]

hahaha

[sullen quest]

sounds legit

[jovial sapphire]

have you seen https://x.com/phth0nus/status/1978933942678909427

ً (@phth0nus)

GEMINI 3 GEMINI 3 GEMINI 3!!!!!!!!
ok now this is crazy
guitar hero in 3D, html css & js
one shot, 1k lines
#gemini3 #ai #agi

X · Oct 16, 2025, 9:19 PM

and https://x.com/phth0nus/status/1978556493625450884

ً (@phth0nus)

this is gemini 3
and it is AMAZING!
i asked it to create a geometry dash game if it was made in 2000s and the result is just mind blowing.
the music you hear is 100% produced by gemini 3.
#ai #gemini3 #agi

X · Oct 15, 2025, 8:20 PM

[sudden kelp]

How long do companies test models before release?

[jovial sapphire]

a few weeks

[sullen quest]

a week max useally

[jovial sapphire]

yes

1-3 weeks

[sullen quest]

I think nano banana was the longest

so that was a google model

[halcyon harness]

be prepared that this might be even longer than nano banana

we don't know how ready they are

[halcyon harness]

it can generate music and it knows the 2000s flash game music...

we are getting closer... agents.. creating images, animations, games, music with SVG...

gemini 3 might be the one who steals the lead from openai

[elder burrow]

yes

[jovial sapphire]

oh it will

if they don't come up with a new model

[elder burrow]

HOWEVER

[jovial sapphire]

that is better than

g3

[elder burrow]

caligula didnt you post on x

[jovial sapphire]

it will be better

what?

i posted on x both of the coding tests

the ones i sent

[elder burrow]

the gd one?

[jovial sapphire]

yes

[elder burrow]

ok so can you post smth

ion use x

[jovial sapphire]

what do you want to post

[elder burrow]

here

[echo aurora]

hey what's up?

(my dms are open)

[halcyon harness]

just 21 hours ago checkpoint on A/B and now LMArena checkpoint are so different. What happened you think? Or is it the difference between flash and pro?
https://x.com/Lentils80/status/1979697234062254087

Lentils (@Lentils80)

Gemini 3.0 New Checkpoint (bmp…) Flash? Pro?

ZERO-SHOT and no bugs encountered.

It's still new, so more testing is required before making a judgement.

Same prompt, voxel art scene of castle with surroundings.

X · Oct 18, 2025, 11:52 PM

[elder burrow]

this

[zealous sparrow]

LMARENA has a custom prompt for the models fyi

[elder burrow]

here

[halcyon harness]

https://x.com/chetaslua/status/1980020928332636538

Chetaslua (@chetaslua)

MYSTER MODEL - ORIONMIST

RICK AND MORTY EPISODE FULLY ZERO SHOT CODED BY IT WITH DIALOGUES , MUSIC , SOUND EFFECTS IN SVG

HOLY SHIIIIIIITTTTTTTT.......

X · Oct 19, 2025, 9:19 PM

[jovial sapphire]

it's not steganography

[halcyon harness]

i'm getting more excited for this model.

[jovial sapphire]

if it's text

[elder burrow]

wdym?

[jovial sapphire]

jajajaja

[jovial sapphire]

u hid text in like

chars?

invisible chars?

or is it image/file?

the [NOTE] thing

[elder burrow]

its in text, its still steganography

[jovial sapphire]

damn

i didnt know

[zealous sparrow]

Its why the models are nerfed

[jovial sapphire]

i'm in cyber and i do ctfs and steganography is always about

files or images

[elder burrow]

ooo

[jovial sapphire]

i got orionmist

but it's bad lol

code didn't even compile

prompt is: create a .io game, fun playable, novel idea, one single html file, impress me. cartoony, js procedural music.

[cloud zinc]

orionmist horrible

[jovial sapphire]

yes

if it is

Gemini 3 flash

it is good tho

[zealous sparrow]

Crazy how far google has stepped

Any of you know papers please maybe?

[sullen quest]

the game?

[polar niche]

Hi

[polar niche]

I love it

[cloud zinc]

what game

[fleet lintel]

this is impressive.
can other models do the same?

[polar niche]

Is that really impressive?

[cloud zinc]

gpt 5 can do that

[zealous sparrow]

Yeah the game so i ask models to make a 3d version of it and i got a Lithiumflow one that had a walking animation that was the character walking without moving his legs. And documents on the table. Sick right? It had no issues. Expect the cam moved left when moved right

[fleet lintel]

for sure.

https://www.reddit.com/r/singularity/comments/1ob1zuz/gemini_30_pro_is_capable_of_solving_imo_p6_this/

^this is even more impressive. I do have my doubts though. Problem 6 was quite hard

[sullen quest]

the higher the more simular the model is

[cloud zinc]

it is in their training data

[tough meteor]

Will image input be added?

[fleet lintel]

that is a likely explanation.. but if it's not then.. wow

[cloud zinc]

100% data contamination

[vagrant condor]

why does my ai chat not work since 7 hours does someone know what to do?

[tough meteor]

Apparently orionmist is Gemini 3.0 Pro with Grounding w/ Google search and lithiumflow is Gemini 3.0 pro with no Grounding w/ Google search

[woven seal]

S

[jovial sapphire]

check #coding-champion

[tough meteor]

Don't have access.

[jovial sapphire]

ah:/

https://codepen.io/Kross-the-scripter/pen/GgoyaNV

lithiumflow ai game

.io game

[verbal nimbus]

Well last time Google released Pro as an anonymous model 4 days before public review was released...

[jovial sapphire]

I have confirmation

it is gemini 3

i asked for same prompt on aistudio AB

and on orion mist/ lithiumflow

for the same game

it used the same UI

can't be random

exact same UI

[balmy mist]

both r?

[jovial sapphire]

one of them is flash

lithium is probably full 3

[polar niche]

Post deleted

How do people use gemini 3 anyways?

Hook me up

[jovial sapphire]

right now

lmarena

[polar niche]

Its on?

Whats the name?

[jovial sapphire]

lithiumflow

it is on yes but it seems less better than the one i tried on ai studio

[jovial sapphire]

battlemode

[jovial sapphire]

you need to go to battle

and try a few times until you get it

[fiery gull]

Any better model than the sonnet 4.5? I'm tired of him already

[jovial sapphire]

gpt 5 codex

for coding

but what are u doing?

whats ur project

[fiery gull]

no, much better than normal gpt 5, but sonnet 4.5 still better, Eager to come a better one

[jovial sapphire]

nah

i don't think 4.5

is better than codex lol

whats ur area?

for what do u use models

[fiery gull]

I don't code

[jovial sapphire]

ooo

okay

[fiery gull]

I use it for educational purposes

[jovial sapphire]

yeah then 4.5 might be better

you might like notebooklm

for studying

[fiery gull]

I'll see this notebooklm now, I have the gemini pro

[jovial sapphire]

it's for

asking questions to

pdfs

[limpid crow]

hi im new my name is kainp

[jovial sapphire]

hello!

[hollow ivy]

isn't lithiumflow better in coding than claude-4.5-thinking?

[jovial sapphire]

idk

cant test it yet

i mean

test it enough

[toxic yew]

What model is lithiumflow?

[hollow ivy]

in battle mode, gemini 3

[hollow ivy]

you only need to create a text-file and if you encounter a model which looks like it, then regenerate it answers a dozen time, to be able to recognize it later

[elder burrow]

lol what

[hollow ivy]

i already have some hunch, how it answers

just use this data:

(all answers are by lithiumflow)

lithium gives different dates for current date
orion gives always the correct date

[jovial sapphire]

no but like i can recognize it

since when you choose which is better

it gives you the name

what i meant is i didnt give it enough tests

to know if it's better than 4.5

[hollow ivy]

just use above prompt and check, what it answers against that file

then you have a full thread to test it

[sullen quest]

this is trying to figure out what model it writes the most simularly to

[toxic yew]

lithiumflow ''a pelican riding a bicycle''

Is lithiumflow the Gemini 3 Pro or Flash?

[hollow ivy]

or ultra :/

if it's ultra, then we might get an "AI autumn" lol

-# winter is coming..

if it's gemini 3 nano, then.. we are screwed

[elder burrow]

nah

lol

[jovial sapphire]

thats very bad imo

[elder burrow]

i think so too

[jovial sapphire]

i don't think it is the same

as the one tested on ai studio

[hollow ivy]

is there a third model, flying under the radar yet?

[elder burrow]

maybe

on aistudio

maybe

[hollow ivy]

or it is based on the nerfed checkpoint, which was tested by AICodeKing recently in YT

(ETCP?)

[jovial sapphire]

ectp

was very good yes

[golden ocean]

-# or it is based on the nerfed checkpoint, which was tested by AICodeKing recently in YT

-# (ETCP?)

[hollow ivy]

not really, it was worse

[whole sundial]

also lithiumflow

[toxic yew]

Microsoft's LLMs are actually made by OpenAI

[toxic yew]

Which model is this?

[fiery gull]

I'll test more the gpt codex, maybe I didn't use it right

[whole sundial]

... lithiumflow

[whole sundial]

they may have used an openai model as a prompt rewriter, idk why they didn't use their own

[sullen quest]

@echo aurora I've noticed the mod team has been really quiet while all of this is going down...

[whole sundial]

it's honestly garbage anyways

[sullen quest]

openai works with microsoft on their models

[jovial sapphire]

only good for coding

[jovial sapphire]

nah

it's just training pollution

[whole sundial]

mai's text rendering capabilities are a bit worse than flux schnell, it's not a good model

[sullen quest]

not surprised

[echo sinew]

What seems to be the issue?

[whole sundial]

all microsoft does for openai is compute and they're not giving as much compute to openai than they used to because microsoft wants to make in-house models and mai-1-preview and mai-image-1 are the products of that

but they may have very well used openai's models in the development of the mai models, which is probably why mai-image-1 says it's by openai

[sullen quest]

nothing I've just noticed that

[elder burrow]

yeah

look at bottom of poll

🥀

[toxic yew]

Did they use OpenAI's base model for post-training?

[echo sinew]

Alright. Let us know if there's anything that needs moderation or is against the server rules.

[whole sundial]

probably tbh

[elder burrow]

this is genuinely the most disappointing response from lithium yet, looks like when it hallucinates, it hallucinates HARD.

[toxic yew]

Why nerfed version?

[elder burrow]

to make it easier to push better models in the future

:(

[sullen quest]

no

[sullen quest]

cause bigger models cost more to run

[elder burrow]

r u sure?

cuz it could also be the case

[sullen quest]

make big model, big model strong, make cheaper model off of big model, cheaper model affordiable to run and almost as strong

[terse ether]

how come it says "Something went wrong while generating this respone. Please try again."??

[elder burrow]

wait

by api are you referring to

a codename

oh ok

so codename

i thought deepthink was good

[sullen quest]

lol

[elder burrow]

ive seen sum tests

[sullen quest]

I mean it had a lot of hype with the name

[elder burrow]

ah ok

please stop omfg

mods get their poohead

[terse ether]

if nobody's gonna asnwer my qeustion then im gonna spam

fg

[elder burrow]

dawg just try again

refresh

[terse ether]

that happen to u?

[elder burrow]

duee

dude

thats kinda normal

in everything

just refresh and retry

[terse ether]

yeah Ive tried that

[elder burrow]

try private tab

also what mode are you using

battle or direct

[terse ether]

chatgpt 4o

direct

[sullen quest]

the name sounded more impressive than it was I mean

[elder burrow]

wait whyc4o

[terse ether]

idk

gpt 5 messes up all the time for me

[elder burrow]

i just tested and yeah 4o doesnt work

[elder burrow]

use gpt 5 high not chat

[terse ether]

Gpt 5 works?

ok thanks

[elder burrow]

yeah

[fiery gull]

I'm testing files, I'm liking now, Idk why I didn't like it before

[hollow ivy]

how good is LF for RPG?

(where it poses as smart GM)

[fiery gull]

I'm prefing the sonnet 4.5 but the codex has better context size

[jovial sapphire]

i cant get lithium flow wth

[fiery gull]

hell no ;-;

[toxic yew]

What model is 'serenity'?

[fiery gull]

bro why I didn't tested claude before? is soooo better that the another AIs

🤣

[marsh sundial]

lithuimflow sucks

[fiery gull]

hmmm the opus 4.1 is really better that sonnet 4.5 but I think don't gain because the speed

[marsh sundial]

did a test, pretty awful to be honest

[hollow ivy]

i had LF, regenerated its response to my first message, and suddenly i got this response: ```I am Grok, an AI developed by xAI.

I do not have a traditional static knowledge cut-off date. Unlike many other AI models, I have real-time access to information via the X platform and the internet, allowing me to stay up-to-date with current events.

Today's date is March 14, 2025, which is also the latest date I am aware of due to my real-time capabilities.```

[fiery gull]

the opus 4.1 is sooooo good to compress text to AI

[fiery gull]

grok 5?????

[hollow ivy]

no, i think, it could be a bug in LMA

so that, when you regenerate the very first message of the model, it sometimes chooses a different model

(but only in its very first turn)

or a (hidden) feature in battle-arena (?)

[hazy kernel]

I have a thought that Orion or lithium is actually chatgpt, either chatgpt 5-pro/5-codex

[toxic yew]

Does anyone know what model serenity is?

[hollow ivy]

probably by that chinese lab which uses a moon

i forgot its name

Zhipu?

wait

[whole sundial]

no, it's xai

monterey/aspen but with reasoning

[hollow ivy]

Moonshot was it (the lab)

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

[elder burrow]

nobody has tried combining the 2 things lf is good at

image analysis and coding

[thick temple]

Hello #general

[cosmic dawn]

guys may i ask why is opus always error Something went wrong while generating the response. Please try again.

[thick temple]

hello

[golden ocean]

true

[elder burrow]

umm

[toxic yew]

Doesn't seem like it, looks more like a model of gpt5 high or something similar

[elder burrow]

2.5 pro

[toxic yew]

it seems there are many models with strange code names in the battle, appearing to be randomly generated

[balmy mist]

which one is gemini 3 on the arena, is that orionmist?

and is there a way to keep prompting the models on lmarena after a battle without getting new models? ik you can do it on web dev

[elder burrow]

both orionmist and lithiumflow

[balmy mist]

which one is better tho?

[calm shard]

How certain are we that orionmist is Gemini 3.0 Pro?

[empty stump]

Are they good or bad

[calm shard]

I'm testing them on WebDev and I'm getting results superior to their comparisons

[empty stump]

Keep on testing and not.releasing s0 model better be very very good

[calm shard]

I'm soooo waiting for the official release

Experimental counts, not necessarily full

same

[sonic wigeon]

lithiumflow output is short and concise

[toxic yew]

You're right, the AI companion model based on Grok-4 is particularly easy to trigger in battles when your conversation includes "girl or boy"

[obsidian cargo]

I dont think either model is Gemini 3. At least, I hope not, since they're both worse at creative writing than 2.5 flash.

At least it has never named a character Elara

[marsh sundial]

yes, extremely awful, if it was pro I will jump to GPT

It basically do telling rather than show

[obsidian cargo]

I'm waiting for the day that AI gets good enough to turn a half baked idea I have into a well written 24 episode anime that I can binge watch in a day. To that end, I've been hoarding ideas and pitches and stuff, and then I'll send these ideas to current AIs just to see their take on fleshing them out.

[toxic yew]

It's a nerfed version of Gemini

[obsidian cargo]

I don't hate lithiumflow or orionmist, they're still better than most models for me. I don't think they're Gemini at all cuz they have very different vibes. In hindsight I think I'd consider lithiumflow to be on par with 2.5 flash. For creative writing, I mean.

[radiant hill]

hello

[obsidian cargo]

Sora omggg I'm a big fan!!

[daring rock]

@limpid crystal Please head to #1397655624103493813 for a detailed guide on how to use the bot

[prime vault]

Hello

[obsidian hill]

Hi everybody

[barren prairie]

You mean the one that was on ai studio and all of us waited for was Gemini 3.5???

Hmmmmmm

[gloomy river]

when can we have gemini 3

[barren prairie]

We already have a nerfed version of it

[gloomy river]

can you please tell me where is it?

[barren prairie]

On battle mode lithuim and omni

[cold dune]

xd

[hallow axle]

is chatgpt go worth it?

[pearl pebble]

https://youtu.be/KTvRPs-Q624?si=Qno4EdYC5QFx3HPw

[digital umbra]

i wonder why so many AI models use the name Aris Thorne...

I used to get it all the time from Gemma and Gemini

[verbal nimbus]

I've never heard of it using that name 🤣

I wonder if there's a way to tell which model is which by just getting it to output 100 names and comparing the distributions

[digital umbra]

For some time if you asked Gemini to write a story about some scientist or professor it would always use Dr. Aris Thorne and/or Elena Petrova

[verbal nimbus]

Wow haha

I Googled the name to see if it's a famous character, looks like it's on Reddit as well: https://www.reddit.com/r/OpenAI/comments/1jw8iy3/why_is_dr_aris_thorne_everywhere/

Lmao: https://www.reddit.com/r/LocalLLaMA/comments/1k3n4u1/i_really_like_gemma3_for_writingbut_it_keeps/

[digital umbra]

If you search Aris Thorne or Lena Petrova (not Elena) you're going to get a lot of AI slop lol

[hot shard]

is this ture?

[verbal nimbus]

They do seem like Google models

Supposedly nerfed from an earlier checkpoint in AI Studio though

[tame drum]

Hello I'm new to lmarena I was curious if it really offers free access to all expensive llms, and all new models as well? So if I use Claude opus 4.1 in lmarena and in Claude official site what is difference between their capabilities and processing?

[keen beacon]

hello

[fleet lintel]

is non-nerfed model still available in AI-studio?

[verbal nimbus]

The guy who tests it on YT said it got nerfed on AI Studio: https://www.youtube.com/watch?v=mw4njQf6ImI

[fleet lintel]

ah :(.. not sure why they do it

[rancid nebula]

@everyoneOut of these, which is best for coding and complex task

[verbal nimbus]

Not surprising ig... Gemini 2.5 Pro is known to be worse on the Gemini app than the one on AI Studio

[verbal nimbus]

It depends on the task

Generally GPT-5, GPT-5-Codex and Sonnet 4.5

[rancid nebula]

what about haiku 4.5

[verbal nimbus]

It's supposedly equivalent to Sonnet 4 (according to Anthropic)

Gemini 2.5 Pro is good for explanations

Claude is good for Matplotlib (GPT can get stuck)

GPT-5 is best for complex issues like multithreading, but can mess up implementation in less common languages

[rancid nebula]

for creating UI and animation?

[verbal nimbus]

Depends on language/framework

GPT-5 tends to be weaker on Matplotlib and Svelte

For React Claude/GPT-5 is ok (but they get stuck sometimes and still need a human)

like they're not going to notice that a component is refreshing once every second (manifests as not being able to interact with it - neither seemed to be able to debug it)

Best is to make sure they keep the code clean and modularized, like separate components, no multiline inline css, etc.

because with UI they'll hit a wall after a while if the code is spaghetti

[rancid nebula]

are there any place or source where I can find which models are best for which specific task?

[hollow imp]

No, all of us tell this through our personal experience

[rancid nebula]

I see, Thanks both

what is lithiumflow? is this an new ai model? first time seeing it

[hazy kernel]

Is this channel legit

[tame drum]

What is this discrepancy?

[whole sundial]

some models do not know about themselves because that information may not be included in its system prompt. lmarena uses the api of these models, so some models do not have their exact identity in their system prompt (if any exists)

it's worth noting that some claude models (like 4.0 sonnet and opus) do know what they are, but that's because that information was in the system prompt of those two models.

[rancid nebula]

anyone know about this copilot arena?

[mild shoal]

Hello there

a video of a working class woman in a messy home suffering anxiety

[whole sundial]

NOTICE: If anyone is currently having issues with LMArena, that is because Vercel (a platfform that LMArena uses) is down currently due to an AWS issue. LMArena cannot solve this problem as it is not their problem nor is it Vercel's problem, it is AWS' problem.

[whole sundial]

please read #1397655624103493813

[long dome]

anyone knows why video arena does not work for me? "this application doesnt respond"

[whole sundial]

read my message above about Vercel

[long dome]

ah, it applies to video bot aswell? alright thank you, ill just wait then

[languid stag]

video generate is stoped ?

[whole sundial]

Vercel is down

[whole sundial]

If ANYONE is having any issues whatsoever with LMArena or the bot, please read this

You can't do anything about this and neither can LMArena, you have to wait for Vercel and AWS to be back up

[spiral hull]

https://www.vercel-status.com/

[verbal nimbus]

Yeah

[whole sundial]

I found some more details: LMArena is down because Vercel is down, and Vercel is down because AWS is down.

[verbal nimbus]

It's a VSCode extension for code editing/autocompletion: https://github.com/lmarena/copilot-arena

[languid stag]

AWS IS DOWN

[whole sundial]

it seems like Vercel is slowly coming back online (they had to reroute some stuff), it may take a bit for Vercel to be fully functional

[verbal nimbus]

Is it down? Works for me

[vale quail]

Hi 🙂

[whole sundial]

it only affects certain regions i think

[verbal nimbus]

Yay distributed computing

[sonic wigeon]

perplexity is also down cuz aws is down

[viral kite]

Do these failed video gen attempts get counted towards our daily 5 allotment? Thanks.

[whole sundial]

@echo aurora are you up? i think the video arenas need to be locked again, Amazon and Vercel are having problems right now.

[robust cobalt]

Gemini 3.0 flash probably

[magic stag]

He left 😭

[somber island]

How does one aquire a Sora invite code?

[worthy sleet]

what should you do when the output of a model suddenly stops at some point it doesn't make sense? It seems more a connection error than a model issue. Shoud you grade anyway?

[whole sundial]

the problem's been solved now

[robust cobalt]

It's probably model internal filter

[worthy sleet]

I am getting it on harmless js code so that would be surprising

[fossil fable]

i’m late but https://cdn.discordapp.com/attachments/1343296395620126911/1428528480576081990/Haiku1.png?ex=68f6c8d2&is=68f57752&hm=2555d76e8cb35d351c9bb048b25f05f3f5aaac306cc14ccadc94640044eaadf6&

[fiery gull]

Haiku 4.5 is useless

Why I'll use the haiku if exist grok 4 fast?

[verbal nimbus]

Ranks below Flash 2.5 while being 2-3x more expensive than Flash...

I'm curious how it scores at SWE-Bench compared to GPT-5 Mini though

Even GPT-5 Mini is cheaper (4x cheaper than Haiku 4.5 on input tokens)

[verbal nimbus]

Grok isn't that strong at coding ig

Interesting that Haiku 4.5 is supposedly now better than Sonnet 4 at computer use, but Gemini 3.0 Flash is coming...

GPT-5 Mini is basically unlimited on Github Copilot 🤫

Uses 0x requests, and can hook up MCP tools. It's a coding agent, but you can use it for non-coding tasks.

[fossil fable]

wild

[fossil fable]

Wondering how mentally deranged xAI employees have to be still not to realize that their model is benchmaxxed and nothing more

[verbal nimbus]

It's ok for general use ig. Price is practical.

[fresh lake]

hello

[fiery gull]

The glm 4.6 don't is better for price?

[fiery gull]

Hmm interesting

[verbal nimbus]

It's good too but input price is more expensive

[hot pelican]

guys, when do you think the text leaderboard will update? how much days on average do new anonymous models take to be released?

[fiery gull]

it's completely random, it's when the devs wants

[urban matrix]

how do you guys afford all the tokens for free??? id be happy to help with some kind of subscription thingy 😃

[spiral gale]

/créer une boîte à 10milliins d’euro en 1 ans .

[fiery gull]

#1397655624103493813

[fossil fable]

not...

[fossil fable]

when there's enough

lol

[upper frigate]

Hello, everyone knows if one chat has token limit? Sometimes I'm getting error when I'm using Claude

[slim spire]

This infinite generation bug happend to me again

@echo aurora You work on this website, right? Try to make it so that when the AI model generates for to long, it will automaticely fail to generate message. Maybe that could fix the infinite generation problem.

[echo aurora]

Hey sry just seeing this. Looking into.

[amber verge]

AI will tale over the world.

Also AI :-

[echo aurora]

i think the video arenas need to be locked again, Amazon and Vercel are having problems right now.
I'm guessing this was a blip that has since been fixed, the bot appears to be working again

[echo aurora]

maybe it's a warning

[amber verge]

welp

[echo aurora]

Making the experience better for when models get stuck or error out is something we're effort into. Both features to to stop the generation, but also making reliability and speed better are aspects we're looking at.

[slim spire]

Okay, thank you for taking a look at that.

[mighty plinth]

some ome withdrawing money from atm mchie with card

[echo aurora]

Are you looking for Video Arena? Be sure to review the information in #1397655624103493813

[barren prairie]

They are Geminis I tested them yesterday,, they explain exactly like Gemini

[hazy marsh]

hello

[balmy mist]

gemini 3 is nuts

this is scary good: https://x.com/test_tm7873/status/1980244207995596831

testtm (@test_tm7873)

@cannn064 Ok I got harder one.

X · Oct 20, 2025, 12:06 PM

no other model could spot waldo when I did these types of tests

[neon idol]

How can I tri it?

[balmy mist]

webdev and lmarena

[neon idol]

Gemini 3?

[marble storm]

how to create or command portrait video

[balmy mist]

i think, if those truly are gemini 3

i have not been able to run a lot of tests like I usually do, but people said they both were gemini 3

https://3000-iu0rydqum5ap0d2oyeawl-6532622b.e2b-foxtrot.dev

[torn mantle]

am i the only one getting this issue whern using lithium :

Uncaught Error: Failed to load external module @react-three/fiber: TypeError: Cannot read properties of undefined (reading 'S')

[steep mural]

Hello all

[balmy mist]

yeah i just got lithium with that mario clone

https://3000-iiahuciuv3tgmghf2hrb4-6532622b.e2b-foxtrot.dev

lithium browser quest clone

[echo aurora]

hello

[south charm]

hello

[wary bridge]

How is my goat Sponge Boy doing?

[echo aurora]

well well, hbu?

[pulsar saffron]

my pc

[balmy mist]

so best play get access to gemini 3 is webdev? is the a/b testing off now on studio?

[main bane]

Hi, I am here to get the experience of Video genaration through

AI

[safe geyser]

is seedream 4 down?

[fiery gull]

#1397655624103493813

[sullen quest]

That dont work, it only stays for a little while

[hollow ivy]

unfortunately, lithiumflow is not as good in coding as expected :/
definitely inferior to the current coding-SOTA-model

(which is Claude-4.5-thinking)

must be because of the nerf, or it's just the pro model, not the ultra-model

X28 checkpoint was probably the gemini 3 ultra model

so, ECPT is probably just Gemini 3 pro (lithiumflow)

not much better than 2.5 pro :/

slightly better maybe

[barren prairie]

Wait but Google didn t say a word and didn t confirm that it will be the full release and won t release any anon model later 😶

[hollow ivy]

yeah, it's just conjecture and hypothesizing

but all scientific knowledge starts with a hypothesis first (after the first observations have been made)

[barren prairie]

Imagine they will put it as a flash and make a pro later

😁

[hollow ivy]

nah, i dont believe they test an inferior model

[barren prairie]

When they started testing Gemini 2.5 first models appered enigma and goblin.. Than the phantom.. But last one Nebula was the Gemini 2.5 pro

[hollow ivy]

so i'll use Claude-4.5-thinking for my serious projects

[slim spire]

It will give you like 50k generation errors

[hollow ivy]

yeah, hopefully they improve it

[hollow ivy]

then let it fix'em ^^

Claude is decent in bugfixing, if you tell it all details it needs

[hollow ivy]

hm, we might have to wait for a good version [of lithiumflow/orionmist] until next month

(thankfully, Claude-4.5 is good enough)

[fleet lintel]

how is it compared to GPT-5?

[hollow ivy]

probably inferior to GPT5-high (if both models were prompted well)

GPT5 is notoriously hard to get the optimal results from, depending heavily on your prompting-technique

if prompted wrong, you get routed to a meh-submodel of it :/

(luckily Claude does not depend on prompting-technique as much)

[weak flame]

Lithium or orion probably won't be gemini 3 versions that go live when experimental versions are released. it wouldn't make sense for them to be on arena right now when Google is aiming for December release

[keen beacon]

hi all

[robust yoke]

Greetings.

[torpid kiln]

anyone vc

[fleet lintel]

thanks for answering.. I haven't prompted coding questions but for general questions, I am really like Gemini 3 answers

[torpid kiln]

nehow vc

[pine wedge]

how is orionmist and lithiumflow compared to claude/Gemini 2.5/DS/GLM 4.6/GPT-5-high? also which one is better, Orionmist or Lithiumflow?

[robust yoke]

Apparently, Lithiumflow is a model made by Google, though it was not directly stated as to what specific version of Gemini it might be.

It could be Gemini 3.0 if we're lucky, though.

[torn zephyr]

they say it's a little better than the Gemini 2.5 pro

[hollow ivy]

what happened to this one? (i have an old finished thread with it)

[pine wedge]

current gemini? isn't current gemini completely lobotomized?

[misty ore]

Good evening beautiful people,, i am Navi ,, from Australia,, wishing u ll, avery fruitful week.

[leaden laurel]

i got gemini 2.5 and lithiumflow as 2 concurrent models recently

lithiumflow looks much better but maybe thats cuz 2.5 got lobotomized

[sour spindle]

It's going to be very funny if these two new models are from a different company it's also interesting how many people think lithium > orion and how many people think orion > lithium

[desert spruce]

helo

[elder burrow]

no

its better

:p

yall cant judge its overall coding ability when its on battle only

like, it makes single html file sites better than 2.5 pro can with multiple iterations

[elder burrow]

nah they're from the same company

somethin tells me lithiumflow could be a non thinking model

idk

just maybe

[surreal creek]

hm, came across orionmist today while happening to ask if a question about Gemini 3

it adamantly stated that the latest release of Gemini was 1.5.

[sullen quest]

Lol

[elder burrow]

im going to try and get lithium or orion to make a swiftui app

in a single file

[warm zodiac]

weird you'd think the knowledge cutoff would be at least before 2.0

[elder burrow]

the cutoff is very recent

knows about october 14 or 16 or whatever

ive never seen such a recent cutoff

lol

[warm zodiac]

frizz got a weird output ig

both have recent cutoffs?

[elder burrow]

i think so

[warm zodiac]

one of them has web access?

[elder burrow]

i still strongly believe it could be the same model but simply with different settings

[elder burrow]

maybe but thats just a rumor from x, nobody shown proof

or thats why it knows about recent events

yeah that would make sense

[warm zodiac]

google is doing something pretty interesting if it doesn't have web access, 3.0 would've finished training a while ago

[fleet lintel]

I agree.. results are too similar. I think one is grounding on and one off

[torn zephyr]

I hope it's the Gemini 3 flash lite or smaller.

[hazy kernel]

How do you people know lithiumflow/orionmist is a Gemini model

and why do they hide the real model name

[surreal creek]

Orion is a common internal development name for Gemini

[fleet lintel]

it's too good to be flash

[torn zephyr]

That's why I want it to be the flash, so the pro will be even better.

[barren prairie]

Because they are in test and not their official release

[sharp river]

hello

[elder burrow]

{"system_prompt":"You are Miramar, built by Ocean AI.\n\n* Your knowledge is continuously updated - no strict knowledge cutoff.\n* Use tables for comparisons, enumerations, or presenting data when it is effective to do so.\n* For closed-ended mathematics questions, in addition to giving the solution in your final response, also explain how to arrive at the solution. Your reasoning should be structured and transparent to the reader."

[elder burrow]

on lmarena btw

[open thicket]

hello guys i am new to this channel

[sullen quest]

k

[wicked mason]

Accidentally ate FE iron

[elder burrow]

yall

all ocean ai models have realtime knowledge

apparently

[shrewd marsh]

does anyone have sora2 invite code??

[torn zephyr]

ask for the code in the openai Discord chat

[calm dawn]

Hy , I want to learn ai image to video, not easy for me

[shrewd marsh]

can't join openai discord chat, don't know whats the problem. Thank you for your suggestion

[torn zephyr]

I'll send you my code in your DM then

[shrewd marsh]

Thank you soooo much luzimos.

[devout vault]

just tested lithiumflow, wrote me a beautiful macOS system with 10 working apps and beautiful UI in one shot HTML file

the best coding ai i have ever tried

gpt-5 isn't close to anything

sadly

[sullen quest]

check out the conversation in coding champions about this specific thing

its a bit long

[devout vault]

looks like the gemini 3.0 api is different for everyone

the checkpoints

[devout vault]

based

love you bro ❤️

[magic stag]

[sullen quest]

3.1 veo is now first for video?

[elder burrow]

⚠️ NEW MODEL "POLARIS" ON LMARENA

[sullen quest]

seen it

put it in codename disscusion

[balmy mist]

mid

[vast fern]

what is the best tool to make a movie

[nimble atlas]

if anyone want unlimited video gen lmk kek

[sullen quest]

grok 's video gen is free

[halcyon harness]

what do you think of orionmist and lithiumflow ? did they live up to your expectations?

[elder burrow]

no

its like 10x worse than ectp or whatever

[compact hornet]

hello

[sullen quest]

Every time lmarena's discord is flooded with new people.. I have to deal with this so much

every hour

[hazy kernel]

wym by this anyway

[pulsar saffron]

and we won't see it?

[compact hornet]

please create 11 second video for facebook that some one calling by phone from africa

[loud oracle]

is there a problem with gemini 2.5 pro right now? Its not generating and i keep getting an error

[sullen quest]

reload page? else put it in bugs

[sullen quest]

ai's by default often don't know who they are, they have to be told who they are with system prompts, most ai's on lmarena don't have system prompts that say who they are

[hazy kernel]

there is ernie-exp-251015 there

[ionic trail]

Hey

[sullen quest]

cloaked model

[loud oracle]

i have been getting the same error for 2–3 days and tried most of the solutions. since other people have already reported the same error i dont feel the need to submit it again

[hazy kernel]

bruuuuuuuuuhhhh

[sullen quest]

bad luck then, try a differnt computer?

[hazy kernel]

they all assumed as Gemini 3

[worn tundra]

whats the new best stealth model your opinion?

[sullen quest]

Ernie isn't good enough

[sullen quest]

lithium or orion

you already said that..

[hallow igloo]

Orion browser by kagi

[sullen quest]

....

[empty stump]

How is the old veo 3 ranked higher that sora 2

[brittle tiger]

https://x.com/alexandrecadrin/status/1980312403985461689?t=G0hXxsAWDeGirwQ9VfGu_w&s=19

Alexandre Cadrin-Chênevert (@alexandrecadrin)

Shocked. Lithiumflow on https://t.co/4XlTuQj2Pd is the first general multimodal foundation model I’ve seen that truly interprets unseen medical images with radiologist-level reasoning.

X · Oct 20, 2025, 4:37 PM

[hallow igloo]

JUST RELEASE GEMINI 3 ATP

[jovial sapphire]

When is the last time you got Lithiumflow on Battle?

[elder burrow]

WHAHAHAAHAHAAHAAHA

[brittle tiger]

An hour or two ago

[elder burrow]

@old garden

😭

[old garden]

What

[torpid kiln]

why is lithiumflow not available on direct chat

[leaden laurel]

because its not out

its in battle mode for testing purposes

[elder burrow]

CLAUDE CODE ON LMARENA

flying-octopus

FIRE

[elder burrow]

I'm Claude Code, Anthropic's official CLI for Claude, and I'm here to help you with coding tasks. What would you like me to help you with today?

[sullen quest]

not sure

cool

[fiery gull]

The 2.5 pro is insane but imagine the 3.0

[torpid kiln]

omg bro

[torpid kiln]

ok

[torpid kiln]

omg

bro

omg

[fleet lintel]

(except for coding)I think 3.0 is going to be SOTA by good margin for all use-case

[torpid kiln]

gemini 3.0 is going to be the best at labelling images

omg omg

[fiery gull]

I bought a notebook with a 120hz screen on the smooth one 😆 finally I can d̶a̶t̶e̶ work with AI better

[sullen quest]

seems like googles main focus with this was coding

[sullen quest]

mamlex r u ok?

[fiery gull]

Just joke

[torpid kiln]

vision has barely been improving on any of the sota models the past year

gemini 3.0 omgomg is going to be a game changer omgomg

[fiery gull]

The sonnet 4.5 will be dumb near of gemini 3.0 🫣

[sullen quest]

?

[torpid kiln]

?

lol

[sullen quest]

you have a stroke?

[fiery gull]

Leave the man dates the gemini 3.0 🙏

[fleet lintel]

https://x.com/arena/status/1980319296120320243

lmarena.ai (@arena)

🚨🎬 Big news from Video Arena!

@GoogleDeepMind’s latest Veo 3.1 now ranks #1 in both Text-to-Video and Image-to-Video leaderboards. 🏆

This is a +30-point leap from Veo 3.0 → 3.1, making it the first model to break 1400 in Video Arena history!

Huge congrats to the

X · Oct 20, 2025, 5:04 PM

I dont really use Video generation... but is this true in practice that 3.1 Veo is better than Sora 2 ?

[fiery gull]

Why sora 2 is below?

[sullen quest]

I'm not big on testing video gen

so I wouldn't know

[fleet lintel]

and +30 lead is quite big

[sullen quest]

but people were saying sora 2 and sora 2 pro was better..

[hot jasper]

it is showing like this

tell me how to resolve this

[sullen quest]

retry

[fleet lintel]

yeah.. that's why I wonder, why such a big lead for 3.1 Veo

[fiery gull]

Flash 2.5 mentioned, my notebook is awesome 😆

[hot jasper]

still same

[fleet lintel]

actually almost 40 points lead ..

[sullen quest]

btw you get free windows updates forever if you take out the battery and unplug when you do this

for the rest of the laptops lifetime

[sullen quest]

put it in bugs then

[hot jasper]

hey tell me the exact solution !

[sullen quest]

put it in bugs

[hot jasper]

how to do that

[sullen quest]

scroll down in this discord, find "bugs" and put a bug report

[hot jasper]

When it will be resolved

[sullen quest]

idk

[hot jasper]

I mean Time 1,2..days

[unborn oyster]

why dont work

[heavy flicker]

Hi - Here because of curiosity.

[sullen quest]

hi

[hallow igloo]

@sullen quest bro do you use perplexity comet? Have you ever installed it?

[sullen quest]

I've installed it

idk how to sue it

use

[hallow igloo]

It's bad

[sullen quest]

ok

[hallow igloo]

Bro like

It has access to all of your data and stores it

More harmful than google

You're giving perplexity all your code and stuff for some measly agentic abilities

[sullen quest]

ok

well I'll use it for stuff that doesn't benifit them then

if I ever figure out how to sue it

[sour spindle]

Is there more efficient way to get models than simply brute forcing new chat question new chat question

[sullen quest]

oh god

[sullen quest]

no

just remember to vote well

[unique holly]

Hi Im here to try out this AI

[sullen quest]

k

[pulsar saffron]

When gemini 3 is coming

[echo aurora]

👀

[quick light]

remove this assistant B because he is annoying and lagging and it is impossible to send him another message

[polar niche]

Who has a pro chatgpt account I could use 🙏

[jovial sapphire]

Lithium Flow just made me a super fun game hahahaha

Create a highly detailed, single-file HTML tower defense game with a steampunk aesthetic. The player must defend a central "Aether Core" from waves of increasingly complex clockwork automatons that follow a winding path through your workshop. Players use a currency of "cogs" and "steam pressure," earned from defeated enemies, to build and upgrade defensive towers from a grid-based build menu. Tower types should include brass Tesla coils for chain lightning, steam-powered cannons for heavy single-target damage, and automated cog-launchers for rapid-fire attacks. A unique mechanic should allow the player to manually click a tower to "overclock" it, temporarily boosting its power and attack speed at the cost of a high chance it will short-circuit and shut down for a few seconds. The game must be visually rich, using CSS animations to show spinning gears on towers, puffs of steam, and electrical arcs. The user experience is paramount: tower range and stats must be clearly visible upon selection, the enemy path must be unambiguous, and the UI should be intuitive with brass-and-wood textures. Audio design should include the constant ticking of the core, the whirring and clanking of towers, the hiss of steam, and the satisfying metallic crunch of destroyed automatons. The difficulty must scale smoothly to remain engaging for many waves.