#general
1 messages · Page 181 of 1
most stuff on twitter about it is fake
logan has tweeted about it
if u even know who he is
all i did is ask for a source
yeah bro, did u get a lobotomy
is gemini 3 better than chatgpt 5
i aint, cuz i told u
logan posted about it
ok thanks
by 50 miles
no sh*t?
duh
X IS UP NOW
nah
IS this site down ?
SWE is just being stupid trust, i seen it do goated gens..
sonnet better
antrophic paid SWE trust, like i swear. How do you NOT see so much twitter praising and you still put it below sonnet 4.5
SWE is not real
facts
twitter is now back at my website
twitter are paid shill
what about reddit
also bro, putting it below GPT 5.1 is Diabolical. the hell is SWE doing, mane?
GPT 5.1 has no creativity at all and uses the same style for websites and SWE still put it higher than gemini
who tf made that chart
yo is the viddeo arena down?
chatgpt prob
ok be honest, do you think SWE rigged it, I hardly doubt that GPT 5.1 beats gemini 3..
its a "verified" benchmark
verified my a-
why?
benchmarks are not realistic
just curious
also fax
Idk I stole it from x.com
when it debuts on LMArena it goin to dethrone every model so deada- SWE cant do sh
ui maxxed
google is something stupid.. my Free gemini account has "gemini 3" but my paid account doesn't. 🫠
AI slop maxxed
wydm u have gemini 3

on gemini app, i see gemini 3 option now
Bro as long as they know how to balance things, good use of space, responsiveness etc.. its ok, you can tell it what style to use
damn
WHY IS GEMINI 3 NOT ON AI STUDIO
is chatgpt down?
yes
as like as grok and gemini.
different launch schedules probably. I would guess 4 more hours max before it appear on ai-studio
show.
I don't though..?
i think it just rolling out
It picks from one style tho
same here.
challenges.cloudflare.com is down.
proof
bro theres no "gemini 3" on gemini app
ok but show screenshot
Also i googled SWE is rigged
and i see articles people yelling about it
So true...
it says 2.5 pro for me
Bro saw that in yesterday dream lol
Force it to use different ones..
too lazy.. i only have discord on desktop
hmm, thats weird
Use windows button+ print screen for screen shot
show screenshot
or ur lying
Also i expected you to get gemini 3 model because you got that random thinking model
riftrunner is pretty sure still alive on LM and this didnt even last 6 hours
that's true!
why are you lying tho
Shiit
why would i lie.. this is not even worth lying
Bro is not lying bro saw that in future
wasn't polaris alpha still alive when 5.1 came out?
yeah but pretty sure it dead now
did you hear that it apparently got swapped with canvas model
idk
idk tbh
Depends on the region i guess? I don’t see it in my app..
its what @quartz light said
probably some rollout started.. will take few hours to reach everyone
Yes for a couple of hours then it was took down..
guys is that true
black screen
you check console
gemini 3 release gonna be delayed cuz of cloudflare lmao
google has its own servers
so no
oh ok
im already logged in
breh
i still think they use cf for some stuff
i cant log in, i keep spamming continue with google @quartz light
what is cf because i have the brain of 2 peanuts
cloudflare
ah
like every cdn would be CF protected
BROO
CANVAS IS SUPER GOOD
check out my ps2 menu
this is identical to my riftrunner gen...
THE CIRCLE IS GOATED THO
maybe riftrunner not being bound by lmarena sys prompt is buffed
the circle says its better than riftrunner
cuz riftrunner couldn't make good circles for me
@quartz light says the new canvas model is riftrunner, but maybe that its not bound by lmarenas sys prompt, and kind of tweaked by google its goated??
okay
this is better
than the one we had
on cursor
@quartz light whats your evidence the new canvas model is riftrunner
is voting working on lmarena?
twitter is up for my country now
gemishe
when was the last time cloudfare was down for this long?
no idea
people call SWE bench flawed after seeing gemini 3 perform bad in it
SWE bench drama is real
they put gemini 3 under gpt 5.1 btw 💀
100% gemini 3
if this is riftrunner, and we somehow prove it. I am glad my goat gets to shine and i never lost belief
riftrunner wasnt good
this your gen?
maybe they tweaked it and put it on canvas?
What's gonna happen to gemini 3 x28
yes
exactly
x28 was FAAAR better
than riftrunner & the other checkpoints we had
like literally no other model was close
✅
✅
can you prompt this Bomb defusing simulator something like keep talking and nobody explodes but singleplayer. Dont forget a bomb exploding sound.
i will compare riftrunner to it
do you not know what KTANE is bro
This is crazy
can someone show how look like x28
ok wait
i have a really good one saved
please show the prompt as well
i couldnt find it
but
Cloudflare down
this one is really good
Cloudflare Error
riftrunner gave me a better one im sure..
cant show it since codearena is down but
it's actually
try it
its good
this was riftrunner, use vsc since codepen wont work
Mhmmm different UI styl
but explosion sound
is better in canva's version
Gemini app, canvas
yooo what the hell
they keep changing the model on canvas
yeah idk whats goin on
yeah graphics are better in riftrunner..
wait i didnt look closely
💀
pelican's head is good but the body
this is me lol
this is one of the best ive seen
Wait gemini 3 pro is available rn on aistudio build?
ohh i forgot lol
its ok (x
yes that one was the best showcase
it was x28
ye
@jovial sapphire
wait ill try the same prompt on canvas rn
i think
nooo
cuz cloudflare
not ai studio
o
canvas
canva on the official gemini site?
seems to have better outputs
cloudflare isnt letting anyone see aistudio apps rn
yes
ah on cursor it's Gemini 3 pro preview. so it's not meant to be the full version
riftrunner winning with the UI style is diabolical tho
Try this again maybe? I need to see a quality gen.
Gemini 3 is on canvas
can you try this prompt again: Bomb defusing simulator something like keep talking and nobody explodes but singleplayer. Dont forget a bomb exploding sound.
i want to see them beat riftrunner
alr
this is the preview version?
Cloudflare down panic?
when's gemini 3 coming out?
people are saying its rolling out on mobile rn
old news now.. we are not waiting for gemini 3.5
What even is cloudfare?
I really like the grok 4.1 but, bruh, 3 trilhion parametres elon must? I don't want a new gpt 4.5 :(.
isnt grok like... bad?
you got the gen back yet?
yeah, the kimi k2 thinking is soooooooooooooo better 🤣
cloudflare is almost fully healed
and have people still calling the ai chineses inferiors, the chineses AI are going really good
its getting better.. In my opinion, latest one is decent
lmao are they faking their placement on lmarena again?
true
yes.. its decently ok
0 for free users yes
as always
whats "gemini-3"
this is it?
imma try it rn
yeah try it
Where is that?
It's certainly fun to talk to, but it's garbage at coding.
what u talk with it about
was it the same prompt on the riftrunner version
bro, I don't trust in lmarena arena more, I want to belive it, but bruh
yep
yeah same
also why is supergrok so expensive... aint no one paying $30
we shouldnt trust SWE for putting gemini 3 under gpt 5.1
grok looks powerful on benchmarks, until you try it yourself...
lmarena needs to fix this problem... i think it is not that hard to manipulate votes
Exactly. Had to double check I was using the right model because its results were so bad.
i think theres some dirty cheating behind xai manipulating the votes..
lol, we need to test itself the AI, for me the sonnet 4.5 is the better of all, but the benchmark said the gpt 5.1 is better (I really like the gpt 5.1, but the sonnet 4.5 for my use is better)
I think elon musk bribed lmarena
gpt 5 deada- beats gpt 5.1
WHATTTTTTTTTT, HELL NO HELL NO
This is literally just RIFTRUNNER!
yeah its very similar
i prefer using it in canvas rather than lmarena
the 5.1 humiliates the 5.0, the 5.1 is the 5.0 good 🤣
why are new grok models always appearing so fast on lmarena? it makes it so obvious that the votes are rigged... it always takes a while for new models to appear on the leaderboards but not for grok
holy peak.
im making a minecraft clone, lemme see
think this is confirmation riftrunner is on canvas
Can u tell me about the pfp i saw in every reel in insta
the grok 4.1 is like the llama 4, the version 4.1 is pre trained to win in lmarena and X, you understand? the MOST messages in lmarena don't is code or some thing is more casual messages
im so sick of corruption being everywhere... cant have the truth anywhere... not even on ai leaderboards lmao
https://019a93d2-e61b-75e0-9c63-7d54a4b3b681.arena.site [RIFTRUNNER LMARENA]
https://gemini.google.com/share/5be2d18d3e87 [GEMINI CANVAS]
Tell me this is not the EXACT SAME!
makes sense
so like whats the best ai model right now? chatgpt? claude? gemini?
Minecraft clone, one-shot prompt https://g.co/gemini/share/ce56c6398eb7
@zealous sparrow @quartz light
broken
if you dont dismiss the gui it wont let you do anything but move [you cant dismiss it]
agi is heckin clossed
i told it to make a mobile version tbh
are u on pc
yes
cuz it works on my phone
Bro
you can move but you cant dismiss the gui on the pc...
lol
Lmarena back y'all?
wait imma make world generation for it
It seems like the bot is working again, yes.
nvm it broke
The model is allergic to edits?
idk, when i asked it to fix the mobile controls it flopped the whole thing
Can we run games in gemini?
yeah saw it on the gemini server
Broo how do you run this in gemini
ii think my gen is done
i made another ps2 gen
how do y'all have gemini 3?
This game so funny
I think i got gemini canvas too yeah
Please share your cool creations with Gemini 3 on canvas.
red screen of death kinda sucks
riftrunner did it better
cuz i told it to generate the whole PS2 startup, then leading to the red screen of death
if u make a new prompt for just that
it does better obv
ah ok
What the heck mannnn!
Bro can anyone tell how to run this in gemini
Is it free or do we need to pay
I have a gemini pro
u will be able to access in 2 hours
2.5 pro canvas
the actual model is not out
still bein tested
So bunni is a test user?
Confirmerd or rumors?
So people with gemini pro can access this for free
source
Intuition
no its like testable if you go to 2.5 pro model on gemini and pick canvas then prompt
Ok George bush
not free. limited
In AiStudio, models are usually available without limits
https://gemini.google.com/share/cc64c91db81e
Prompt: recreate the ps1 startup and use this as audio myinstants.com/media/sounds/ps_1.mp3
if your asking that links to a myinstants audio file, it didnt make it itself
I hope the limits on free API calls will be no less than 250 requests, as with Gemini 2.5 pro.
GUUUYS
i was first
i hope they remove the confidental tag so it doesnt end like kingfall
preview only
GEMINI 333
omg
check AI studio.. it's out
OOOMMMGGG
its rate limited rn
YESS
careful because the model is kind of ratelimited
BRO I SKIPPED NEWS
oh... it's a bit expensive. i kinda knew and predicted it
YOOO
looks like nano banana aint coming today
1.6 times more expensive
already reached
surprised, that's more expensive than gpt 5? Better performance, but i expected cheaper
yup. 1.6 times
I can't even test it.
is this an accident?
the demand is too high
😭
The demand for Gemini 3 is very huge so dont expect to use it often
I am not surprised at all.. it's much better and google knows about it and hence charging the premium.
I mentioned that few days back that it is a possibility
today is see this: "LMArena didn’t respond in time" and nothing is happening. Why?
@echo aurora Waiting for gemini 3 to debut on the leaderboards
I hope Gemini 3 will actually be as free as Gemini 2.5 on AIStudio.
We hope these RateLimits are just due to overload.
true, but i expected it to be cheaper because it's cheaper for google internally
nah it wont, its more expensive
release for me too
bro gemini 3 doesnt work
ratelimits probs due to overloads yeah
I can use it so, if you cant. It's overloaded
ahh im late lol
it says error for me
Gemini 3 Pro is the next generation in the Gemini series of models, a suite of
highly-capable, natively multimodal, reasoning models. Gemini 3 Pro is now Google’s most advanced
model for complex tasks, and can comprehend vast datasets, challenging problems from different
information sources, including text, audio, images, video, and entire cod...
i think google is thinking that people are just going to pay more because of higher quality. And they decided to do so
Today is going to be a very interesting day.
its overloaded
WE NEED NANOBANANA PROO
damn i have a meeting during the event 🙁
NO WAY+
NANOBANANA PROOOOO
I FEEL SPECIAL
Why is Gemini 3 in the Confidential tab?
Models from this tab usually disappear quickly.
Which is it?
Let's goooo
Yeah the OVERLOAD ratelimit is crazy for rn
They just released it and it was confidential till release
Why would they put it up with confidential?
lol
The requests per minute quota is insane rn
Can’t even use it
Prob forgot to take it off
Ya it is
IT WORKED FOR ME
Hey why it says for nano banana it's new??
FINALLY
Is nano banana pro???
Super expensive
GUYYSS ITS WORKING
ok i gen cant connect to vpn💔
It's OVERLOADED the demand was INSANE
GEMINI 3 PRO IS WORKING
what the hell
I don't have it,
really
adoption is more important ... if they think that expensive model will result in customer going to competitors then they would have taken some losses. But they know that they have a better quality product and higher price is fine
bro its making the PS2 startup for. me
lets compare it
Wheres gemini 3 anf nanobanana 2
No, you’re tripping. That price is too expensive.
https://ai.studio/apps/drive/1i2jPvT0JEhegBf7SkJPBi6r-g6wbI7eo
Here's the PS1 startup. The audio is supplied by myinstants, it didnt find it.
told u already man. only gemini 3 pro releasing today in preview
flash and nanobanana will come in december general release
me too!! but shhhhh...CONFIDENTIAL
What should be the first prompt
lol.. yeah it's confidential 😄
"You've reached your rate limit. Please try again later." .. i cant even 🙁
Lame
It doesn't matter, your answer will be:
anyone have a working model? Did you get a response?
It got overloaded already
🥳
nano will prob follow soon
bruh page not found
atleast I have paid version to use. i m already out of tokesn on free one 🙁
what
i cant look at it, just send the code file
I'm done
its gone
no its overloaded
its still there
react
react
React
WHAT IS THE DIFFERENCE
aalright
what is AIS Applets
gg
idk what that is
AIS: artificial intelligence super
when is lmarena gonna get gemini 3 pro
soon, but for right now the model is OVERLOADED and unusable
Bright shiny toy sure but how do they know the model then to vote on it
I apologize to Elon for anything I said about grok 4.1, at least it actually did something! Even if it's all lies and hallucinations. This is pathetic, I'm shorting the fk out of Google. You have one chance to make a good impression in the AI world, and this has been a total failure so far
where nano banana
I have reached the rate limit with 1 prompt 🤨
Bro. THE MODEL HAS LIKE 1000000 REQUESTS and you expect it to function perfectly [overloaded]
How many prompts did you do wtf
It got Overloaded due to the omega demand
is cloudflare still down?
My request wasn't even answered and I received the limit.
Zero
Idk
When will gemini 3 release
For people who reached rate limits: No you didn't they are high due to the models overload.
yeah it Overloaded 20 SECONDS after release
Gemini 3 is AGI
Google: 📉
use your TPUs google... give us the free access
trying 😮
You are just mad because they made a SO good of a model, everyone jumped onto it and overloaded the model...
hope itll be good
Gemini 3 has been pulled from the confidential category, It will not end up like Kingfall 🥳
hello
It's so good they should have called it Gemini god.
reboen with gemini 3
buuuuu
the demand is simply too insane
just use via api
ahhh on Cli might work???
We all need to wait for google to get their servers used to the demand.
i would like to use it but always go on overloaded
They were not prepared
or api
New error
no
just pay them and all works
i had no single issue yet
it didn't tell me im out yet
I guess AGI is the new code symbol for "not functioning"
ill steam in #1340554757827461215
finally also in desktop brower for me
???
bro deosnt listen to anybody💔
until a min ago it only worked on mobile for me
Is Gemini 3 already available via API?
AGI ain't free homie
no, i can use it fine in roo code, come #1340554757827461215
Now not confidential. Just new
Also as we predicted, this model had to have a paid factor
Gg guys
gemini 3 is free on AI studio dawg
tell them to block it on build mode abuse
thanks!! will try
Oh
thank you
oh you saw it now?
its working
nope
NO PROBLEM!!!!!
build mode is working chat
dang its good
Serious? 😑
no single error yet
use build app
i know
i swear to god why did cloudflare go down today💔 😭
soon on my api for free too
i meant for everything but coding
i genuinely cant use vpn
I am taking a break... even on my paid account, i m getting quota issues.
i think its only in build mode currenlt
This is LMarena, not Poe, we expect and demand greatness here. Humanity depends on us and our mission for AI greatness. We take our responsibility seriously 🫡
Is it telling you to try again tommorow or is it a different one
Socialists need to wait for lmarena to use Gemini 3 for free.
Someone send gemini 3 in #1372229840131985540
and build mode takes soo much tokens... no hope to recover anytime soon
why does the output say "Gemini 2.5 flash"
or just use in free apis like mine (soon)
Do you know any method for unlimited movies from Veo 3.1?
I think you can try it in Google ai studio
simple: pay
Lmao 😂😂😂
yoooo
Nope
bruh
Keep making new accounts?
seems to work!! thanks
build mode is using gemini 2.5 flash
lol where did the other one go
Ok but where is nano banana 2. I hope at least that will be free
?
How to create a Google account without a phone number and without verification?
I can't see this in normal gemini
Where?
Didnt say it was easy...lol
google api
why are you spamming the link
no you can change
how
cloudflare itself has uam now, crazy
Nah 2/12$ 🤧
Yeah, it's on Google ai studio, it always comes late on the app
@vivid coral
We wont see gemini 3 on LMArena too soon, simply no API yet
is the leaked bechmark real?
I FINALLY CAN USE VPN
Who has managed to lost his virginity with Gemini 3? I did 🤪
yeah it works in app builder omggg i love google
Well, write how to do it.
testing nowwwww
gemini 3 currently works in app builder
It's by Google, why should we trust that. We are the only benchmarkers here that are official in the AI world.
The normal gemini 3 seems to be coming back
Visualize planetary orbits around a sun, with ellipses traced by moving points representing planets at varying speeds.
Who else has received this exact message?
the frick is the context this low 😭
Do you know any method for unlimited movies from Veo 3.1?
me, on the main chat area. but if you go to builder it seems to work.
Literally AGI
its like so broken rn
Yes everyone is getting it, it crashed already
no the overload caused it to freak out
Do you know any method for unlimited movies from Veo 3.1?
no, i made new chat and it worked
Do yall guys have the original 3D solar system PROMPT for EVALS??
Do you know any method for unlimited movies from Veo 3.1?
ong IT NAILS IT AT FIRST TRY
Free users only get 1 prompt per 5 hours I guess
good
xD i use my api key since a while now 🤣
i hope they prevent the free abuse
Do you know any method for unlimited movies from Veo 3.1?
MODEL IS FIXED NOW

woorks
well, guess i'm not allowed to share where to get free gemini 3 api, sry
YOOO
you need a billing enabled API key btw
I wonder what the limits will be without an API key
Do you know any method for unlimited movies from Veo 3.1?
a few prompts
pov: i use it in roo code 💀
Oh s*** it's working some now 👀😲
im asking it to make os
@quartz light gemini 3 is solving your bench im seeing it do good so far
Only a few prompts for free.
works!
@quartz light this is the result without code execution enabled
Well I'm happy to pay if it works
Hmm, maybe in dm?
Even if it doesnt say what the model is? Its gemini 3?
im probably wasting my api limits but im gonna try it on ro bloxstudio scripting
Can someone clarify free rate limits
Longing Google....stock is about to double 📈
@quartz light Gemini 3 result with Code execution enabled for the [NOTE] bench
I bought 10k of SOL
Ong we going to the moon
dude i guess i was proven wrong.,.,
this real?
lol, bad gemini, bad
i call BS on the SWE bench, definetly higher
this also happened on LMArena for riftrunner sometimes so its natural
now worked
🙂
maybe tool calling is not as good?
youtube like in 1999
is it better than sonnet?
Thinking is set low or high?
high
from our testing yeah, SWE bench are a bunch of liars or they got a worse model for evaluating
Hoping the opus 4.5 😈😈😈😈
oh
lol
If this isn't AGI then what is
Does "Media resolution" have any effect when working with text?
was riftrunner better than 2.5 pro?
ofc
they removed gemini 2.5 pro on ai studio
some old dall-e 3 image of youtube in 2007
chat is this too much to ask for LMArena's devs?
anyone tested if it's also in gemini cli available?
pro
or pro
nah its gonna in december
They will not do it, simple. Not only adults bench AIs, and this could get them sued.
with flash
well huggingchat doesnt have censorship 😭
they didnt receive a lawsuit
they should write, on their TOS, any content that the user generates that gets them in legal trouble is their responsibility, not LMArena's.
they supply the AIs
or maybe not remove guardrails, but allow jailbreaking
To a certain degree sure, but if it gets too ilegal, no.
Hell no, is away, but is geting close to agi 🤤
LuaU is a difficult language for AIs
wdym
like allow normal jailbreaks I found on reddit or smth else
k then
is gemini 3 on AI studio with limits?
dawggggg gemini 3 in lmarean when
i mean, personally luau is actually a really easy language and if it excels at html and or python there would be no reason that its kinda mid
tho for granted python and or html are common
Yeah, but heres the kicker. AI companies train AIs for python and nhtml. Less for LuaU
@echo aurora
this is just a preview btw
but other than that, they really should add the edit, stop generating, rename chat buttons immediately lmao
When the API releases...
SWE bench is not better then that
when will that release
Couldn't say
Hmm despite this appearing, I can still use gemini 3.
yeah its real leaked one
https://i.snipboard.io/doDiZB.jpg
.svg of spider web
is grok 4.1 good ??
grok 4.1 goin to fall off bro
votes were in some way manipulated
dang, first prompt was bad results, let me reprompt
grok is 1 on lmarena
It is pre trained for lmarena and X, in inteligencie is worse that grok 4
i thought that was banned after llama incident
You need to do good prompts
why is it suspicious lol
Gemini 3 is amazing
the point is, if you want a good ai, it has to be creative
Good evening, I just joined the server so I want to say hello 🙂
between grok and gemini we are so freaking spoiled mang
Grok feels fraudulent
guys Gemini 3 on build mode is worse than the chat mode
i just tested it
chat mode is better
Gemini 3 coming to lm arena when now
cuz its using limited token there
grok just won because its got a good personality, arena isnt a iq test
Sys prompt!
It's failing the 6 finger test.. Oh man.
dude
do u have billing enabled or smth
cuz ur the only one who can use it on chat
chat mode is free now
go use it
fr
I can use it on chat too
I don have billing fyi
GUYS THE WAR IS OVER!!!!
grok 4.1 died already
barely 15 points ahead
rofl must suck to be elon rn lol
CRAZY
100 Points AHEAD!
i think when the full model is actually released and not benchmarks of it you will not see as many failures regarding six fingers etc
whats the prices for gemini 3 pro btw?
a kidney would be enough for 1M tokens
18$ above 200k [output]
1.5 times more than gemini 2.5 pro price
ah thx, will keep below for free users then ❤️
4$ for input
In vision it's 100 points over next non-gemini model
NB1 lets you edit photos of famous people as long as it's SFW. Does anyone know if NB2 is going to add safety filters to block editing public figures completely?
gemini is the goat at vision tbh
no flash yet?
definitely prob
only 50 points over 2.5 lmao
THAT SHI IS CRAZY
flash is going to steal the show, mark my words!!
yeah, flash is in december
You clearly don't understand how much big is data
flash comes later
flash and nano banana next month
It fails the 6 finger test but its probably because, the test makes AIs think weirdly
could you tell it specifically to count each finger left to right
All high end models are competing in a range of 10/20 points even less and then gemini 3 comes with over 50/100 points difference
AIGI IS FUGGEN CLOSED
It's an issue a lot of models suffer with. Probably will get better with time..
GPT-5-high is "65" .. sixty-five points behind gemini 3 ?? what happened to OAI?
wow.
dayum
are we really ready for g3?? lol
mind you not they asking gov for loans
actually you can feel how good the model is. It is giving so much better answer in general
check official notes
how are ur SVG looking good
skill issue
model card confirmed
46% HLE with tools. OMG!
that's in december i heard..,.,
pump those numbers to 90% we need AGI
does anyone know rate limits of gem 3 in ai studio
you are late
is it like another vscode or cursor clone?
AI Studio
Is Gemini 3 not yet available via GeminiAPI?
ohh i saw it there already
?
API
its free for now
we dont know
Oh, indeed. 🙂 I prefer AI Studio for speed and stability, but certainly cool for A/B tests.
tuff
lmarena thinking is low
wait when's it officially releasing tho? i got myself a google ai pro subscription thinking it was gonna release on gemini.google.com
december
this is interesting
Been awhile, but posting new results. https://x.com/DillonUzar/status/1990813243405647898
Context Arena Update: Added Gemini 3.0 Pro Preview (Thinking, 11-18) to the OpenAI-MRCR leaderboards. It establishes a new state-of-the-art in context performance, taking the #1 spot on all our AUC leaderboards and for nearly all pointwise scores.
All results at: http://contextarena.ai
The 2-needle results are incredible, maintaining a 99%+ pointwise score at <=128k tokens and a strong 72% at 1M. Even on the difficult 8-needle test, it achieves an impressive 54% pointwise score at 128k.
The performance curve is interesting: on 2-needle, it's a nearly flat line of near-perfect recall up to 128k. On harder tests, the degradation slope steepens past 128k, with a clear performance shift in the 128k-256k range (likely around the 200k mark seen in prior Gemini models).
It dethrones the previous champions: openai/gpt-5:thinking at 128k and the top google/gemini-2.5 models at 1M.
2-Needle Performance (@ 128k / @ 1M):
- AUC: 99.4% (vs 96.7%) / 81.2% (vs 78.3%)
- Pointwise: 99.0% (vs 95.0%) / 72.2% (vs 68.1%)
(going to have to retire 2-needle soon)
4-Needle Performance (@ 128k / @ 1M):
- AUC: 84.7% (vs 74.1%) / 49.9% (vs 49.5%)
- Pointwise: 80.9% (vs 70.6%) / 34.3% (#2, behind Gem 2.5 Flash Thinking)
8-Needle Performance (@ 128k / @ 1M):
- AUC: 67.8% (vs 50.3%) / 34.5% (vs 28.0%)
- Pointwise: 54.2% (vs 40.0%) / 24.5% (#2, behind Gem 2.5 Flash)
A significant leap over all prior models tested, establishing new leads in AUC performance across all context lengths and difficulties.
Enjoy.
free gemini 3 is crazy 💀
why is my plican so fat
@echo aurora we need that beast🙏
Is there any issue with thr bot..I am not able to generate videos
ate too many berries
how much you made? share the joy with us! (atleast joy of numbers 🙂 )
theres a deep think model?
only 73 lmao
its old
i am seeing it for a while.. never used it
a robotics model its old
idk i just saw it
learnlm still exists and is used for notebooklm i think
yeah but who has access to deep-think
APi key owners?
will be a while till it ships then
Can someone make a comparison video between Bard and Gemini 3
can we use gemini 3 in gemini cli?
where is grok?
Told u lol
Quite sure you will be able to .. if not now then in few hours
google was so based and knew 4.1 sucks so they didnt even put it there
gemin3 omg
thank god for gemini restoring my faith in humanity
Gemini 3 is FIRST
Now I can say
Google Deepmind does it again
Is it Gemini 3 high or low
low
Models like "gemini deepthink" and "gpt 5 pro" dont matter to most people.
They are too expensive to use daily
even though deepthink 42% on HLE is impressive, I think 37% of normal pro version is more impressive
use it on AI Studio
it has better thinking
and longer tokens
where 36?
i mean 37.5%
fax
the jumpp
Gpt 5.1 is best
Where the heck is gem 3 search @tiny palm?
great start ...
ITS OUT?????
lol.. ragebait?
..bru
LLMs mostly level up with hardware scaling. That's why Google always wins because of TPUS
pingp ineapple..
lmao
hell yeah
its on gemini app now
yes
I'll let the team know
so can we confirm that rift runner was gemini 3
Gemini 3 THE CHAMP
Not anymore. It's not even close.
Gemini 3 pro is a freak
hehe
where 3flash
tbh i prefer Gemini canvas for apps
I think the model on LM is without search
Hasn't roll out for me yet
We all knew Google was freaky since Veo 3 and gd is 3
Sam Altman is having a deep think right now...
gpt-5.1 is pretty ass from my experience
yep , think so too
Gemini 3 pro da boss
its better than riftrunner in AI studio
google is tryhard
