general

[proud bobcat]

Minimax 2.5
GLM 5
DeepSeek V4

[honest verge]

Like glm did the same

[proud bobcat]

Testing first

Probably all the Chinese ai companies agreed

[golden ocean]

claude sonnet 5

[honest verge]

Anyways it comes in February I'm pretty sure

It's almost ready

WAIT MINIMAX 2.5 TOO?

Lol

I thought it won't come out until summer

[violet thunder]

i hate this day

i can't create the game -_-

guys

how to delete cookies i forgot

[proud bobcat]

Browser settings

[bleak lake]

Glm usually launches around the releases of claude

Pretty good ngl

[proud bobcat]

Claude Sonnet 5 is supposed to drop maybe tomorrow

I’m so excited for this week

[bleak lake]

It was predicted on polymarket that claude will have the best model by the end of feb

[violet thunder]

bro how i can upload history to the arena.ai?

[bleak lake]

and prediction was spot on it seems like

[proud bobcat]

Qwen 3.5 just comes out and achieve agi

[proud bobcat]

Can’t

[honest verge]

GLM 5 IS OUT LOL

[proud bobcat]

Recover your account

[honest verge]

I didn't know that

[proud bobcat]

No tech blog post yet

GRAAHHHH

GIVE ME MY BENCHMARKS

[proud bobcat]

Give me some prompts to use with it

[honest verge]

I haven't used it yet

I only discovered it now

[low quiver]

Why isn't the timeout being fixed at all?

When will they fix it?

[surreal zephyr]

whatafu claude/anthropic*

[surreal zephyr]

because its model issue brotato chip

[proud bobcat]

Yeah it’s defo a Claude issue

[low quiver]

No, this is a problem with ArenaAi

100%

[proud bobcat]

It’s not

Trust me

Claude will perish randomly

[violet thunder]

where my history bro nah -_-

[low quiver]

Hardcoded timeout

[proud bobcat]

Well

Here’s to your

New history

[honest verge]

Wait glm 5 agent is also available?

[surreal zephyr]

arenai just shuts down models that get stuck thinking for 10 mins

and opus does that every time

[low quiver]

Claude Opus 4.6 was actively writing the text, but then everything stopped due to a timeout

[spare rune]

Let me try it’s roleplaying stuff

[proud bobcat]

Yes

[left lodge]

What? Reasoning is always native? It's that they have it or not there's nothing in between. There are model in which can disable reasoning but that's just limiting their reasoning to 0 so technically now it's a non-reasoning model

[proud bobcat]

You don’t understand what I mean

GPT is tailored towards specific prompt training

Why do you think codex always has a few prompt examples that do great

[honest verge]

Crazy how fast we got glm 5

It wasn't even 3 months

[proud bobcat]

However when you ask gpt to do anything else it has a seizure and dies

Claude doesn’t

Because it can gauge a prompt by itself without needing that reinforcement learning for it

[left lodge]

Yeah that's not reasoning or whatever but it's just repeating its own training data. It will always be good at it.

[terse shuttle]

what happened with claude?

[proud bobcat]

Exactly

That’s what I mean

Claude has the ability to tackle any project because it doesn’t only rely on reinforcement learning

GPT does

[left lodge]

Somethings wrong with the site cause model is working but then this is happening

They did some updates to website I am experiencing new issues now

[manic lagoon]

Is it me or anyone else facing issue after entering the prompt, getting the same error every time of "Somthing went wrong..",

[left lodge]

Opus-4.6 thinking somehow automatically works after some time check on it after some time lmao

I got results from 4.6 thinking this way

Non thinking is working some time , and time is random too it's sometimes some second and sometimes it goes a minute and then throws off

[manic lagoon]

time to tag admin

[left lodge]

Minimax 2.5
Glm 5
New deepseek version
All very soon 👀

[spare rune]

It usually takes around 1 hour to a day

On all models

I think

[spare rune]

Or probably even less

[honest verge]

From quick testing I can say glm 5 is significantly better than 4.7

I didn't test it fully yet but it's impressive

[left lodge]

For me the site isn't working its full of issues just like arena.ai

[violet thunder]

guys

i can bypass limit

just open in guest window browser service

[left lodge]

That's a new session, you can't have old chats, like opening new account

And if you login you get your limits tied with the account

[honest verge]

Where's minimax 2.5?

I can't see it

Or it's stealth yet

[hollow ivy]

Has anyone ever tried vibe-coding in https://nim-lang.org/ ?

[surreal zephyr]

bad model

[honest verge]

That's why mistral is better

mistral 6.7 large extra ultra max codex cursor high

[surreal zephyr]

https://cdn.discordapp.com/attachments/1018470269326729286/1470906684347253038/togif.gif

[honest verge]

Actually it's mistral not opus

[left lodge]

👀

[west geyser]

will arena.ai have its own video section soon?

[left lodge]

What do you mean by video section?

[radiant swift]

it has

[west geyser]

like arena has text coed and image

[west geyser]

i cant find it :(

[radiant swift]

choose battle

[west geyser]

ok

oh wait i found it

tysm

apprecaite it

[left lodge]

Its only available in battle mode and only 3 generations every 24 hour

[proud bobcat]

No no it’s gonna be addressed in Mistral 6.7.6.9 Extra Large Ultra Max Codex Cursor High Speed

[brittle hollow]

Did the models for image disappear??

[left lodge]

They all are there , where are you?

[brittle hollow]

I can see the others but not for image

[left lodge]

Reload the website

And use white-mode

Hehe

[brittle hollow]

Oh ok it was a bug got worried 💀

[brittle hollow]

https://tenor.com/view/my-eyes-burning-eyes-light-ouch-hurt-gif-16411003

[toxic verge]

Browser reset data

[honest verge]

Big day for Chinese ai today

We got minimax 2.5

Glm 5

Deepseek update or new model

Crazy

[polar brook]

Anybody got a hold of Seedance 2.0

I tried it and its crazy

[left lodge]

So what the heck is this? I am disappointed in GLM5
https://x1k9h162sv41-d.space.z.ai

I have tried 10 new sessions and only 2 of them worked and this is exactly the output they give, I think we have to wait for this stable experience but this will not change the output I think

[hardy charm]

anybody here experienced the same message?

[polar brook]

Yep

Trying Nano Banana Pro again and same message

[left lodge]

yeah the whole community
#1417174113092374689

[polar brook]

Ever since lm arena changed thier website design it got bad

[hardy charm]

lmao.

[toxic verge]

Evolution

There’s a lot of cool features and a bunch of add-ons and stuff they added, which are really convenient and really nice

[left lodge]

Evolution with degradation

[polar brook]

yep

[toxic verge]

Name of the game

No different than any other AI service

[proud bobcat]

New DeepSeek model is fast

Holy hell

[toxic verge]

How fast

[left lodge]

They used to be very very slow

[proud bobcat]

DUDE THIS MODEL IS GOATED?

WHAT

[proud bobcat]

Actually lightning fast

I’m using it for math right now and holy moly

[proud bobcat]

Never was for me BUT

This is significantly faster

I’m getting full answers in less than 3 seconds

[left lodge]

I will see

[toxic verge]

Wow gal don’t got seed dream

[left lodge]

Unofficial**

[toxic verge]

Damn

That is fast

[surreal swallow]

I noticed that 'gemini-3-pro-image-preview' (Rank 3) has disappeared from the Image Arena leaderboard. Is it temporarily down for maintenance or has it been removed?

[proud bobcat]

Accurate too

[proud bobcat]

Probably down for maintenance or upgrade

I heard Gemini is doing a refresh this month

Possibly further fine tune

[left lodge]

Obviously, it will be right. Look at the question you are asking.

[proud bobcat]

Yeah I know that

But it’s FAST

Faster than it was before

[left lodge]

yeah it looks faster

[proud bobcat]

I’ll have to test it with some coding and such but for now it’s defo an improvement

And even if not very noticeable, context handling is a lot smoother

It doesn’t get choked up as much

[left lodge]

Minimax UI is better than ZAI

[north obsidian]

It has one use only 🙁

[surreal swallow]

That’s good to hear! I actually preferred that model over the 2k version, so I was a bit surprised to see it gone. Is there a rough ETA for when it will be back on the leaderboard?

[proud bobcat]

I don’t really know?

[left lodge]

https://u03oj3fyxceg.space.minimax.io/

[proud bobcat]

Perhaps maybe later today or tomorrow

Dude I’m so excited for GLM 5 benchmarks

[surreal swallow]

I see. Thanks for letting me know! I'll just keep an eye on the leaderboard then.

[proud bobcat]

Nodnod

[peak sapphire]

@echo aurora Why are not all models available in the manual image generation mode? For example, in the photo generation battle mode, I came across an image first from the Citrus model and then from the Unami model, and I really liked them. But I can't test them manually — I have to use the Battle mode over and over again, sometimes ten times or more, just to get them to come up. Please fix this shortcoming, and not only in the photo generation mode.

[proud bobcat]

Those are cloaked models

[left lodge]

Instead of sun minimax created an entire whitehole

[proud bobcat]

Models in testing by other companies

You can only access them in battle mode

Ai labs like to test their models on arena to see how well they do and what to refine

[left lodge]

Those are anonymous models.

[surreal swallow]

I keep getting this error message. Is it because I've been rate-limited for generating too many images, or is it a system issue?

[polar brook]

anybody knows a way I can use Nano Banana Pro for free and unlimited?

[left lodge]

There is nothing free and unlimited in this world

[toxic verge]

Amen

[left lodge]

Namam

[polar brook]

Yep

[left lodge]

#1417174113092374689

[toxic verge]

Gem flash is trash

Has the same exact problems the original one had

[surreal zephyr]

Here http://careers.google.com/

[proud bobcat]

It’s always been solid

I wouldn’t say it’s bad

[surreal zephyr]

bro ofc its not as good as gpt, but its still decent

[proud bobcat]

Bro cannot stop meat riding gpt

😭

[toxic verge]

Yeah, I guess you’re right is decent

[polar brook]

Im in the website now. Whats the next step?

[proud bobcat]

GPT is turbo ass

[toxic verge]

I just get pissed off at it

[proud bobcat]

I personally use Kimi instead

It just has better thinking

[surreal zephyr]

u dont have it so shush

[left lodge]

yeah kimi improved by a lot it is just amazing right now

[toxic verge]

Honestly, I can’t lie no more censorship is really making ai very unpleasant to use

And I don’t even mean NSFW

[proud bobcat]

I swear when 5.3 codex comes out on arena and it’s ass I’m frying you

[surreal zephyr]

https://cdn.discordapp.com/attachments/1446560503668146178/1462543800097570836/togif.gif

[proud bobcat]

Bingo!

[left lodge]

Fr 😭

[proud bobcat]

The safety tax

[surreal zephyr]

kimi would NEVER

[toxic verge]

No, I mean it’s gone way too ridiculous

It really turns me off from ai

[placid verge]

bro is the freakin nano banana fixed yet

[proud bobcat]

Did you know Claude apparently processes 65K tokens of safety prompts before responding

65 thousand tokens

What a waste

[proud bobcat]

Use DeepSeek in that case

DeepSeek is just goated

[toxic verge]

That the thing not tryna run from model to model

And it’s not like I’m doing anything bad or even harmful so it doesn’t matter and

[left lodge]

people don't realize how big is the 200k window

[toxic verge]

This is a disease that plagues all of AI, not just a single model

To heavily controlled too expensive for regular people to run high-end high-quality models

No control

[placid verge]

like yesterday it kept saying everyday terms of service violation even though of used the same prompt hundreds of times before

[proud bobcat]

Safety is defo good but my god does it interfere with a model

GPT is lobotomized because of it

[toxic verge]

Yeah, but we went beyond safety here, dude

And I don’t in the rain I’m talking about all models in general and all of the AI universe

And what’s messed up it’s gonna drive a lot of people to use underground LLMs

[shrewd citrus]

freaks just ruin everything don’t they

[proud bobcat]

No it won’t 😭

Bro has never heard of openrouter

[polar brook]

Gooners and freaks be ruining it for us

I accessed Seedance 2.0 on byteplus but now im not seeing it

not even on Dreamina

[toxic verge]

For seed dream

Just like Sora

https://youtu.be/JS1pLZiOC6o?si=IOyiJSvk1cNUEpeV

[north obsidian]

It's there but only in the Chinese model

The jimeng but it's almost impossible u access it

[remote vapor]

soooo umm-

• glm 5
• minimax 2.5
  just came out, hm?

[proud bobcat]

Yes

[remote vapor]

excuting, ey?

I just hope ms dear dear qwen also comes out if hiding.....

[left lodge]

Its not actually fast or maybe I didn't got the test but I am sure that is just animations you are experiencing in the app , try on web you will see the reality.

[shrewd citrus]

glm 5 is kinda ass

[left lodge]

Glm looks like a very minor upgrade, minimax is a better upgrade

[shrewd citrus]

tried it out for coding and it didn’t do well

[left lodge]

Yep same here

[left lodge]

Maybe they improved in some other aspects like agentic behaviour cause now there's agentic mode available for glm 5 not 4.7

Minimax M2.5 ocean 🌊

Glm 5 is so bad here 😭

I hope they done something in other domains

[proud bobcat]

It still fast on web for me too

Perhaps I got the special DeepSeek privileges

[proud bobcat]

Perhaps result of heavy load due to launch?

[left lodge]

It gives me the new 1 million context window and too but it's slow compared to others still

[proud bobcat]

Im just better than you

Get rekt

[left lodge]

But that should not affect the actual output, it can cause that it will just not work or work 50:50

[proud bobcat]

We don’t know if they quantized it yet for the launch

So

[left lodge]

What are you saying 😭

I don't get a word

[proud bobcat]

The Chinese ai gods have FAVORED me

[left lodge]

:v

Bro wrote a article here 💀

[proud bobcat]

…

Thanks bro?

Lmao

[left lodge]

He deleted xd

[proud bobcat]

Me when I don’t go to the channel specifically labeled as video arena

[surreal zephyr]

lol they removed nsfw, the model is going to fell off now that 99% userbase is gone

video generators have no other usage besides self made nsfw lol

grok controversy proved it

[hushed gyro]

nano banana users would increase by 1000% if they allowed nsfw

wait did the model train with nsfw content?

[sweet cove]

🦥

[toxic verge]

Real talk

[surreal zephyr]

FansBanana - trained on onlyfans

[hushed gyro]

no like fr did google include nsfw in the training data?

[surreal zephyr]

ofc bruh

they dont exclude that

but its reinf trained to not make nsfw

[mortal spear]

Followers badha do

[left lodge]

If all of that is true and you are actually capable you should not be asking for job here. This is not a place for it. We are consumers. We you don't hire, we just consume things.

[pulsar crystal]

when is deepseek coming out?

[golden ocean]

it's too shy to come out so early

[quartz light]

same result as 4.6 opus, actually

opus 4.6 is the only model that often does this for the prompt

[shrewd citrus]

Lunar new year

[pulsar crystal]

that is in 6 days right?

[shrewd citrus]

yep

[cedar tide]

☠️

[hushed gyro]

how to unreinforce it

[surreal zephyr]

💔

[hushed gyro]

bro is it even possible

[surreal zephyr]

No

Lmao

[proud bobcat]

Stares

[surreal zephyr]

Its like they run random prompts, if it generates nsfw it gets killed and replaced

[hushed gyro]

so then why do corporations censor everything in fear of government regulations and lawsuits from karens

[proud bobcat]

“How to unreinforce it” we wish too man 💔

[proud bobcat]

Well you don’t really want a model making problematic content

It’s quite simple really

[surreal zephyr]

Because CSAM

[hushed gyro]

then we need to change the law to blame the user not the company

[surreal zephyr]

https://cdn.discordapp.com/attachments/1412020704924401724/1471182491384676647/watermark.gif

[surreal zephyr]

Thats already the law

But company gets hated by dumbasses

So its PR issue

[hushed gyro]

then change the society's perceptions on this issue

[hushed gyro]

cuz i thought reinf = reinforce

[proud bobcat]

That’s already the law

But it’s also up to the company to have proper guardrails

[hushed gyro]

then why can some models run locally without any guardrails?

why cant google do that? open source their model and the user can do whatever they want

[proud bobcat]

Because they are local open source models

The company cannot moderate that

And also Google won’t open source their Gemini architecture

[pulsar crystal]

you can get uncensored local models
if they have open weigths, someone will take the safety stuff away
but you have to have a good pc

[proud bobcat]

That’s literally market death

[surreal zephyr]

Yes reinf = reinforce

But you cant unreinforce stuff bruh

[left lodge]

No?

I have provided results from opus 4.6 before in this chat you can check those

[hushed gyro]

guys why arent the devs fixing the errors?

[proud bobcat]

Backend issue

They need to diagnose it

[surreal zephyr]

Because they using opus

[hushed gyro]

waves? really?

[surreal zephyr]

Ya

[left lodge]

They were trying to cook but they burned the whole because they forget they are devs not chefs

[proud bobcat]

You can tell they stuffed a ton of stuff into the training data

I had gpt 5.2 codex try to make a Minecraft clone and it didn’t even run

Needed Kimi to fix it

[hushed gyro]

guys NB Pro is literally the best image model since its release and will be for the next few months, and this is the only platform where they offer it completely free of charge. This is extremely disappointing, given that the company literally had a rebrand (which was totally pointless) and is now worth $250M.

also after the issues occured the quality of the generated images have gone down significantly? they really need to make us customise our own system prompts!

it would really help!

[shrewd citrus]

it’s not worth 250m lol

[proud bobcat]

GLM 5 BENCHMARKS ARE HERE

[hushed gyro]

glm 5? new model?

[left lodge]

Kimi k2.5 instant
https://019c4be9-1e0d-71f3-aabd-d2f24690fb30.arena.site/

Claude opus 4.6 thinking
https://019c4bdd-ca2f-74ce-a4dd-a2e6fadc9d68.arena.site/

Reality ↑ {results from one prompt}

[proud bobcat]

Yes

[hushed gyro]

dont you realise that glm models have on average lower quality than other open source models?

[left lodge]

Where are the model names 😭

[hushed gyro]

so im not really sure if glm 5 is a groundbreaking example

[proud bobcat]

https://z.ai/blog/glm-5

Wait let me

Shorten the link

[proud bobcat]

What.

Does bro know?

[hushed gyro]

im testing the model rn

[left lodge]

Why you cropped it?

[proud bobcat]

Oh my bad

Sorry

Phone shows up differently

So I wanted most quality in a screenshot

[left lodge]

We are launching GLM-5, targeting complex systems engineering and long-horizon agentic tasks. Scaling is still one of the most important ways to improve the intelligence efficiency of Artificial General Intelligence (AGI). Compared to GLM-4.5, GLM-5 scales from 355B parameters (32B active) to 744B parameters (40B active), and increases pre-training data from 23T to 28.5T tokens. GLM-5 also integrates DeepSeek Sparse Attention (DSA), significantly reducing deployment cost while preserving long-context capacity.
Glm 5 is double the size of 4.7 to gain this increase 💀 bruh

[proud bobcat]

To be fair they squeezed the most out of their old dataset

Now they’re gonna squeeze the most out of this one

[left lodge]

Yeah glm models are always on agentic behaviour

They are below opus 4.5 in coding but above all others (including gpt-5.2-xhigh) and quite good in agentic stuff

[proud bobcat]

its not that far of a gap anymore though tbf

[obtuse heart]

Why is nb pro always so broken in arena? It's always "something went wrong with this response"

[left lodge]

Yeah but these benchmarks numbers are kind of specific by task so better is better in that task. But general intelligence is different scenario

[echo aurora]

It can be a variety of reasons for why you're getting this error, it's recommened to try the steps in this article: https://help.arena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

[proud bobcat]

nodnod

i think glm 5 is not groundbreaking just yet because effectively

this is just the first iteration

[obtuse heart]

I see thanks

[proud bobcat]

im sure the architecture will have every drop of power squeezed from it

[left lodge]

Yeah they are trying to close the gap as much as possible in any way but that could be not a good way to close the gap , rn there is much greater pressure on china labs than before.

X-AI is not having a good time. I hope they survive

[toxic sphinx]

When's seedance 2 coming to the site

[echo aurora]

No problem. It's also worth noting that it's common for this issue to be caused by rate limit. Unfortunately, the error message isn't descriptive (something we absolutely need to change), but you this article here will explain how to verify if it is rate limit - https://help.arena.ai/articles/8931786544-arena-how-to-rate-limit

[proud bobcat]

lmao you know gpt begged them to fudge this

[left lodge]

5.2 Refine like what is that bruh 😭

[proud bobcat]

if anything the pressure isnt that bad

they dont have to explicitly worry about losing funding and research is shared

[proud bobcat]

exactly 😭

no kimi k2.5 on the leaderboard either

nor glm 4.7

i smell bias

[left lodge]

Yeah

[proud bobcat]

they havent even tested any new deepseek models

come on dude

[left lodge]

There is grok 4 refine too 😭 but it's still so low lmao

Yet so expensive

[proud bobcat]

they probably had to explain every problem to gpt in detail and then giveit the answer

[left lodge]

They are just giving answers in prompt at that point 💀

Do you have any likeness for ChatGPT ui/ux?

[proud bobcat]

its

fine

[left lodge]

I just hate how the way it looks , just so messy and unnecessary complexity

[proud bobcat]

i prefer claude and deepseek's ui the most

[left lodge]

Yeah Claude ui is calming

Deepseek ui is clean

We're bringing some of Claude’s most-used features to the free plan.

File creation, connectors, and skills are all now available without a subscription.

Compaction is also now on the free plan.

Claude summarizes earlier context automatically, so long conversations can continue without starting over.

Claude is being generous somewhat 😭

[proud bobcat]

claude really needs to fix their pricing because good lord

[left lodge]

The limits are still low and kind of wierd like I only can send limited messages It doesn't if I used haiku or sonnet or even if I enabled thinking or not.

If it says limit reached you can't use anything

Not even haiku non-thinking

[quartz light]

yes*

from my experience

although i like opus

it produced that same issue

a lot

[proud bobcat]

i mean claude is more used for api and enterprise

its not like gpt

[quartz light]

until i changed the prompt to mention creation of custom shaders n stuff

[left lodge]

That's why see this I came with examples

Models can output dull results sometimes but I mostly got good results

[left lodge]

My prompt is very basic and simple

[civic spindle]

remove the rate limits this ai slop fell off ever since

[left lodge]

That will require a subscription of 2 million dollars can you pay?

[quartz light]

kimi used three/addons/objects/Water.js

this test is so dumb

[left lodge]

Yeah both used that

[civic spindle]

maybe they should get more money, not my problem

[left lodge]

Then don't except to get the best at free , nothing is free

[quartz light]

no

wdym?

[civic spindle]

not the point
they should get richer

worst platform in existence
rate limits 🤡

[quartz light]

??? @left lodge

i checked the code.

[left lodge]

All models I tested in here had the same available tools and stuff

[quartz light]

dude.

this test is dumb.

as ####.

[left lodge]

When did I say this is the best benchmark?

[quartz light]

this is a waste

no point

misleading

[left lodge]

Not a waste

It shows its capabilities at this specific domain and case

[quartz light]

it does not

[left lodge]

Some models are great some are really bad at this

[quartz light]

again, this test is really bad

it does not tell the model not to use water.js (which is incredibly basic)

[civic spindle]

no one gives a crap about your test and this slop of a platform stuck on rate limits

[left lodge]

I showed this test and you replied that it was same as opus 4.6 thinking and I showed it is not? So why are you blaming the test now 😭

[quartz light]

because the test doesnt really show much

[desert abyss]

Please let's be respecful toward the other users.

[quartz light]

i just mentioned ive tried this too with opus and it produced the same error which is interesting

[left lodge]

Stop changing the topic lmao

[quartz light]

???????

[civic spindle]

remove the rate limits

[left lodge]

Try again then?

[quartz light]

hes not a developer

[deft nova]

Arena is soo peak

[civic spindle]

i wanna use this platform for free and make unlimited images

[left lodge]

Ahm

[quartz light]

yes but i said it is misleading

[left lodge]

I didn't said anything?

[quartz light]

because you never mentioned why kimi may be dull, people would just think kimi did a better job

[left lodge]

I am not leading anything

[left lodge]

I gave same prompt to every model and tried multiple times

This is not a one good chance

[quartz light]

https://cdn.discordapp.com/attachments/1205990758348230666/1295568791706669066/attachment.gif

[civic spindle]

tell this pineapple kid to remove the rate limits off of this platform its so bad atp
most unbearable and most insufferable platforms that turned greedy every since

[quartz light]

"this pineapple kid"

gtfo

[left lodge]

I literal said I tried the test with this model and it didn't perform well lmao and showed proof what's misleading I don't get it 😭

[quartz light]

i hate the rate limits too but why say this

[civic spindle]

trust must hurt

[quartz light]

they cant entirely remove the ratelimits that would be abused

[left lodge]

I suggest you to use the holy block tool!

[dreamy ivy]

anyone tried GLM 5?

[quartz light]

do they always talk like this or what

[proud bobcat]

https://tenor.com/view/ahem-shit-loser-seriously-gif-5731140

[proud bobcat]

its very solid

not great great

but solid

[left lodge]

Yeah some people do and we can't do anything lmao
Its just their thinking

[dreamy ivy]

aight

[left lodge]

Ahm

And ahm https://z.ai/blog/glm-5

[quartz light]

• pineapple is a mod
• pineapple is not a kid, pineapple is botanically classified as a fruit
• how can they be greedy when they doesn't charge you anything
• removing ratelimits entirely is a bad idea, but we definitely need an increase for when using arena while logged in

[woven harness]

stop complaining you’re being really annoying and you should be at least a little grateful that people are spending time and money so others can try these models if you don’t like it you can always find another site, also, shut up and calm down and be respectful to other user cuz nobody treated you bad 😭

[left lodge]

Midnight is silent now and gave me a permanent warning cause of a similar fellow

[dreamy ivy]

I agree partially

[desert abyss]

Thank you so much for your support guys! That user was hit with the ban hammer, we don't need that kind of negativity on this server.

[dreamy ivy]

yo he got banned

damn

[left lodge]

Yay

[woven harness]

yayyy

[quartz light]

YES

YES

[woven peak]

is it just me or gemini models in grounding (haven't tried others) not display sources of websites they use?

[quartz light]

my bot detected

[woven peak]

i dont know if they even search in the first place

this only happens in arena

[left lodge]

Yeah they listed that in their blog
https://z.ai/blog/glm-5

[woven peak]

google ai studio is fine

[sleek phoenix]

@echo aurora the padding between the logo and the title is different depending if it's an actual logo or the generating animation

[woven peak]

example

[left lodge]

We don't see all websites they got back from search I think idk how they select which one to show.
But like there is similar case that when grok access twitter it doesn't show that in sources and instead say no search used

[left lodge]

Yeah same as I described for grok

[woven peak]

so it searchs just decides whether to show it in the box or no?

[sleek phoenix]

i think the sources dropdown just shows which websites the AI put a citation in its response

[left lodge]

Not sure but there has to be something

In official platforms they list all the sources that the model got back from search tool but here we don't see all

[woven peak]

i think it just decides whether to show it or not

[woven peak]

plus its only 2 websites so its definetly that it might be searching for more websites just doesnt display

[quartz light]

THIS BENCHMARK IS MUCH BETTER

@left lodge @proud bobcat

[proud bobcat]

huh

[left lodge]

@echo aurora What is the criteria for showing sources? See older messages for context.

[proud bobcat]

its the same benchmark

[left lodge]

I think that's one is from their blog I saw it

[woven peak]

in the message where it showed 2 sources i asked it to cite sources and put it in the text box and not sources box it told me where it searched for (i did include it in the prompt where to search for so it might just be looking at the prompt but it also included 2 random sources in the sources box so im unsure) and in the one it didnt show sources (the first prompt) i asked it what i wanted it to search for

[left lodge]

I have a better view one.

See

[quartz light]

ok noob

smh

😡

[left lodge]

Wut lmao I provided you with a better view for the table 😭

[quartz light]

im jus kiddin

[bright shard]

What is P-image?

[left lodge]

People really are wired

[left lodge]

Image generation AI model

[quartz light]

no

[bright shard]

Which company?

[quartz light]

im kiddin

[left lodge]

Idk do a search

I am not a search engine

I don't understand why people don't use search engines and just ask here and wait for hours

[woven peak]

i also gave it the output in another chat and asked it to verify if all it said is true and it gave sources

[left lodge]

Models usually get 5 or atleast 10 sources for search queries. From I have seen on official platforms and api documentations

And in here source drop down shows one or two and sometimes more

No model searches for a specific webpage

[quartz light]

yall

they released many glm 5 versions now

[left lodge]

Many?

Like flash , vision supported?

[quartz light]

weeelll

no

like not even all official

[left lodge]

Uh

[quartz light]

GLM-5-FP8 (zai-org/GLM-5-FP8) (details above)
GLM-5-GGUF (unsloth/GLM-5-GGUF) (details above)
GLM-5 (unsloth/GLM-5) (details above)

just quants

but still useful info

[left lodge]

Bro that's just Quantization
Yeah

[quartz light]

1. GLM-5
╰ repo: zai-org/GLM-5
╰ link: HuggingFace ↗
╰ task: text-generation
╰ params: 753.9B
╰ downloads: 0
╰ likes: 41
╰ updated: 2026-02-11
╰ inference: novita · zai-org
2. GLM-5-FP8
╰ repo: zai-org/GLM-5-FP8
╰ link: HuggingFace ↗
╰ task: text-generation
╰ params: 753.9B
╰ downloads: 0
╰ likes: 3
╰ updated: 2026-02-11
3. GLM-5-GGUF
╰ repo: unsloth/GLM-5-GGUF
╰ link: HuggingFace ↗
╰ params: 753.1B
╰ downloads: 0
╰ likes: 2
╰ updated: 2026-02-11
4. GLM-5
╰ repo: unsloth/GLM-5
╰ link: HuggingFace ↗
╰ params: 753.9B
╰ downloads: 0
╰ likes: 1
╰ updated: 2026-02-10

[woven peak]

i gave same prompt in google ai studio and it only gave one source but clearly in thinking you can see it is looking at alot of sources unless its hallucinating

if its not hallucinating i think whats happening is that it decides whether to display a source or not

maybe only decides whether to show a source or not if it can directly quote it

[left lodge]

AI studio got another ui revamp
Its like 4th or 5th revamp they are refining and testing new ui

[left lodge]

If I ever had to run a model offline I would use kimi k2.5 as my main go to

All these other models are only good at specific domains

[left lodge]

Hm

Speed and quality matters too, many times these providers use smaller models

Aye

Glm 5 in arena

[obsidian shell]

people cant just stop releasing models

[quartz light]

novita is usually one of the fastest

[left lodge]

They are just accelerating so fast

[left lodge]

Nice than its better for end users

[glacial mulch]

why not

[proud bobcat]

glm 5

https://tenor.com/view/homelander-homelander-the-boys-homelander-sad-homelander-its-peak-homelander-peak-gif-14542009839452529163

[proud bobcat]

WHAT.

[left lodge]

Hm

[echo aurora]

I'm going to send this to #ask-here and will followup when I get to your question.

[proud bobcat]

its so expensive

https://tenor.com/view/black-man-crying-cry-crying-crying-emoji-crying-meme-gif-3223392509479903783

[quartz light]

LOOOL NO WAY THEY JUST BANNED ME FROM THEIR DISCORD FOR POSTING THIS

AND ITS NOT EVEN CORRECT NOW

[distant spoke]

Deepseek New model

[left lodge]

Okey

[quartz light]

Z.AI JUST BANNED ME

[proud bobcat]

testing yes

[proud bobcat]

wait what

[left lodge]

This is something to see 👀

[left lodge]

Lmao

[quartz light]

CRAZY

I hope it was an accident????

but since its a permaban i cant even appeal

lmfao

[left lodge]

😭 sometimes they crashout so easily

[quartz light]

there was a guy who pinged an entire role right before my message

or after idk

but they probably tried to ban that guy not me

cz i didnt do anything

[left lodge]

They might be lazy and might never unban you xd

[quartz light]

they wont ever care to check

they havent realised

im sure

[quartz light]

discord's ban thing should be clearer on who you're banning

it should show some of the user's recent messages

╰ repo: zai-org/GLM-5
╰ link: HuggingFace ↗
╰ task: text-generation
╰ params: 753.9B
╰ downloads: 0
╰ likes: 72
╰ updated: 2026-02-11
╰ inference: zai-org · novita

0 downloads

i wonder how long itll take

[left lodge]

It is clear but people ban quickly

I think the time has come to make a new discord account for a single freaking server lmao

[fickle venture]

Fym GLM 5

We just got GLM 7

Which one is better

[burnt sinew]

What

[fickle venture]

I mean why they dropped GLM 5 they should've named it GLM 8

[left lodge]

No cause it's not that better

They are not apple

Which switch to 26 directly

[fickle venture]

So GLM 5 sucks?

[swift ermine]

What is multi file app?

Can some1 explain

[left lodge]

No

Its better than 4.5 its no so much good that you would skip 2 generations for it

[echo aurora]

This has a really good explainer: https://www.youtube.com/watch?v=lAFsaT5oi8g I'll be posting this tomorrow to the server, but it's live now so enjoy!

[left lodge]

The video on Discord announcement is 160MB 💀

Better watch on yt

[swift ermine]

Ty pineapple

[fickle venture]

Great update

[left lodge]

That is the thing in code arena which had multiple files right? With listing tool and more?, self answer yeah I think

[open wind]

Please add the ability to delete/edit change's files in Code Arena. @echo aurora

[echo aurora]

Glad to hear you like it!

[echo aurora]

Could happen! Best place to tell us about feedback is in #1372230675914031105

[fickle venture]

Tell ai to do it

[left lodge]

Yeah I haven't noticed that if it have delete file tool, it can only edit and create I think

[fickle venture]

At this point people will have their own website on arena 😭✌️

[open wind]

Unless I tell him to, he creates and edits different/inappropriate files, and I have to tell him to delete them.

[open wind]

sure ı'm writing now

[fickle venture]

When sharing the code you made like as link does it expire or stay online forever

[left lodge]

But delete tool can cause chaos because models like gemini delete anything and then say sorry 😭 lmao, Sorry isn't bringing back those files

[echo aurora]

The intent is for it to remain online and available

[left lodge]

I want to suggest a lot here but I will later in #1372230675914031105

[stray aspen]

is glm good

[fickle venture]

Here is a feedback, we could use Google drive so anything we make goes to Google drive

[fickle venture]

Idk tbh wait for it to be on leaderboard

[left lodge]

No don't

Everyone don't use google, like me I don't use google

[fickle venture]

You only login using google

[left lodge]

No we can login using email too

[fickle venture]

Why you don't use Google?

Cuz of phone number?

[left lodge]

Privacy

[tender patio]

@echo aurora are there any plans for a vs code executor thingy?

[left lodge]

Google tracks every single click

[tender patio]

Similar to cursor

[tender patio]

what

[left lodge]

#ask-here

[fickle venture]

That won't happen

[tender patio]

Never know man

would be cool

[fickle venture]

But just put it in #1372230675914031105

[tender patio]

https://cdn.discordapp.com/attachments/1406520995254239295/1460236287322624100/togif.gif

[fickle venture]

Free models, free website, free hosting dude that's too much ✌️

[proud bobcat]

another W from arena

glory to pineapple

[little ginkgo]

Meh everything is so fire now

Just missing a txt file upload

W arena tho

[fickle venture]

Pineapple And arena team

[tender patio]

Im not saying what we have now is bad

im just saying it could be cool but not needed

it would make more people use it also

[fickle venture]

#1372230675914031105 hopefully the developers accept it

[proud bobcat]

arena users after locking in and making entire minecraft clones

[echo aurora]

This, but mostly the team, Aryan is an absolute rockstar

[proud bobcat]

https://tenor.com/view/soviet-union-propaganda-stalinium-gif-16289670004460099406

bing chilling

[tender patio]

#1471211109452349472

[proud bobcat]

the machine must keep moving

[left lodge]

Don't know which one lmao there are so many aryan in here xd

[left lodge]

Official people in here should have tags like moderator and should be listed separated in the member list

[fickle venture]

I gave a star hopefully

[fickle venture]

It's better they be hidden cuz people will just target them

[north obsidian]

The video on announcement is not loading for me

[left lodge]

That's just theoretical

[fickle venture]

#general message

[hushed gyro]

Oof GLM 5 was received negatively

See this is why since 4.5 I don't trust Z. AI releasing a decent model anymore

They're all dogwater

[left lodge]

One bad thing shouldn't stop the benefits

[north obsidian]

Ow thanks lol

[fickle venture]

67

[left lodge]

Glm are not general models they are made for agentic tasks and high tool calls reliability

Lol

[hushed gyro]

89 🤭

[fickle venture]

Vro

Really bro?

[stray aspen]

lol

[hushed gyro]

OK fine 😔

[left lodge]

Nah no need to delete that

Its what it is

[fickle venture]

Hopefully no one saw that

[stray aspen]

is glm 5 good

[hushed gyro]

Cuz like other servers get triggered whenever I make a slightly "offensive" joke (eg related to someone's name or ethnicity)

Like they need to chill out

[left lodge]

We had a laugh and I think it's good

[hushed gyro]

It's not even that bad imo

[fickle venture]

Aryan is a developer on arena bro

We should respect him

[left lodge]

Yeah people sometimes just crashout so easily

[left lodge]

Bro he is not having any motive to harm ?? He is just sharing what he thought

[hushed gyro]

Well no offense to him it was literally on his name which is shared by millions of people

And it's literally just a swap of letters

Anyways...

[fickle venture]

I didn't crashout it's just I feel bad

[hushed gyro]

Well... Everyone is different at reacting to jokes

Cuz I do a lot of dark humour

This is light work for me 😂

[left lodge]

You need to feel less bad idk why you feel bad

I suggest train yourself so you don't feel bad lmao

[fickle venture]

https://tenor.com/view/speed-speed-trying-not-to-laugh-speed-rage-speed-angry-ishowspeed-gif-5724215244380832722

[left lodge]

🤣

[hushed gyro]

Is speed wearing a hijab?

[left lodge]

Wut

[hushed gyro]

Or just a loose headcover it's not clear

[left lodge]

Lmao yeah I think

[fickle venture]

https://tenor.com/view/speed-speed-trying-not-to-laugh-ishowspeed-ishowspeed-not-laughing-not-laughing-gif-14283606936869876625

[left lodge]

It is going on his nick

[hushed gyro]

Defo a hijab

Anyways...

Your opinions on GLM 5?

[fickle venture]

Boring

Jk I haven't tryed it

[left lodge]

Little upgrade from 4.5 but still focused on agentic task

I mean they are archiving what their motives are

[hushed gyro]

Tbf calling it 4.8 would be more appropriate

Btw look at their DC server

Chats on fire lmao

[left lodge]

Just numbering this and that

I like current scenario of kimi k2.5 and minimax m2.5 lmao

Exactly same naming scheme

[spiral basin]

What is best model for coding

[shrewd citrus]

mistral

[fickle venture]

Claude Opus 4.6 Thinking

[proud bobcat]

id call it a big upgrade tbh

[left lodge]

Its definitely a upgrade I would have called a big upgrade if they could have atleast cooked everything sota model with enough margin in atleast agentic behaviour

[echo aurora]

Can you reload the Discord app?

Are others also having troubles with the video? I've tested on app and desktop and appear to be running just fine for me.

[left lodge]

Its cause the video is too large it will take a lot of time,the size is 160MB

[proud bobcat]

i had no issues

[shrewd citrus]

works fine for me

on mobile at least

[left lodge]

Well I had loading issues :p
Maybe slow speed rn

But yt working fine

[proud bobcat]

wait havent we had multi file apps before

or is this like enhanced

[left lodge]

Its now generally available

[proud bobcat]

ohh

i see

[left lodge]

Earlier it was in testing

[echo aurora]

Yeah fully rolled out officially

[simple moth]

the video worked fine for me like 20 mins ago

[left lodge]

Pineapple I quicker than you :v

[simple moth]

on mobile

[echo aurora]

True true

[left lodge]

:p

[echo aurora]

Making you CM

[left lodge]

ooo

[proud bobcat]

all hail Error

[echo aurora]

https://klipy.com/gifs/hail-the-victorious

[remote vapor]

will.... we get qwen...?..... I want my qwen...

I want my karp, my qwen back 🍰💖

[proud bobcat]

well qwen 3.5 is set to release around this time

i think

chinese new year

year of the horse

[remote vapor]

no one sais it yet, but maybe...

[proud bobcat]

oh no we know

[remote vapor]

where did u hear about the release?

🐴 💖

[proud bobcat]

well the team published documents on huggingface related to qwen 3.5

[stray aspen]

qwen will suck

[proud bobcat]

and in their new release of qwen image 2 they snuck in some references to qwen 3.5

[proud bobcat]

ragebait

[remote vapor]

I kno.... I read them! I saw the 9B and 9B-text!

but..... no release.

[simple moth]

yo pineapple, is the vscode extension still working. like are the endpoints from there still working. Becuase I wanna revive ts and make a full agent including MCP server support.

[proud bobcat]

well since glm 5 and minimax 2.5 released today

id wait to see tomorrow

[remote vapor]

I also saw the reference to 3.5 in the image post-

yassss I want qwen!!!

i am - the number one qwen fan -

[stray aspen]

wdym minimax 2.5 released today

[remote vapor]

(uncontested.... unless someone wants to)

minimax released, yes

[proud bobcat]

bro does NOT know?

[remote vapor]

couldn't be a minimax fan 😭

[proud bobcat]

qwen, generate me a python script to give this guy testicular torsion

[little surge]

Can someone help me? My chat seems to be stuck in a loop and is generating indefinitely.

[proud bobcat]

make a new chat

backend glitch

[stray aspen]

where do i use it

[little surge]

Bruh but i cant download nothing unless its done

[remote vapor]

stop the generation.

[little surge]

You cant stop it

There's no stop button

[proud bobcat]

minimax site

[left lodge]

A win I would say is that glm5 is performing better in our code mode than their native platform @echo aurora
https://019c4e06-95f6-7556-9214-d93636303262.arena.site/

[proud bobcat]

well

just wait a bit

[quartz light]

@left lodge the model is much bigger now

so im not really that impressed

[proud bobcat]

glm is better at full stack

i think if you lock it down to just html it doesnt use its full capacity

[stray aspen]

good

[quartz light]

that looks cool

[proud bobcat]

also btw heres a little suspension test claude did
https://019c4e03-cee7-7ec4-9988-9d4f7c7035a5.arena.site/

[little surge]

Ive been waiting for like 7h now

[proud bobcat]

just start a new chat brochacho

😭

[left lodge]

Result from their own platform using their similar full stack mode

[proud bobcat]

arena tends to have diff prompts and results because

well

[little surge]

Crazy

[proud bobcat]

i think the models can put in more creativity

[left lodge]

The result is using Fullstack and actually it performs better if we just keep it to html in its native platform

[proud bobcat]

well maybe im wrong

i dont know

well regardless

looks like kimi k2.5 thinking is the king of open source coding

[left lodge]

Yeah kind of we have to wait and see how good minimax 2.5 & glm 5 performs in other peoples test

Another generation from glm 5
https://019c4e0b-9172-79cf-b33b-0cf55a257023.arena.site/

[proud bobcat]

wait

wait

[tawdry coyote]

where is the archived chats page !!! 😭

[proud bobcat]

i have the ultimate test

[left lodge]

It is performing much better than their native platform , crazy

[proud bobcat]

i think they have worse system prompts

[left lodge]

You have to scroll the clips/blocks they doesn't look like it but you can

[tawdry coyote]

found it https://arena.ai/search?filter=archived

thanks

[proud bobcat]

oh holy hell

glm 5 is SLOWWW

[left lodge]

No?

[whole swallow]

Maybe cuz of the sys prompt

[proud bobcat]

but it mogs GPT 5.2 high

[left lodge]

Bro stop doing that prompt on thinking model its waste of compute 😭

[tawdry coyote]

exactly my thoughts

[proud bobcat]

i am the one who wastes water

[tawdry coyote]

but i guess its a reasonable first test

[proud bobcat]

jokes aside though gpt 5.2 sucks at questions like these

like really bad

[left lodge]

I think a lot of compute is wasted in how many letters and very similar tests

[proud bobcat]

its still a stress test

[left lodge]

Yeah there is the seahorse emoji one too which breaks gpt models completely

[proud bobcat]

if a model cant tell me how many instances of a specific letter there are in a word do i actually trust it

[left lodge]

It happens differently cause of how their tokenizer works is what I have heard

Because eventually models translates our words into tokens before computing

What is this bro 💀
Go here it seems #share-prompts

[proud bobcat]

https://tenor.com/view/martin-scorsese-absolute-cinema-absolute-cinema-meme-martin-scorsese-meme-martin-scorsese-absolute-cinema-gif-7188464701423971028

[woven harness]

so funny to see how there are 200k users using the video arena and only 5 using the other channels

[pulsar kiln]

brothers i suffering with claude opus 4.6 thinking in lmarena.ai it send me everytime "something went wrong...

[little surge]

lol yeah

[loud verge]

Fr?

[little surge]

yeah look at the arena

#video-arena-1

[loud verge]

People are bored, lol.

[left lodge]

But wait wasn't the discord support for video generation dropped?

How many hours are left?

[loud verge]

I thought the same.

[quartz light]

GLM 5 STILL has 0 downloads

🤗 HuggingFace — 4 hits

1. GLM-5
╰ repo: zai-org/GLM-5
╰ link: HuggingFace ↗
╰ task: text-generation
╰ params: 753.9B
╰ downloads: 0
╰ likes: 142
╰ updated: 2026-02-11
╰ inference: zai-org · novita
2. GLM-5-FP8
╰ repo: zai-org/GLM-5-FP8
╰ link: HuggingFace ↗
╰ task: text-generation
╰ params: 753.9B
╰ downloads: 0
╰ likes: 9
╰ updated: 2026-02-11
3. GLM-5-GGUF
╰ repo: unsloth/GLM-5-GGUF
╰ link: HuggingFace ↗
╰ params: 753.1B
╰ downloads: 0
╰ likes: 7
╰ updated: 2026-02-11
4. GLM-5
╰ repo: unsloth/GLM-5
╰ link: HuggingFace ↗
╰ task: text-generation
╰ params: 753.9B
╰ downloads: 0
╰ likes: 5
╰ updated: 2026-02-11

-# 📊 5920 tracked · 🕐 2026-02-11 19:18Z

[left lodge]

Okay bro stop this

[loud verge]

Bro could have posted a screenshot or two instead.

[left lodge]

This isn't a 3 billion parameter model

Which whoever can download and run

This is a freaking ~800b size model

This needs a 2016 nasa server

To run

[toxic verge]

🤣🤣🤣

[woven harness]

is it possible to run this locally?

[mystic olive]

Yrs

[left lodge]

Yes

[mystic olive]

Get it from Ollama

[left lodge]

You just asked the possibility

[proud bobcat]

there probably are people who've downloaded but the file is taking a long time

we're talking about a model larger than deepseek

[mystic olive]

I have it. Didn't face a prob

[proud bobcat]

what the hell is your internet speed?

[mystic olive]

Enough

[left lodge]

Lmao he is lying

[proud bobcat]

id figure

[mystic olive]

🤫

[remote vapor]

@proud bobcat where did u get the info from that Qwen will release around new years?

was it just an assumption based off if that image model reference?

[proud bobcat]

bro has a nasa pc

[mystic olive]

Bro why the hell would I lie

[proud bobcat]

well chinese new year has always been a big time of celebration

and most labs are defo done with their new models by then

remember that despite gpt and claude being the main ais here

china primarily uses deepseek and qwen

[left lodge]

I haven't seen any blasting news in the news so can't say it's truth

[remote vapor]

hehehe "here" >v<

[mystic olive]

See the graph. Chinese still prefer primary ais we use here

[toxic verge]

True

But I think that Western AI is at a disadvantage at this point

[hushed gyro]

Guys nano banana Pro still has errors

[mystic olive]

Yeah. I think it's an api issue

[remote vapor]

who would prefer GPT over kimi?

[proud bobcat]

are you actually dense

https://tenor.com/view/aniyabea-yt-man-stare-ab-yt-man-stare-aniyab-gif-10494397403125281154

[left lodge]

Yeah somewhat atleast the ones which are available to general public

[hushed gyro]

Not me because I have extreme distrust with OAI since early 2024

[toxic verge]

https://youtu.be/MWMe7yjPYpE?si=nooxXMpup2OwJNRS

[mystic olive]

Bro r u crashing out @proud bobcat

[proud bobcat]

"ah yes the chinese can access american models"

casually forgets the great firewall exists

[hushed gyro]

I heard that google AI studio is cooked and the model is unstable even on the Gemini app

[toxic verge]

I agree with him I think that generative AI has peaked

[proud bobcat]

granted you can use a vpn but the deals for ai in china are just better

[woven harness]

vpn

[remote vapor]

yeye it was ironic question >v<

[left lodge]

Now there is a new look, ui revamp

[mystic olive]

Disagree

[toxic verge]

AI is so complicated. It’s not just about data set and code. It’s so much other factors in all of this.

[proud bobcat]

brochacho

20 dollars for gptslop a month

or deepseek for free

qwen for free

glm for free

kimi (mostly) free

[toxic verge]

You got politics you got the hardware there’s so much that goes into it

[sand crown]

What is the new announcement talking about? It was already doing it

[left lodge]

China AI are not bad and they provide more thinking limits than west

[hushed gyro]

Is NB Pro free on there

[proud bobcat]

also deepseek is THE sponsored ai

[left lodge]

Idk

[toxic verge]

But this is only temporary until they get market advantage

[proud bobcat]

like hospitals, schools, and startups in china use deepseek

[toxic verge]

They’re not gonna be free forever or open source forever

[shrewd geode]

Seedance 2.0...

[proud bobcat]

why not

they have nothing to gain from being closed

[hushed gyro]

Btw kimi just removed their legacy k2 model

What the hell? Those could be backup when 2.5 is at high demand? Do they not think of contingency plans or smth?

[mystic olive]

Yeah obviously they will show they use deepseek to stay hidden.

[toxic verge]

What’s the point of having a business?

[proud bobcat]

k2.5 instant exists

[proud bobcat]

what is bro talking about 😭

[hushed gyro]

Yes but k2 is easier to use jailbreak on

[toxic verge]

You think all these people are investing all this time and money just to give out free large language models

[left lodge]

Still the current situation matters the most

[toxic verge]

No, I agree with you

[hushed gyro]

So guys any solutions / workarounds to the NB Pro error

[proud bobcat]

china has a much more "for the people" view

they subsidize

they reward research

[left lodge]

GLM-5 takes 4th place on Vending-Bench 2. Above Claude Sonnet 4.5, the state-of-the-art model less than 6 months ago. China seems to be 6 months behind the West. By June they will be ahead if the trends continue. More in this thread on why we don't think this will happen.

[proud bobcat]

companies share info

[toxic verge]

Money makes the world go round

[proud bobcat]

communism unequal capitalist worldview

here we are used to companies constantly competing

[mystic olive]

Alr bro I give up. I actually just made a stupid arguement

[proud bobcat]

in china competition is more friendly

theres no aggression

if deepseek makes something revolutionary they wont gatekeep that

as such GLM 5 uses deepseeks' sparse attention

[hushed gyro]

OMG NB Pro worked for the 3rd time today! Let's see if it works again...

Guys I really need some ideas for solving the errors

[surreal zephyr]

thats crazy good for opus

[proud bobcat]

granted its been bug reviewed by kimi but most of it is claude's doing

[surreal zephyr]

its still buggy but impressive for noncoding model

heres codex

suspension

[devout helm]

its generating since yesterday its possible to refresh it or something like this?

[surreal zephyr]

wait no i cant find the clip

💔

https://streamable.com/kaptqa

here

(fully custom and replicated to other clients, no default physics/collisions used)

[proud bobcat]

could use more tweaking but this is good

hey

maybe 5.3 codex is good

maybe ive been too harsh

but we shall see

[surreal zephyr]

its good but it has to know what to do

less creative than opus

but at coding? its insane

[surreal zephyr]

(theres a reason why wheels have to sometimes clip thru terrain and thats the engine issue sadly)

if you alt tab the fps throttles to like 5 fps

so it causes other issues if hard prevention is added

[remote vapor]

maybe Qwen 3.5 will beat opuzzzzz (no it won't, but I don't care >~< )

[surreal zephyr]

now trying to add animated water (worst engine ever bruh)

(no you cant use dynamic meshes in rblx)

[remote vapor]

what engine?

.... did opus recommend godot again?

[surreal zephyr]

5.3c beats it at everything but creativity

[surreal zephyr]

bobox

[remote vapor]

its not even out yet!

ohgosh..... bobox..... sigh.....

[left lodge]

<@&1349916362595635286> instant detection

[surreal zephyr]

it is out lol

[hasty thorn]

I can't log in to Google on my computer

[left lodge]

Damn

I was not expecting this

Kimi k2.5 and now glm 5