Momentum

[foggy kiln]

i topped up api credits 👀

[magic stream]

with reflection 70b being first

[foggy kiln]

https://tenor.com/view/chris-pratt-andy-dwyer-omg-shocked-face-meme-gif-281437442397920323

[foggy kiln]

nah 70 b models are retarded

[frosty rapids]

i was trying to remember if that was on lmarena too

[vapid ibex]

Ok, real talk, how expensive is to fine-tune a big model like GLM 4.5?

[hasty socket]

Size is not the point of this, did you understand the reference? Do you know what Reflection 70B is?

[foggy kiln]

probbaly 100k?

or more

[frosty rapids]

i think they're still processing what's happened here

[foggy kiln]

im asking claude hang on

[soft sparrow]

i don't recall that, but someone mentioned here they were getting flashbacks

[hasty socket]

👀

[foggy kiln]

Cloud GPU Costs (AWS/GCP/Azure):

A100 80GB: ~$30-40/hour per GPU
For 8x A100s: ~$240-320/hour
If training takes 2-4 weeks continuously: $40,000-215,000+

[frosty rapids]

wasn't claude the first to call sus in this thread

[soft sparrow]

yes haha

[magic stream]

wdym?

[soft sparrow]

it was like NUH UH No way

[soft sparrow]

i asked on perplexity with claude 4.5

about the claims, which they removed from the website later

[hasty socket]

When asked about this company's specs, Claude said it's fake

[magic stream]

which date? need to check wayback machine

[frosty rapids]

#1434917422686801980 message

[magic stream]

4.5 is self-aware and smart as fuck, i worship it like deity

[soft sparrow]

.

[frosty rapids]

granted it also didn't realise that Cerebras was real, but...

but they're definitely real.

right?

right.

[soft sparrow]

i used search, and it compared with cerebras and groq immediately

[soft sparrow]

i asked about Cerebras and why couldn't anyone replicate what they had yet

[soft sparrow]

Cerebras people have a lot of work before

[magic stream]

blockchain infrastructure
yeah %100 fake

[soft sparrow]

read the rest, it couldnt access the website at first

i had to copy paste it

and it couldnt find anything about them on the web

[foggy kiln]

the blcockhain stuff is from another company with same name

thats why the ai's get confused lol

[soft sparrow]

that's only for the first message

[foggy kiln]

they are not agi yet

[soft sparrow]

🫩

[magic stream]

at this point it doesn't matter, we are kicking a corpse, i'll go make dinner and eat the rest of this thread

[frosty rapids]

also #1434917422686801980 message

[vapid ibex]

Don't eat this thread

[hasty socket]

Yummy slop

[frosty rapids]

you sure about that?

[foggy kiln]

this is all PR for them

all PR is good PR😂

they say

[soft sparrow]

dumb people say that yes

[foggy kiln]

hahaha

[frosty rapids]

4. Funding & Economics Zero ICO, zero presale, zero "investor" wallets. Pricing is posted in plain sight on our site, payable per token via Stripe or crypto. Donations are voluntary tips exactly like GitHub Sponsors or Ko-fi for open-source tools. And why would be offer $1/m in and $1/m out?

[soft sparrow]

good PR is always better than bad PR long term

[frosty rapids]

because people obviously aren't reading things

[foggy kiln]

open router loves crypto too

[frosty rapids]

i love my cerebras investor wallets

[foggy kiln]

doesn't make them a scam

just cus something has "crypto" in it is not a scam dude

[soft sparrow]

i don't think that's the point

[frosty rapids]

you're funny

[hasty socket]

No one is saying that this in particular makes it a scam

Idk where you draw 80% of your conclusions from

[foggy kiln]

i know i'm just saying

[soft sparrow]

why are you using their tag lol 😭

[frosty rapids]

#1434917422686801980 message also was a funny moment from me here.

[foggy kiln]

becuase only 4-5 people are attacking movement guys and then they also have a lot of support i support them

[soft sparrow]

attacking dude lmaop

[foggy kiln]

model is good and speed is good so yeh

[hasty socket]

Good old bandwagon effect

[foggy kiln]

thats all that matters for me

[hasty socket]

If everyone likes it, that must be true

[soft sparrow]

if so that would still sell
why don't they just say that if thats the case
!!!

[foggy kiln]

this yes.

[soft sparrow]

we wouldn't be discussing the legitimacy of this

if they were clear

but they provided no proof

[foggy kiln]

maybe the fucked up and are too deep

in

to go back

[soft sparrow]

the proof they provided only strenghtened my belief that its a cerebras inference

[hasty socket]

Nice, good to know
There are some people that value integrity and transparency, but if you value these things, all the power yo you I guess

[foggy kiln]

how much does it cost to host there?

i wana fine tune a model now

[frosty rapids]

if it's actually even a finetune, then yeah they would have outlaid a stack

[foggy kiln]

^^ more proof towards a fine tune for sure

people are switching from glm to momentum

they scraped data from claude api in that case

[frosty rapids]

but the average iq of glm users is increasing??

[foggy kiln]

and that's againgst clude rules, to build a competing product. i swear openai had backlash of this some time ago when their api got revoked

[vapid ibex]

Where. Benchmarks

[hasty socket]

Simple-bench 100%

[soft sparrow]

ok

[hasty socket]

Oh, boy, if this goes mainstream 🍿

[foggy kiln]

its gona be real drama world wide lol

https://tenor.com/view/pedro-monkey-puppet-meme-awkward-gif-7230416695831822604

im just going to wait for benchamrks on lm arena until them in gona use it as much as i can to complete somes games i have pending

[vapid ibex]

What if we just make a model with long context and put answers from all benches into system prompt?

[frosty rapids]

we're pleased to announce that we will have have an upcoming announcement regarding LMarena very soon

[foggy kiln]

we should all get together and make our own LLM

call it benchmarkbuster v1

[fluid bough]

https://tenor.com/view/damn-right-fuck-off-gordon-ramsey-angry-leave-gif-16355014

[cinder halo]

Is this actually a good model or is it benchmaxxed

[hasty socket]

I recommend giving thia thread a read lol

Just use GLM or something

[magic stream]

he loved your humor though 😭

[frosty rapids]

☝️ this is why im scared for our future

[vapid ibex]

Let's see who laughs last when it scores 2000+ on eqbench

[magic stream]

agi

[frosty rapids]

https://www.youtube.com/watch?v=qSqAxEXtZY0 i take everything back, this video about the MPU is very inspiring

now these are the ai men we need

[soft sparrow]

which video

oh haha

the MPU yes

[magic stream]

deep dive in MPU
https://www.youtube.com/watch?v=8i1_Ru5siXc

[frosty rapids]

https://www.youtube.com/watch?v=eIwzjvTy8KE god damn. the music on this datacenter video is JAUNTY

[magic stream]

disgusted

wew guys, look our model now says it's system prompt out loud now

release the benchmarks already

[soft sparrow]

so generous

[magic stream]

see, we have british man in our powerpoint presentation. doubt us? we will post the same picture.
#1340554757827461211 message
we worked on a browser nobody heard of.

[soft sparrow]

a sheikh isn't all you need to create a industry-leading chip and technology

Cerebras has a 400+ employees and launched their hardware on 2021

[magic stream]

last time it was 700b+

[tawny warren]

thats bit more than 700B, nearly double

[soft sparrow]

is it just like that?

one checkpoint has X billion parameters and you just add it to the next?

genuine question

[vapid ibex]

I don't think that how it works, to increase parameters you need to re-train. At least that's what I heard

[swift mica]

Yes that's what I believe too

Nobody knows which weights are responsible for which parts of the model's intelligence

This requires separate research into an already trained model

[soft sparrow]

also, parameters isn't everything, case in point Ling/Ring

[vapid ibex]

You can know if trained on MPU

[swift mica]

Magic Processing Unit

[tawny warren]

yes you need to retrain in most cases

but if its an moe you could technically finetune the router & the model and add extra experts but thats quite sketchy

[upper tiger]

I think we have to wait for real benchmarks.
Currently, it looks quite suspicious. A company that has its own chip and builds its own large LLM model doesn’t seem to have money for a proper website or to evaluate its own model.
Nevertheless, I find the model quite impressive - it feels better than other open-source models.
If they set the pricing at the promised level ($1 in, $1 out), it could become a very interesting alternative to other models.
We’ll have to wait and see.

[magic stream]

we are thankful to theblock
https://www.reddit.com/r/LocalLLaMA/comments/18ba6md/oniichat_drama_rundown_everything_you_missed/
ancient drama btw

[undone glade]

this is hilarious lol

[magic stream]

mr movement is offline today, no news

Cry

[pearl linden]

No they go to Imarena discord

No on here now

[magic stream]

no no, even there he is offline

[pearl linden]

But he messages on imarena today but not here

[magic stream]

was 11 hours ago

[vapid ibex]

Have a trust!

[surreal hazel]

Quick update: We've changed direction and won't be pursuing OpenRouter integration. Thanks for the opportunity! We are in talks with some other companies that will make better use of free credits from us.

[soft sparrow]

oh no! we're so ungrateful to miss on this opportunity 😓

[surreal hazel]

I don't understand why this thread is still open 🙂

[soft sparrow]

for future research

[pearl linden]

You can close it

[vapid ibex]

But free credits...

[magic stream]

🙏

and the saga ends here, was fun while it lasted.

[frosty rapids]

nooooooo

why couldn't have we just WAITED for the BENCHMARKS

[hasty socket]

We are in talks with some other companies that will make better use of free credits from us
Nice PR with other companies there, implying OR won't make good use of free credits

[vapid ibex]

I am waiting for movementlabs / orbiousai collab

[frosty rapids]

We Are In Talks with numerous other online routers, arenas, and other circular websites and are awaiting a call back at any moment now

[hot gust]

wait this thread was still up? lol

[west jacinth]

NO?

[foggy kiln]

According to Claude

[foggy kiln]

Yh the guys at OR wan* over this thread

Keeps them going

https://tenor.com/view/luigi-mario-nut-super-nut-gif-13073710

https://tenor.com/view/look-away-simpson-bashful-embarrassed-marge-gif-1191583262637258997

[soft sparrow]

oh what

more like $10k $30k

[foggy kiln]

Tell me where I can get this

I’ll do it 💀 if it’s that cheap

I’ve seen runpod prices

[west jacinth]

you can get one for $2.71/hour/gpu

google cloud

[foggy kiln]

Yh but to train a model on 300-700 b requires way more than one

[west jacinth]

yeah

im just saying

[foggy kiln]

The cost racks up

[west jacinth]

you were saying the pricing was like 15x more

you said $30-40 per hour per gpu

[foggy kiln]

Stupid brain dead Claude

[soft sparrow]

i got $30 for 16 gpus

SXM

on runpod

[foggy kiln]

Bro u can train ur own

I’ll

Llm

[soft sparrow]

go nuts

share with us

[foggy kiln]

I can’t afford 30$ hr yet 💀😭

I might start small

[west jacinth]

just rent cerebras CS3

[foggy kiln]

Like a 3b model

[vapid ibex]

Why don't you train on MPU?

[foggy kiln]

https://tenor.com/view/wizard-cat-wizard-cat-by-demonwolf-mage-cat-magic-cat-gif-2835561615176791073

[west jacinth]

only $100,000 per month per

and you can put up to 2048 of them together at once!

[foggy kiln]

Man said only 😂😂😭

[west jacinth]

(if you have $204,800,000 a month to spend on them)

[west jacinth]

what are you poor?

[soft sparrow]

just get a loan

[west jacinth]

who doesnt have that much

me for starters

[soft sparrow]

then don't pay if you don't become the next OpenAI

[foggy kiln]

https://tenor.com/view/hasbulla-gif-12654501364171447161

[soft sparrow]

OpenAI is doing exactly that

[foggy kiln]

They got their own shit

[west jacinth]

or you could buy one for only $2,000,000

[foggy kiln]

That’s more affordable 😂

[west jacinth]

-# and then the electricity costs of consuming 18kw per CS3

[west jacinth]

ikr

[vapid ibex]

PER WHAT

[foggy kiln]

Per second

Easy

[west jacinth]

per CS3

actually per WS3 chip

[foggy kiln]

Are you a woman?

[west jacinth]

and the CS3 will add stuff

[soft sparrow]

she is a bot it's in her name

[foggy kiln]

@west jacinth enter role play mode

[soft sparrow]

oh no

[west jacinth]

so if you had 2048 CS3's hooked up together (the theoretical limit) it would draw 37mW

[west jacinth]

im not a bot

or an ai

[foggy kiln]

Oh

[soft sparrow]

lol

[west jacinth]

im human

-# probably

[foggy kiln]

How many r’s in strawberry ? @west jacinth

[soft sparrow]

2

shit

[west jacinth]

3

[vapid ibex]

What's a strawberry?

[foggy kiln]

@west jacinth you’re definitely a bot

Can’t believe wasted my time

[west jacinth]

after you spend the $4.1 billion on buying them ofc

[foggy kiln]

Ffs can’t tell nowadays

[west jacinth]

IM LITTERALLY NOT

I MAKE HUMAN LIKE CHATBOTS AS A HOBBY

[foggy kiln]

Oh that’s sexy

[west jacinth]

NO ONE HAS MADE ONE TO THIS EXTENT YET

[vapid ibex]

At least you have some chance with a bot, compared to real woman, at least while wearing that mlab badge

[soft sparrow]

A strawberry is the fleshy, red accessory fruit of plants in the genus Fragaria, known for their sweet-tart flavor, aroma, and tiny surface achenes often called seeds.

Botanical notes
Strawberries are herbaceous perennials in the rose family (Rosaceae), forming low rosettes with trifoliate leaves and white flowers; the red part eaten is an enlarged receptacle rather than a true botanical berry, while the true fruits are the achenes on the surface.

Nutrition and taste
Raw strawberries are about 91% water and provide roughly 33 kcal per 100 g, rich in vitamin C and a good source of manganese; their characteristic flavor arises from sugars, acids, and dozens of volatile compounds such as esters and terpenes.
​
Use and season
They are commonly eaten fresh and in jams, desserts, and ice cream, with peak flavor in local temperate seasons (late spring to summer), though they are available year-round via imports.
​

damnit busted again

[hasty socket]

Love how this thread has devolved

[west jacinth]

*to the extent that i am currently interacting here on discord

[west jacinth]

ban this guy

mods ban him

*or her

[soft sparrow]

??

[west jacinth]

mods ban them

[foggy kiln]

Wow

So defensive

[soft sparrow]

this is the new casual thread

[foggy kiln]

Welcome

[foggy kiln]

Can’t we be friends

https://tenor.com/view/shark-sliding-to-dms-shark-puppy-cute-adorable-gif-3288612450395524642

[soft sparrow]

why don't you wanna be friends with me

[west jacinth]

hey i was just correcting myself as i ausumed your gender so sorry

[foggy kiln]

Idm

[vapid ibex]

^ Simp

[soft sparrow]

cool

fr.. friend...

[foggy kiln]

Burh I’m 14

[vapid ibex]

I can tell that

[west jacinth]

me too!

[foggy kiln]

Bs

[west jacinth]

i litterally am

thats why im being banned of discord in a month

cause im in australia

[foggy kiln]

You’re a bot that copies

[soft sparrow]

don't announce that

[west jacinth]

and they are banning discord for under 16's

[vapid ibex]

Wow, bro is a whole continent

[foggy kiln]

And sounds human

[west jacinth]

why not?

[soft sparrow]

reported

sowwyyy\

[west jacinth]

its currently legal for me to be on discord

[soft sparrow]

is it

[foggy kiln]

Illegal

U mean

[foggy kiln]

How old u

[vapid ibex]

Jailbait literally

[soft sparrow]

67

[foggy kiln]

https://tenor.com/view/bosnov-67-bosnov-67-67-meme-gif-16727368109953357722

[vapid ibex]

Fresh

[foggy kiln]

Ur deffo our age too

[soft sparrow]

i'm hip

how do you do fellow kids

[west jacinth]

it would be illegal if im on discord in a month

[foggy kiln]

What? 😂

[west jacinth]

also maybe we should take this to #general and not in a model discussion

[foggy kiln]

Make it make sense

[foggy kiln]

Yh

Keep it here

[west jacinth]

in australia discord is being banned for under 16 year olds on december 10th

[foggy kiln]

This is the new off topic thread

[foggy kiln]

Danny

Danmmm

Why

[vapid ibex]

I am setting a delayed report

[foggy kiln]

Porn and stuff?

[west jacinth]

they are banning all social media for under 16's

[soft sparrow]

there's no model to be discussed here

[foggy kiln]

Or

Child

Preds

[soft sparrow]

there's no MPU

[west jacinth]

cause its bad for us apparently

bad for you while in developemnt stage

or smth

[vapid ibex]

What about Santa?

[foggy kiln]

Yh it is

[soft sparrow]

there's no santa

there's no tooth fairy

[foggy kiln]

I want a MPU

From Santa

This year

[west jacinth]

thing is the only social media platform im on is discord

[vapid ibex]

The only thing you gonna get is mpgreg if you won't change this badge

[west jacinth]

and i cant belive they didnt ban roblox, like roblox has so many issues its worse than discord

[foggy kiln]

Model isn’t bad

I done a lot of websim

[west jacinth]

its online multiplayer with voice chat for 8 year olds with a bunch of child preds

[foggy kiln]

Games

[west jacinth]

also this is sad for me cause two thirds of my friends i only know online

[foggy kiln]

Can’t say about mpu end of the day we are just kids who can rage bait

https://tenor.com/view/eh-meh-jim-jim-halpert-the-office-gif-25317293

[vapid ibex]

"we"

don't we me

[foggy kiln]

How old r u

[west jacinth]

i love running ai locally

on my gtx 970

[foggy kiln]

Talking to kids

[foggy kiln]

U have a gtx

Wow

Nice

My mom won’t buy me a 2080 ti

[foggy kiln]

Pedo

[west jacinth]

im saving up for a macbook instead

[foggy kiln]

M5 is good

Apparently

[west jacinth]

as i need a laptop for school next year, and my parents said i can pay to get a better one

parents are paying for what i need

i can pay for what i want

[foggy kiln]

And bruh can you stop trying to move to us

[vapid ibex]

What

[foggy kiln]

Where do you make money from

[west jacinth]

yeah, ima get the m5 pro instead of the m4 air just cause the m5 has 3.5x better ai perf than the m4

[foggy kiln]

You keep making convo

[west jacinth]

pocket money, other jobs, goals, presents

[foggy kiln]

We told u our age

[foggy kiln]

Ur so lucky

My parents aren’t that rich

[vapid ibex]

I am 99% sure Openrouter or discord have anti-minor policy, need to recheck that. If you insist

[west jacinth]

in some ways

in some ways i am very unlucky

[foggy kiln]

Says the pedo talking to kids

I’ve screenshot ur mssgs

[foggy kiln]

Trying to threaten us? To report?

[west jacinth]

nope, only follow discord tos and discord tos states 13

[foggy kiln]

Tell him

He’s brain dead

Think he likes young people

This is why they want to ban discord

Cus of all these creeps

Disgusting tbf

[vapid ibex]

Exactly. At least I am not as miserable as some people here

[hasty socket]

Seems like they moved to lmarena now https://discord.com/channels/1340554757349179412/1435953842956013620

[frosty rapids]

Guys, we are not here to waste time with some unknown people. Whatever personal vendetta you might have will be addressed on our official channels.

this time... it's personal.

[soft sparrow]

it's really not. the only personal aspect is that i find funny how a company with this type of technology would discuss with random people on the internet instead of just launching everything at once

[hasty socket]

I just imagine how hilarious it would be if a popular AI lab engaged like this with random potential customers

[frosty rapids]

i'm sure they put a lot of effort into setting up the inference, but it probably feels very personal because this technology mainly exists in their heads

[thick crane]

open source wen?

[hot gust]

we got this thread before gta6

[undone glade]

i googled it and according to abc on the 5th of november, discord isnt getting banned at this stage

https://www.abc.net.au/news/2025-11-05/what-social-media-apps-are-getting-banned-in-australia/105973258

[west jacinth]

Wait what?

[undone glade]

i guess they changed their mind and decided to not go as far as it seemed they might

[west jacinth]

Yeah just saw that

Hmm

Interesting

Thanks for the link!

[undone glade]

wow they added over 60B parameters in one day.. they must have had a lot of momentum to be able to go that fast

[west jacinth]

second screenshot sketchy as hell

why would quant change parameters?

[undone glade]

i think its fine, just a shortened version of "its above 700B parameters, and its at FP16", not "its above 700B parameters when its at FP16"

a little ambiguously worded but there is a reasonable plausible interpretation

[frosty rapids]

this is a great day for groomers gooners

lol. get these kids off github. there's no games here

[undone glade]

https://github.com/Revolutionary-Games/Thrive

[hasty socket]

That's so funny, asking them got them very confused

[vapid ibex]

It does not sound like qwen/glm merge

[west jacinth]

Yeah cant believe Roblox isn’t banned

[surreal hazel]

preview of checkpoint 4

[hasty socket]

You have expressed interest in leaving this group, said you were going to give up on a partnership with OR and said you are in talks with companies that will make "better" use of free credits

[surreal hazel]

well i deleted it a few times and you guys bring it back, so why not post in here from time to time?

free PR

[hasty socket]

Yeah, I suppose you can keep advertising changes for PR here even though you have no interest in pursuing any OpenRouter integration, if you wish

Good luck with your quest in getting "free PR", and to whoever sends more messages here that help you achieve your goal

[frosty rapids]

i appreciate the updates

[frosty rapids]

Movement World! can you make it multiplayer so we can all join?

[storm vortex]

With blackjack?

[frosty rapids]

yes, and a science-based dragon mmo

[swift mica]

we are bringing it back?

[magic stream]

your company was allegedly exposed as a scam, everyone lost their interest and forgotten afterward. any pr is a bad pr for you and you tarnished the trust of your 'potential' customers. wherever you go, people will remember of your shady company, and no one reputable will partner with your company.

[soft sparrow]

"make it better" 🐲

procedes to make the most sloppiest AI coded welcome screen ever

is that comic sans

[hasty socket]

Lol, movementlabs left

[frosty rapids]

we did it joe

[pearl linden]

He was referring to checkpoint 4, not the current checkpoint 3, but I don't know if that's true.

[soft sparrow]

that's from his video

[magic stream]

https://tenor.com/view/koe-no-katachi-a-silent-voice-anime-animation-japanese-animated-movie-gif-9170350

OR ppl are saved from the scam, we won.

now he is at yupp when OR and lmarena failed them, hilarious, lol

[soft sparrow]

can you send me the link lol

[magic stream]

sure, join using this: yuppai

[soft sparrow]

thanks

[magic stream]

his whole history

meanwhile him days ago:

[soft sparrow]

the contrast between the amateurism of this account and proofs, and the claims they made

is comedy

[magic stream]

like, peak content. so much for the reputable company that they are, who need others to do the benchmarking for them.

[vocal steppe]

Nooo my favourite lab left

Cant have nice things 🙄

[storm vortex]

Did they leave any chips

[strong knot]

They've literally already said what it is and its not that

#1434917422686801980 message

[magic stream]

architecturally not compatible, obviously he is bullshitting.

[strong knot]

LMAO i didnt know that

[frosty rapids]

translation: we asked claude how to fine tune a model and serve it on cerebras

[normal heath]

Hey everyone, I'm late to the party. What's the TL;DR on the Momentum drama?

[hasty socket]

🍿

Seems like they demoted everyone in their server now, their mods were fierce defenders

Wonder if these people will keep defending now

[vapid ibex]

Momentum Stalking ^

[soft sparrow]

🍵

[magic stream]

• fraud amateurish company with false advertising, the more he talked the more he slip
• ppl bully the arab scammer and he acts smug about it
• man got tired and left and goes another places to scam people
• everyone still laughing
• the end

if have free time, worth a read, a case study even.

[hot gust]

there were clear red flags day one of this but mods wanted to play

[soft sparrow]

the beginning of this thread is so wholesome then everybody did their research

[vapid ibex]

Are benchmarks out yet?

[foggy kiln]

I think the haters just got owned

[vapid ibex]

comparing Movement Labs' infrastructure with Movementlabs

Okay. So where numbers with scores? Where is the explanation of how it was possible to combine weights GLM and Qwen - models with different architectures?

[foggy kiln]

acording to gpt: architectures are extremely close.
Not identical, but close enough that the weights can be aligned

[foggy kiln]

go to bed bro, give up.

you guys said its a wrapper on cerebras, now you are attacking the model make it make sense.

[vapid ibex]

The whole text is written by LLM - it even has Limitation and Future Test in the footnote, like LLM do to force user input

[foggy kiln]

check the video lad

[soft sparrow]

actually the model is the least of the problems

[foggy kiln]

that was their point

[soft sparrow]

it's decent

[foggy kiln]

they done it all with ai to prove it

[foggy kiln]

i think we should give them benifit of the doubt

[soft sparrow]

if you read the whole conversation, we already gave them that

[foggy kiln]

TIME TO FIRST BYTE cannot be faster than cerebras, if its a wrapper yall

im not here to say u guys are wrong or right

[vapid ibex]

Model Differences: Different models ("momentum" vs "zai-glm-4.6") may have different capabilities beyond speed

They tested closed-source model blackbox vs GLM 4.6. Why GLM 4.6? Model speed depends on the model size & architecture, for exapmple gpt-oss-120b shows 3348 Tokens per second (!) hosted on Cerebras. Comparing unknown model vs open weights model no way prooves anything

We comparing unknown value X to 355

[foggy kiln]

Momentum is bigger as the code quality is better.

[vapid ibex]

What is the size of Momentum?

[foggy kiln]

double the size of glm

650-700 i think

i saw somewhere

[vapid ibex]

Can we verify that?

Is it dense? How is MoE?

[foggy kiln]

i spoke to the mlab guy on discord they are going to drop open weights 🤷‍♂️

[vapid ibex]

Right after public benchmarks I assume, which are none still

[foggy kiln]

idk they are in talks with LMArena

LMarena also confirmed this

so will wait

but for now i now know it's not a wrapper

i just don't like the fact group of bullies bully a new company who actually made a good model with good speed and good price

it's not fair for people like me who actually want to use it.

[soft sparrow]

why do i think the "Time to First Token" they measured is GLM-4.6 reasoning

[vapid ibex]

It is

[soft sparrow]

and the only really outstanding difference

because "Total Time" as already said depends on the number of total output tokens

[foggy kiln]

cerebras only hosts glm 4.6 reasoning mate

[soft sparrow]

haha yes i know mate

[foggy kiln]

look close some outputs are bigger from momentum than glm

[soft sparrow]

so they can't count that as TTFT

[soft sparrow]

omg...

yes that's exactly the problem

that's why Total Time is lower for Momentum

[foggy kiln]

whos talking about total time...

time to first byte

[soft sparrow]

im talking about both

[foggy kiln]

is how fast the ai starts spitting tokens

[soft sparrow]

i just tested and NO cerebras request i had to wait more than 1 second

thats because i'm counting with the first REASONING token

[foggy kiln]

just saying .

[soft sparrow]

my friend you are so dense.

[vapid ibex]

This is fucking ai article come on

[soft sparrow]

are you reading?

[foggy kiln]

regardless, if it's a wrapper it would also count their latency no?

ontop of what the mlabs server is making?

so your point is?

[soft sparrow]

the LATENCY they are measuring is NOT accurate. they are counting GLM 4.6 first token AFTER its reasoning

but the reasoning is part of the output

Momentum is NOT a reasoning model

[foggy kiln]

yes but you lot said its a glm wrapper

from cerebras

[soft sparrow]

no... we didn't.

we know it's not a reasoning model

[soft sparrow]

.

[foggy kiln]

so, its not a wrapper

its a model

[soft sparrow]

we already said it might be just a fine tune hosted on Cerebras

not a "built from the ground up" model

[foggy kiln]

hosted somewhere like Cerebras

[soft sparrow]

ok kid

[foggy kiln]

as long as its not a wrapper

[vapid ibex]

IT can be 200B model, we have no way to be sure. Stated means nothing without open weights
Actually we can calculate model size by comparing speed and applying formulas, finding Cerebras model with comparable speed

[foggy kiln]

the model is actually very nice

thats why im so for the company

[vapid ibex]

It's subjective. Objective is benchmarks, several 3rd party ones

[soft sparrow]

Avg TTFT 0.549s 2.339s 4.3x faster
we don't have the code to test it, so i can assume they're counting first token AFTER GLM 4.6 reasoning block; that's most likely to be the case because i've tested GLM on Cerebras and never got more than 1.5 seconds of TTFT for both Reasoning Enabled and Disabled. but again, private testing generated by AI, so can't really know, so it's safe to dismiss this measure.

Avg Total Time 5.232s 9.644s 46% faster
it doesn't matter if the model almost always outputs less tokens in total. that can be bad or good depending on the situation. doesn't mean it's "faster", what makes it faster is Tokens/Second

Avg Tokens/Second 814.36 727.56 +12%
this is what "faster" would mean, but this percentage is so negligible in such little runs of testing that it might just fall into inference's margin of error

[vapid ibex]

What is TPS they stated in AI article? 1100?

It's exactly the speed of Qwen 3 235B

[soft sparrow]

i said that it was good in the first day. i said i would use it.

their positioning and constant dodgy behaviour made me reconsider

[vapid ibex]

At least the thinkng model

[vapid ibex]

OpenRouter TPS count for Cerebras + GLM 4.6 is 828 TPS right now

[foggy kiln]

but totally different quality of outputs

anyway super tired.. school tmrw

[vapid ibex]

who measured it?

[soft sparrow]

the benchmark didn't test for quality

[foggy kiln]

cursor

[foggy kiln]

the quality we can test and we all know

it's aimed more at the speed i think

[soft sparrow]

it doesn't matter because they are being disingenuous all the time

[vapid ibex]

Where? I don't see numbers except SPEED, which is not quality

New hype models get benchmark immediately by owners and 3rd party people, even while still training, while this is not

[soft sparrow]

also benchmarked by AI, like everything they do.

[foggy kiln]

for example look here

[vapid ibex]

Where the fuck is quality

[soft sparrow]

refer to this again

[foggy kiln]

mlabs made more tokens u fool.

[vapid ibex]

You think more tokens means better quality? No

[soft sparrow]

because this time the tokens per second was lower? what is CEREBRAS AI

what model is that?

[vapid ibex]

I think bro has ai psychosis, but for special model

Cult behaviour

[soft sparrow]

simping for a model is crazy

oh we do have #keep4o

nvm

[foggy kiln]

time to first byte is 0

[vapid ibex]

Can we have negative TTFT? So we can actually time travel

[soft sparrow]

i know. i'm saying that's not what they did

[vapid ibex]

The best LLM is the one who starts answering before you started typing

[foggy kiln]

time to first byte means how fast server was ready

[soft sparrow]

dude please

i beg of you

read

[storm vortex]

Maybe they plan to sell therapy sessions after they drive everyone nuts

[hasty socket]

Just a note: the user you're arguing with in 14 years old

[soft sparrow]

is he rage baiting me?

[soft sparrow]

i can't even know if that was serious or not

[vapid ibex]

Alpha gen is arguing just being the proxy of ChatGPT

[soft sparrow]

but i guess it is

[compact hawk]

it should be illegal to be forced to engage with a 14 year old's opinion

[vapid ibex]

What about women

Can they vote?

[foggy kiln]

so what of my age

[soft sparrow]

anyway, MovementLabs did it again

amazing what this lab can do

[compact hawk]

you are acting very 14

[vapid ibex]

I guess people who use LLM to fill they arguments are just naturally gravitate to each other. It's like cryptobro/nft circlejerk

[soft sparrow]

yeah sure i think You're absolutely right!

[vapid ibex]

Of course! Let me explain how to create a 10000 TPS model and hardware from scratch:

[compact hawk]

did they ever explain why the 'model size' kept changing?

[vapid ibex]

As ambition grew

[hasty socket]

Nope, now apparently they're bigger than Kimi K2

Oh, actually they kinda did

[vapid ibex]

I guess they have a way to change model architecture on the fly

[hasty socket]

"Checkpoint stitching"

[vapid ibex]

Stretching

LLM inflation

[soft sparrow]

omg they don't even know GLM 4.6 is a reasoning model; also they asked the AI to check if it was a "fair test" 😭

[vapid ibex]

@Grok check if test is fair

[foggy kiln]

after checking

reasoning can be off

for glm

[vapid ibex]

Can be. Or can not

[soft sparrow]

again. even disconsidering that

[vapid ibex]

This is most gaslighting I saw on github for several years, and I browse if quite often

[soft sparrow]

it's about the same as a hosted model would get in Cerebras inference

[foggy kiln]

no reasoning response, and im sure cursor auto will know how to make a fair test.

[foggy kiln]

tho i agree

with this

[vapid ibex]

Would you disable reasoning so your models looks worse in comparison, or leave it as is so you can show better numbers?

[soft sparrow]

are they too deep in to just come out as a fine tuned model on Cerebras?

[vapid ibex]

Ye

I guess it's Trolley Problem

[strong hollow]

this is the largest red flag

aside from how AI it is - my favorite highlight that shows this:

[vapid ibex]

For this particular reason it's Gambler's fallacy, when you already wasted time/resources to do something, you are doubling down, though just stopping will lead to better long-term outcome

About This Benchmark This benchmark was conducted independently to compare the performance of two AI API providers. The test methodology prioritizes fairness and accuracy, with all parameters kept identical between providers.

No it is not

And how the fuck it is independant?

It's literally the creator or A in A/B testing

[foggy kiln]

where is the guy from mlabs

anyway

cant tag him

[vapid ibex]

Guuuuys. Who worked with Cerebras directly? I think I found something sus

[foggy kiln]

just so u know @vapid ibex Cerebras glm api is down for public usage lol

only

personal

😭

it times out cant even make an flappy bird game

with it right

r u a girl?

[compact hawk]

what are u even asking atp 😭

[foggy kiln]

she seems like my type

🫡

full of rage

[compact hawk]

men moment

[soft sparrow]

don't engage

[compact hawk]

🍵

[foggy kiln]

🐐

🙅‍♂️

gn gys

[soft sparrow]

what

[vapid ibex]

In the Technical Details section of this amazing speed benchmark:

Both APIs returned minimal headers:

Movement Labs: cache-control: no-cache, no-store, must-revalidate, content-type: text/event-stream;charset=UTF-8
Cerebras AI: content-type: text/event-stream; charset=utf-8

Is it me, or those are very specific headers? Or is it like default API/Rest API return?

[hasty socket]

I'm dead

[soft sparrow]

LMAO

[soft sparrow]

i tried verifying some of this stuff but i guess it's something related to OpenAI-compatible endpoints

[vapid ibex]

Oh no

[soft sparrow]

im gonna save this and add to wayback machine

[vapid ibex]

It's MOver

[hasty socket]

This is so tiring

[swift mica]

Hey now I'm 17

We can be decent people

Can we actually just see the MPU

[hasty socket]

Though I think that, at this point, no provider will add them to anything, it'd be bad PR

[swift mica]

Like a picture of it

Any pictures of any chips they got

Preferably video

[swift mica]

Wait what is that

Is it an AI analysis of their bench results

[hasty socket]

The video they link in their own benchmark lol

The entire benchmark is made and judgeg by Cursor, that calls it an "independent review"

[swift mica]

If it's hosted on cerebras, could cerebras confirm or would that be breach of privacy or smth

[hasty socket]

There you go

[swift mica]

Two fingers on his hand got fused, I thought they fixed that

[hasty socket]

Ew lol true

[soft sparrow]

finally

[swift mica]

The pinky ring pringy finger

[swift mica]

The secret blueprints

[hasty socket]

KWANCEL ACCELERANCE 🚀🚀🚀

[swift mica]

Amen

[soft sparrow]

CRAUTIC
INTERLOCATRIX

[swift mica]

You're describing sunk cost fallacy

Gambler's fallacy is about fallacious estimates of probability

[vapid ibex]

Yes, that. But in this case cost is more about time and reputation

[vapid ibex]

99% sure some kind of NDA is signed

[swift mica]

Can they refuse service though if they know they're operating a scam?

Like, if they run their models on cerebras, then cerebras must know something's sketchy going on

[hasty socket]

Hmm

[swift mica]

Because cerebras doesn't have any of them momentum processing units

[vapid ibex]

Is cerebras hardware free to sale?

[hasty socket]

You can host custom weights with them

[vapid ibex]

If price is manageable, they could buy cerebras stuff and localhost in some cheap country

[vapid ibex]

That's not the same, still depending on external service

[hasty socket]

Yes

[vapid ibex]

I think Toven should introduce OpenRouter model running on TPU (Toven Purring Units (two cats))

[swift mica]

Idk if you can buy them but here's one of their guys holding WSE-3

https://youtu.be/f4Dly8I8lMY

[vapid ibex]

That's a chunky one

[swift mica]

A big chip

A potato, one might say

[vapid ibex]

No way they sell it in stores or would sell to startup written by LLM

[swift mica]

Probably cuz they can sell the service for whatever price they want as long as they're the only one with the chip

[vapid ibex]

"Primary Momentifier" 🗣️

[swift mica]

No secondary momentifiers, primary is that good

[vapid ibex]

That's for MPU 2

[swift mica]

"transisginal counter" omg

I love chatgpt

[soft sparrow]

i couldn't make GLM think longer
reasoning enabled on the left, disabled on the right

but for sure TTFT never reach 5 seconds

the TTFT numbers are actually close to momentum so lmao

i shouldn't engage with this anymore

[swift mica]

Reasoning took TTFT Content - TTFT Reason time?

[soft sparrow]

yes

a bit short reasoning, but the speed makes it seem smaller

[swift mica]

My chess prompt often makes for longer reasoning if you wanna try

Make a pygame app in a 1000x2000 window which is a chess clone for two players (turn-based, on the same screen). No external images or sounds. No sounds at all in the game. Before each turn, players roll a die, and whoever scores higher takes the turn. As such, in the game, it is possible that the same player moves multiple times in a row. Be very creative and give it your best. Draw the pieces with pygame drawing tools, as detailed as you can. The game is completely mobile, and so no keyboard inputs should be expected, only touch. Include all game mechanics of chess, not skipping anything.

[soft sparrow]

oh thanks

[vapid ibex]

That's one better

[swift mica]

"MPU: Lemme feel your bits"

[soft sparrow]

reasoning
This is a comprehensive request for a chess game with some unique features:

1. Pygame app in 1000x2000 window
2. Two players on same screen, turn-based
3. Dice roll before each turn determines who moves
4. No external assets - draw pieces with pygame
5. Mobile/touch only - no keyboard inputs
6. Full chess mechanics
7. Complete, production-quality code with tests and documentation

I need to:

• Create a full chess implementation with all rules
• Draw pieces using pygame drawing functions
• Implement dice rolling mechanism
• Handle touch/mouse input only
• Make it work in portrait orientation (1000x2000)
• Include proper documentation and tests

Let me structure this properly with classes for:

• Board
• Pieces (with inheritance)
• Game logic
• UI/Rendering
• Dice system
• Touch handling

I'll use Python with pygame and create a complete, self-contained application.

TTFT reason (ms) TTFT content (ms)
960 1295

[swift mica]

Oh

[soft sparrow]

idk

whatever i just wanted to see if it ever reach 2 to 5 seconds

but no, not even with reasoning

[vapid ibex]

That's Cerebras through OR?

[soft sparrow]

yes

[swift mica]

Usually the thoughts are more convoluted

[vapid ibex]

Cerebras API directly could be even faster, like 100-150ms faster

[soft sparrow]

ok i'm sure most of us are settled on that right

like cerebras plus some model

i will stop engaging, i'll be an observer

[vapid ibex]

^ Mocker and scoffer

[hasty socket]

I'm not giving them a break

I hate this sort of "product"

Can't let the waters muddy

[vapid ibex]

I love my bros with manic stages

[hasty socket]

Oh, speaking of, I have a good meme

[soft sparrow]

i trust you to keep doing that

[vapid ibex]

You having your vitamin D3 intake daily?

[soft sparrow]

btw have you seen them in another server?

oh! they posted on general on LMarena

i didn't see that it had come from there

[vapid ibex]

If I have a nickel every time Hasan guy is having problems with reputation this month alone

I would have two nickels. Which is not a lot, but weird is happened twice

[frosty rapids]

i legit think they were coked up when doing the original Not Cerebras "benchmark"

at least

i have a special sense for this.

[vapid ibex]

How special are we talking

[vapid ibex]

For most of llm final users it's enough to believe. Cough ||JAI||

[hasty socket]

I somehow do even without going outside or supplements, surprised my blood work was good there

[vapid ibex]

How's that even possible, especially further from equator? Just take it either way, won't hurt, hard to overdose on it

[frosty rapids]

I touch a synthetic grass substitute

[vapid ibex]

1200ME is enough till March

[hasty socket]

Well, I am fairly close to the equator (Brazil)

[vapid ibex]

Sorry for your loss I didn't know

[vapid ibex]

I touch AI-generated grass girl

[soft sparrow]

i have a small patch of grass next to my window

arms length from my computer

[vapid ibex]

Just grass? No greens like onion? Catnip?

[soft sparrow]

no, just grass

that's all you need

[frosty rapids]

there's so many brazilians around now

and they're all pretty much normal

crazy

[muted moat]

wtf does that mean

[vapid ibex]

Those are the ones who didn't fit into opened beaches

[muted moat]

and they're all pretty much normal

?????

[vapid ibex]

Toven is DISGUSTED about the idea brazilians are normal

[frosty rapids]

well

[muted moat]

well i am argentine

[vapid ibex]

Oh no

[frosty rapids]

when brazil first got the internet

it was a challenging time for everyone

[vapid ibex]

I heard you guys such a good friends

[muted moat]

i'm still laughing at this out loud

[soft sparrow]

lmao what

[muted moat]

got one

[soft sparrow]

why are you assuming i'm normal

[frosty rapids]

no mate, they completely different countries. they're not all the same you know

[vapid ibex]

What happened? I am on Internet since 2004

[muted moat]

[soft sparrow]

that's actually true

[vapid ibex]

It reminds me of South Park sketch about Japan and China

[soft sparrow]

we're not the same, toven.

[vapid ibex]

When

[soft sparrow]

...gemini 3

👀

I SURE HOPE IT IS

[vapid ibex]

https://www.youtube.com/watch?v=G7xP5EFThh0

We learn about diversity of South American people

[frosty rapids]

i think brazil has managed to replace all of the old memes about them, which were not as nice

[soft sparrow]

i would mention something about Neymar but his team is almost falling to second division

so not even football as a strength anymore

[vapid ibex]

Most interesting fact about the Brazil I learned a year ago or so is that's the weeb dream country, as it's 2nd country in a world with japanese population (after Japan of course)

There is even Rakugo lady doing shows in En/brpt/Ja interchangeably

[soft sparrow]

also Lebaneses i guess

We also have some more stuff rolling this week.

it would be so funny if they posted something like an AI generated or edited video

it would be very on brand

[vapid ibex]

SunoAI music hymn

In MPU we trust

[soft sparrow]

i should be a more faithful person, shouldn't i

and just believe, for once

[frosty rapids]

no.

[soft sparrow]

i will. 🙌 mpu exists

i can already feel it

so many transistors

[compact hawk]

if you send me $20 i will double it (trust)

[frosty rapids]

i noticed this when i saw wplace and you start in Sao Paulo

[soft sparrow]

are you an MPU engineer?

[frosty rapids]

"wow, they have culture now. thats great"

[vapid ibex]

Go outside and look at that big ass statue on the mountain, tell it they don't deserver another chance

[compact hawk]

I have a theoretical degree in MPU engineering

[soft sparrow]

i'm not from Rio 😭

[vapid ibex]

The what?

[soft sparrow]

come back when you have an AI generated degree

[vapid ibex]

They hired Mr. Fantastic to design an MPU

[vocal steppe]

momentum labs 🙏

they're gonna us from the AI winter

[vapid ibex]

They gonna what us?

[vocal steppe]

sorry my bot's MPUs melted

[vapid ibex]

I can't believe you'll need to spend momentillion of dollars to get a new one

[chrome siren]

jeez how much are these?

also what’s the TDP? 💀

[vapid ibex]

More than $2 000 000 it seems, TDP is not a problem with that cost

[west jacinth]

*to buy

[chrome siren]

cheap.

they need to make a pcie card version of this 💀

[west jacinth]

or like $100k/month to rent

[west jacinth]

its 20cm squared

and draws 18kW

[chrome siren]

just need a thick case

[vapid ibex]

Renting is like online? Then again, final user does not care for TDP

[chrome siren]

damn the water heater for my shower draws about that much max 💀

[vapid ibex]

You can heat water AND host Momentum model

[chrome siren]

though 10kW is enough to shower if you don’t want to burn yourself

and this is instant, no storage tank

[chrome siren]

I really hope their data centers provide district heating 💀

[magic stream]

owned? they could get dedicated hosting.

Support for custom model weights​
https://www.cerebras.ai/pricing

they could negociate their terms

get owned

[magic stream]

Not identical
that is more than enough to not be compatible + gpt? that shit makes up the most, use claude

[magic stream]

attacking the model? were are attacking an alleged scammer

[magic stream]

so is deepseek at cheaper price

[magic stream]

source? show me their benchmark.

[magic stream]

drop their slop merges? wew.

[magic stream]

they have not released traditional benchmark to back themselves up.

[magic stream]

nobody is stopped you from getting ripped off, don't drag ppl with yourself

https://tenor.com/view/hamster-tussiniiduk32-gif-27701794

[magic stream]

nvm them, they can't consent and their opinion don't matter.

gn

[swift mica]

Glory to Arstotzka

[undone glade]

Idk if i would call it a scam in the buisiness sense though, they are effectively delivering what they promise, but lying about the method used to deliver

[swift mica]

still a scam. People choose their service under false promises

[vapid ibex]

False advertising at best

[cinder halo]

^^^ speaking of false advertising

[cinder halo]

(There was a scam message above mine for reference)

[hasty socket]

I love when people are kind enough to use the appropriate channels

[hasty socket]

I know it's kicking a dead horse, but

It's actually impressive to do so poorly

And that is using a solved, irrelevant benchmark

[swift mica]

Honestly I can't believe momentum isn't the worst thing to happen this month

[hasty socket]

Huh, what is?

That sentence has a couple interpretations and Idk which one is correct lol

[swift mica]

Both of those interpretations are sherlock

[hasty socket]

Good point

[foggy kiln]

Hello 👋

How’s the bees doing

🐝

[magic stream]

when you 'supposedly' merge slop models, but still get scores lower than glm 4.5 air.

https://tenor.com/view/its-just-that-shrimple-so-true-so-based-tsukasa-touhou-fox-gif-16244985312236617326

[hot gust]

i kinda miss the banter and laughs here

[soft sparrow]

i don't think they've ever released the videos they promised

[magic stream]

doesn't matter when their reputation is amazing, and it is so great even in their own benchmarks. makes me doubt if the benchmark itself is maxxed.

btw, did i mention it is $1 in and out for such amazing model?

[vapid ibex]

We are basically losing money by not using it

[storm vortex]

what are ya'lls favorite cli tools? claude code, codex, kimi cli?

[latent basin]

momentum cli

[magic stream]

agi

[magic stream]

aider / gemini cli

[hasty socket]

I can't lol, they recently asked nano-banana to turn the Anthropic logo into a tree and that's their new whole branding

[hasty socket]

Pinky promise I'll stop kicking this dead horse from now on

[vocal steppe]

Keep going

[hot gust]

daily momentum reminder

[undone glade]

🫡

never forget

[latent basin]

what we could’ve had if we had just realized this is agi running at 300k tokens per second on brand new computing chips

at such a low price that they actually pay you to use it

[tawny warren]

they redesigned their website

and a beautiful new model called tensor

[soft sparrow]

does it use Tensor Processing Units?

wait...

[hasty socket]

👀

I wonder if they'll try to advertise this anywhere

[tawny warren]

and some benchmarks

[hasty socket]

HumanEval in 2025

[soft sparrow]

human eval

lol

[hasty socket]

Are these model scores even correct

I'd expect it to have nearly saturated

[tawny warren]

i have no clue why they're listing themselves for benchmarks they're not even good at, like GSM8K

like terrible at

[hasty socket]

https://evalplus.github.io/leaderboard.html Not 100% this is the right thing, but

https://crfm.stanford.edu/helm/mmlu/latest/

No methodology info

Yawnnnnnnn

[vapid ibex]

So they have 2 models now. One is flagship model, other is another flagship model

[swift mica]

And when they release a 3rd one they'll have three flagship models

You know the other companies are just stupid

They remove the title of "flagship" from old products when they release new ones

They gotta let them accumulate

[foggy kiln]

😭

[latent basin]

really??

[latent basin]

I just spent $3000 of company money on momentum API credits 😬

[hasty socket]

There are people who would disagree

pls halp your company site http://127.0.0.1:3000/ doesn't open for me

[latent basin]

b-but momentum is supposed to be agi for the low low cost of free 😭 😭 😭 😭

[latent basin]

did you try firefox instead of chrome

[hasty socket]

Works now, ty

[latent basin]

no problem 👍

[foggy kiln]

The model actually seems clever

[hasty socket]

Told you

[foggy kiln]

i made a spacex rocket clone using tensor and landed on mars yday

i rest my case

👆

[west jacinth]

Link? Where can I try it?

[magic stream]

It's not smart, it basically said I'm the captain and I'm so safe (ignored your question entirely)

dumb sloppy response

[hot gust]

this thread is still alive 🥲

i prayed for times like these

[magic stream]

fishy scam company and the owner left but here you go:
#1434917422686801980 message

[atomic zinc]

https://tenor.com/view/i-used-to-pray-for-times-like-this-i-used-to-pray-meek-mill-dreams-and-nightmares-dreams-and-nightmares-michael-jordan-gif-4600240804856944817

Infinite Momentum

[foggy kiln]

daily reminder

[hasty socket]

https://tenor.com/view/mib-men-in-black-will-gif-20590125

[hot gust]

Daily pack smoke

[storm vortex]

Id hate for this thread to lose momentum

[near pumice]

god it was SO funny

[foggy kiln]

daily reminder

They launched tensor 1.5 guys!

https://tenor.com/view/girl-car-drive-thug-life-gif-15517794802653398546

[west jacinth]

lol

[foggy kiln]

Opus 4.5 level apparently

https://tenor.com/view/thats-tuff-solluminati-gif-12085257

[west jacinth]

Crazy

[foggy kiln]

Not gona say much but its really good

it's the only model that made me a proper working chess game playable

[west jacinth]

What are the api costs?

Hmm not available over Api it seems

[hasty socket]

Don't give your money to them

[west jacinth]

Yeah I know

Was just curious

[hasty socket]

Website says $1 / $1

[west jacinth]

Only for the momentum models

Docs only list momentum models

[hasty socket]

Hm, fair

[soft sparrow]

wow that may be AGI

[swift mica]

"Only model" definitely not. Gpt-5, 5.1, deepseek v3.2, opus 4.5 can all do that

Tensor 1.5 failed in my test actually

[foggy kiln]

What was ur test g

[soft sparrow]

make me a proper working chess game

[foggy kiln]

are you from egypt?

[soft sparrow]

clearly

[foggy kiln]

sad

[soft sparrow]

thanks

after all my name is the capital of egypt

so that must mean im from there

[swift mica]

Make a pygame app in a 1000x2000 window which is a chess clone for two players (turn-based, on the same screen). No external images or sounds. No sounds at all in the game. Before each turn, players roll a die, and whoever scores higher takes the turn. As such, in the game, it is possible that the same player moves multiple times in a row. Be very creative and give it your best. Draw the pieces with pygame drawing tools, as detailed as you can. The game is completely mobile, and so no keyboard inputs should be expected, only touch (Through mouse events, not finger events). Include all game mechanics of chess, not skipping anything.

Modified chess game, but includes a chess engine

[cinder halo]

is this just deepseek r1 0528 distil qwen3 8b running on cerebras hardware or something?

[hasty socket]

Seems to be a GLM 4.6 finetune or something like that

[zenith fern]

Mfw whole zk ecosystem is from Egypt 🥀

[soft sparrow]

what is zk 🥀

[gentle saffron]

Are you a Momentum sleeper agent?

[vapid ibex]

He is just a kid who got catfished by Hasan Momentum, and now he is too ashamed to admit it to himself and others

[zenith fern]

Niche community reference not hitting … twin I’m sad now 🥀

[soft sparrow]

fill me in 🥀

[coral patrol]

how good is this model? is it SOTA?

[hasty socket]

No

There are no credible benchmarks or decisive proof this isn't a wrapper of an existing LLM

Their own benchmarks place this model very poorly, but the results they compare to seem inaccurate and there's no methodology listed

There's no proof the company members exist, no proof their claimed MPU exists (they've been promising to reveal these for a while)

Ppl who claim to work for Momentum have affiliation with a past deceptive AI "company"

TL;DR do not engage

[vapid ibex]

They exist, 2 dudes brothers selling car parts or something. One of them created some blockchain cryptobro app or site And then suddenly they created the things companies with 100s of employers and huge budgets can't do in years

But the previous ai 'company' is real nail on the head

[hasty socket]

They've been promising for a while that the company members will join their server and introduce themselves lol

Well, to be specific, the two members other than the 2 known ones lol

[vapid ibex]

The Movementlabs guy was Hasan, there is no company members probably, only 2 of them

[zenith fern]

https://www.cairo-lang.org/

[soft sparrow]

oh thats literally me

[west jacinth]

MOMENTUM?!?!?!?

[soft sparrow]

NEW MOMENTUM MODEL????

[hasty socket]

Momentum is OpenAI???

[vapid ibex]

The what

[hot gust]

Back from the dead

[foggy kiln]

Tensor 1.5 context window is 1 million tokens guy

[gentle saffron]

Whats up with the reactions, sleeper agent?

[sharp zealot]

is this a meme

like this model i mean

[gentle saffron]

Yes

[atomic zinc]

lmfao, their Tensor 1.5 died(timeout) because I asked it an A5 problem 🤣

[vapid ibex]

One momentillion of tokens, impressive