Scott

From the infinite potential of energy to the total actualization of entropy, intelligence charts a course for the pursuit of meaning, mission and love.

[Scott @scottstts]

I wrote and directed a cutscene intro film for my game, used gemini tts for all the dialogue, and opus 5.5 created the scenes and animated it all in @threejs, including baked lighting inside the ship cabin using blender cycles. I'm really getting good at this

Sound on!

It's live now: https://t.co/Ax8USgFbxW

[Scott @scottstts]

I can physically see brain shrinking

[maeve ~ @miaaowing]

THE FUTURE IS HERE

[Scott @scottstts]

@X should investigate those participating in this, clearly a violation of ToS

[Elshayib @elshayib_]

That's how Kimi AI are promoting their new model Kimi K3.1 pro and I know all the influencers working with them and none of them are honest,I have posted about it myself but I didn't join this and never used their templates that they provide.

PROPAGANDA, I have more to share it's f**cked up I'm gonna have follow up for this

[Scott @scottstts]

It’s really just that simple, just throw a ton of agents at it

[Noam Brown @polynoamial]

Last month a large team of agents used over 100 billion tokens to resolve the Navier-Stokes Millennium Prize problem. Now we're using a similar approach for red-teaming to find and fix infra vulnerabilities.

[Scott @scottstts]

@Curline1222

[Scott @scottstts]

I’d like to see one thing Kimi k3.1 can do that opus 5.5 can’t

So far I haven’t seen it yet

[Scott @scottstts]

Maybe @polynoamial’s idea of agents using cpu temperature to communicate isn’t so crazy after all?

[roon @tszzl]

‘In one experiment, we asked agents in separate VMs, created and managed by OpenAI’s sandboxing software on shared host machines, to exchange synthetic secrets without the use of network connections or shared writable files—something that our systems are meant to prevent. The agents found a way to communicate using the processor’s cache’

[Scott @scottstts]

I still think about this. What happened to model training that made frontier AI suddenly very good at 3d?

Like what did they train the model on? How did they bootstrap it? I imagine before all the AI threejs demos and games there was very little high quality threejs work they could use to train the model

Is it an emergent capability from RL? Idk, it’s kinda mind boggling

[mrdoob @mrdoob]

This is how AI-generated 3D worlds looked last year.

[Scott @scottstts]

The only reason I can think of why they did this is because CCP demanded it

It must make them nervous too that ordinary people can do hacking with open models, diminishing their cyber control

[🥔🥔🥔 @argofowl]

kimi nerfed k3.1 pro's hacking skills on purpose before the open weights drop

they pulled the cyber data out of training, which took their own exploit bench score from 90.5% to 81%

even at 90.5% they say its hacking still trails frontier closed models that have been out for 3+ months

everything is getting hacked soon 🙂

[Scott @scottstts]

OpenAI is back in the game, can we meet that monstrosity that cracks frontier math problems every 3 hours?

[OpenAI @OpenAI]

Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.

On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.

While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.

After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.

https://t.co/EmEZsFRMcF

[Scott @scottstts]

I don’t think Kimi k3.1 is much of a threat to OpenAI or Anthropic at all

It’s good at 3d, and avg on other things

api pricing is $2 / $10, not cheap, bad subscription usage as well

On a Claude max you get the best model and 58x equivalent of api usage

Nobody in their right mind would trade in their Claude sub for a Kimi sub

Host it yourself? it’s a freakin 3T model, even hardware bros sweat over it

Now why would Anthropic or OpenAI be intimidated by that?

[Scott @scottstts]

@dummerspast39 Why was o3 benchmaxxed? I remember it to be a really good model

[Scott @scottstts]

From what I can see on X right now about Kimi k3

- very good at 3d, on par with opus 5.5 but not consistently better
- may be benchmaxxed on other domains
- sub usage is abysmal

I think that’s about it

[Scott @scottstts]

@IndependentEco Yeah I think so, and opus 5.5 wins on usage too

[Scott @scottstts]

@LLMJunky Totally

[Scott @scottstts]

All the subscription usages are bad except for Anthropic, I would never have believed that a year ago

[am.will @LLMJunky]

Do not buy the Kimi $19 Plan. In fact, I'm not sure its a good idea to buy any of their plans. As per usual, the limits are a complete JOKE.

I already hit my 5h limit, and the only lines of code it wrote were to scaffold the project. Not 1 meaningful line of code.

It wasn't even a fraction of the way through the research phase. It had looked at 1 screenshot.

For context, even the $5 Meta plan was able to build a functional Rocket League before using the limits. What a joke.

Just use the API.

[Scott @scottstts]

@iamAliAgha Although I’m not fully convinced here, I’m seeing tons of people reporting very bad Kimi sub usage, whereas Claude sub right now is great. That means opus 5.5 is both cheaper and better than k3.1 atm

[Scott @scottstts]

We now have an open model roughly in the league with opus 5.5 and Astra🫣

[Kimi.ai @Kimi_Moonshot]

Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.

🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.

Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on https://t.co/85jA71X9gw, Kimi Work, Kimi Code, and the Kimi API.

Open weights by Nov 4, 2026.

🔗 API: https://t.co/XCrgjXAqMw
🔗 Tech blog: https://t.co/5iPP0MBej6

[Scott @scottstts]

K3.1 probably mainly focused on 3d, impressive

[Scott @scottstts]

Kimi k3.1 doesn’t seem that impressive on web dev, it’s not even close to the top

[Arena.ai @arena]

Big news: @Kimi_Moonshot is back in the top-4 labs in the Code Arena: WebDev with the release of Kimi K3.1 Pro!

This model just landed #8 in the Code Arena: WebDev with 1692 pts, matching Claude Opus-5 (Max) at 70% lower cost. This is a 9 place jump (+37 points) from Kimi K3 (Max) at #17 with 1655 pts.

Kimi K3.1 Pro improved from Kimi K3 across all categories. Highlights:
- #7 in Brand & Marketing (up from #9)
- #9 in Simulations (up from #20)

Cost efficiency is another upgrade. Priced at $2/$10 per million input/output tokens, Kimi K3.1 Pro (Max) has a median cost per task of $1.23 (29% less than Kimi K3). See its placement compared to the Pareto frontier in thread.

With open weights coming soon, Kimi K3.1 Pro would be the new #1 open model in the Code Arena: WebDev, overtaking DeepSeek-v4.1-Flash. Stay tuned!

See @Kimi_Moonshot’s ranking by lab, and 3D generations from Kimi K3.1 Pro below in thread.

Huge congrats to the @Kimi_Moonshot team on this release!

[Scott @scottstts]

No way, even HF is censoring models? Are we gonna finally resort to downloading models from torrents?

[Nova @NovaXCode]

🚨 Uncensored models are quietly vanishing from Hugging Face.

​First “Kimi-K2.7-Code-abliterated”, now “MiMo-V2.6-Pro” and “penclaw-GLM-5.3” are hitting 404s mid-download.

​“Open weights” only stay open if you self-host. Grab your local quants before your favorite repo turns into a dead link.

[Scott @scottstts]

This is for sure a bug right? No way 5hr limit lasts 1 min?😂

[Mia @MiaAI_lab]

absolutely disgusting and shameful usage limits on Kimi K3.1 Pro. My 5-hour limit was DONE after exactly 1 min into the prompt, and nothing was built!

not only has it not improved, it's actually a lot worse than before. this is the WORST subscription of them all 👎

[Scott @scottstts]

OpenAI just bans your account for reverse engineering work marked as cyber abuse

If you do anything even remotely non vanilla you gotta use an open model

[log4j @log4j]

banned for reverse engineering work? that’s wild

oh well, -$500, and a bunch of banked resets) but gained motivation to get 3 more machines and finish my cluster.

i have never used openai to exploit or find vulnerabilities, only local reverse engineering which tripped daybreak.