The First Fully General Computer Action Model

2/26/2026 at 12:09:04 AM

This seems like really great research, and the first time I’ve seen overwhelming praise on HN. Congrats!

I wanted to comment though that your title is not doing you any favors, and I suspect that is why this is not getting more traction (which it deserves). I fully expected some half baked GitHub repo, but instead found something truly awesome.

To use your own words, Neel, “ a very different type of computer use model” would have had me clicking faster. I’m not great at titles, however, and maybe there are better ideas out there.

Anyway, can’t wait to see how this develops! Especially looking forward to the CAD work.

by kylenessen

2/26/2026 at 12:42:45 AM

I think you guys are on the right track here. I’d love to learn more about the math behind the FDM. I don’t think folks realize how behind we are on vision, thank you for your work here.

by nextzck

2/23/2026 at 5:10:27 PM

Hey guys! I’m Neel, been holed up in our south park office for the past year working on model training. excited to share our research!

This is a preview of a very different type of computer use model—we train on the internet. Specifically we have 11 million hours of computer video stored on our storage cluster (previously shared https://news.ycombinator.com/item?id=45438496 !) and the model can work in 30 FPS. Since we match the fundamental form factor of computer-use, we can get our model to do CAD, browse websites, and even drive a car using arrow keys. I’m super excited to see what our model can do as we scale more, it's a fun frontier to work on (not language models :) ).

The team and I will be online responding to the comments, so drop any questions.

by nee1r

2/26/2026 at 12:25:56 AM

How do I access this? Any HF or API coming?

Any benchmark comparisons to Fara-7B or Sonnet 4.6, Qwen 3.5 etc.?

by ilaksh

2/26/2026 at 12:06:37 AM

Get ready for the acquisition offers.

by arkmm

2/25/2026 at 11:15:57 PM

11 million hours of data is a lot, did you have to synthesize it at all, or was it purely collected?

by dangoodmanUT

2/25/2026 at 11:14:58 PM

This looks like a really promising approach

In particular the Forward rollout module is very important. It aligns your (effectively) world model with what it expects from the world, and keeping those in sync I think gives this the power it needs to be able to generate the state action pairs to continuously train semi supervised

by AndrewKemendo

2/23/2026 at 5:13:42 PM

I rly liked the point about ctrl-c only being able to be labelled retrocausally. I do think that with enough past context you should be able to know what was copied - in some sense the past does encode the future - but also an agentic decision is precisely the kind where the future is more informative than the past for reconstructing that decision.

It does make me wonder if you should have the inverse dynamics model split into specifically retrocausal and causal. You kind of do this already with the inverse and forward dynamics model, but the idea of a model that knows only about the future training in a feedback loop with a model that knows only about the past is kind of interesting.

I think you could just do a clever masking regime in your diffusion model to achieve the same effect without a whole architecture change.

by clemvonstengel

2/23/2026 at 5:29:00 PM

yeah we actually had some wacky ideas with ctc + a reverse-causal mask but diffusion does just make it all a bit more simple

by g413n

2/25/2026 at 10:37:08 PM

At first glance, this looks incredible to me. The authors train one model on 40K hours of computer-use video, previously labeled by contractors with keyboard and mouse actions, then use that model, in effect, to label 11M hours of computer-use video, which they use to train the computer-action model. The key advance is in compression. Quoting from the OP:

> [previous models] burn a million tokens to understand just one minute of 30 FPS computer data. Our video encoder encodes nearly 2 hours of video in the same number of tokens—that’s 50x more token-efficient than the previous state-of-the-art and 100x more token-efficient than OpenAI’s encoder.

While I was already aware that there are people working on new, more efficient "world models," this is the first one I've seen in action. I'm a bit in shock at how good it is, quite frankly.

I've added the OP, as well as a related 2018 paper on Behavioral Cloning from Obervation (BCO) to my reading list.[a] So far, I've only skimmed the 2018 paper, but it's already evident that it's well-written. I'm no expert in deep RL, and I can understand it. BTW, "Behavioral Cloning from Obervation" is a really good name, with an easy-to-remember acronym.

Thank you for sharing this on HN.

[a] https://arxiv.org/abs/1805.01954

by cs702

2/24/2026 at 2:24:36 AM

This looks extremely impressive, really deserves more attention here.

Are the inverse dynamics and forward dynamics models trained separately? It sounds like if the inverse dynamics model is meant to extrapolate more training data, then perhaps all that means is it takes very little data to generalize directly with the forward dynamics model assuming the right architecture.

by alyxya

2/24/2026 at 5:48:29 PM

thanks! the inverse dynamics model is trained first on 40k hours of data and then frozen to label all 11 million hours. yup! the idea is that it should take a small amount of data to generalize environment dynamics, then you can use a lot of data to understand actions.

by nee1r

2/26/2026 at 12:24:41 AM

Looks extremely impressive! Genuine question - why are you sharing your methods openly? I am grateful for it, but just curious your motivations.

by ripped_britches

2/26/2026 at 12:34:45 AM

Really really cool. I appreciate the article style a lot too.

by user-

2/25/2026 at 11:45:36 PM

dammmmmmnnnn - lots to like here. I'm impressed with the 80,000 parallel website fuzzing desktops. And the 30hz (everything). Amazing.

by vessenes

2/26/2026 at 12:20:51 AM

Very impressive stuff!

Can you prompt it or is it strictly Copilot-style prediction?

by bananzamba

2/25/2026 at 11:42:57 PM

Nice that it can drive a car, but you could just use openpilot.

by LorenDB

2/25/2026 at 11:46:45 PM

Beware of ending up on the top page of "things HN didn't like" with such a comment (see post a few days ago)

by davidguetta

2/23/2026 at 6:42:29 PM

The video compression is very cool. And the small tricks like binning the mouse movements.

Wonder how much data is generalizable across different UIs? ie how good will the model be at using Figma if it’s never seen it before but has seen a lot of Photoshop

by aakashks

2/23/2026 at 6:45:35 PM

this is honestly an issue for the inverse dynamics (for app specific shortcuts etc.) but for general UI learning we still see promising eval trends

by nee1r

2/25/2026 at 10:27:09 PM

Just wanted to say: this is might impressive research.

Really interesting breakdown, proper nerdsniped into this, thanks for the refreshing AI news outside of language models :)

by piva00

2/25/2026 at 10:28:54 PM

May I suggest a driving demo in a parking lot with a mannequin instead of a real world video where it drives way too close to a pedestrian?

Otherwise, very cool and exciting!

by sp1nningaway

2/23/2026 at 5:06:21 PM

Curious about the masked diffusion IDM choice. They mention CTC loss and cross-entropy both underperformed — I'd love to see ablations on that. The claim that typos were "extremely common" with non-causal cross-entropy is interesting but hand-wavy without numbers.

by rio_popper

2/23/2026 at 5:15:55 PM

the main chain of experiments was trying causal => non-causal => non-causal with ctc and CE. i think a good intuition here is that you need a generative approach fundamentally because there definitely are multiple correct IDM labels.

by nee1r

2/23/2026 at 5:06:49 PM

The car thing is very impressive By the way, do you have plans to handle the computer’s audio output?

by ennucore

2/23/2026 at 5:12:47 PM

yeah we've done audio work in the past so we'll def merge the recipes at some point, long term should have full io that a human has (except maybe not generating video for video calls that seems a bit much)

by g413n

2/25/2026 at 10:28:08 PM

Can it defeat captchas?

by wasmainiac

2/23/2026 at 5:23:13 PM

What sort of fine tuning data was needed to allow the model to self-drive? One hour of video of someone driving, or extra labeling?

by ClaireBookworm

2/23/2026 at 5:27:11 PM

i actually drove the car (with arrow keys) around south park for around ~45 minutes as finetuning data, no extra labelling other than that. think the car line graph is super cool because you actually see the videegame prior working

by nee1r

2/23/2026 at 5:29:49 PM

relevant note is that we finetuned by having the human also use arrow keys which keeps it in-distribution but also slower to collect

by g413n

2/23/2026 at 5:34:15 PM

what tasks can the model do out of the box? was each of the examples a different fine tuned model?

by kdrag0n

2/23/2026 at 5:39:44 PM

it's a pretty general policy but this is all super early, it's great at exploring websites so fuzzing was easy, for CAD it has good enough base rates with the few-shot prompt when we do the repetitive stuff, and we gave it checkpoints on each step, the other stuff in the mosaic are just some of our favorite clips from internal evals

by g413n

2/25/2026 at 11:30:14 PM

Looks like it's playing the special stages from Knuckles' Chaotix?

by bitwize

2/23/2026 at 5:08:23 PM

How do you tokenize the mouse inputs?

by ennucore

2/23/2026 at 5:14:15 PM

good question! we use exponential binning (map the mouse movements onto a plane with exponentially increasing tick marks https://si.inc/fdm1/exponential_binning.webp) but tried a bunch of other methods (linear creates too many tokens for the model to learn well). Polar coordinates seem like a better solution but empirically didn't work well because the tokens got too coarse too fast.

by nee1r

2/23/2026 at 5:14:18 PM

we do exponential binning but fwiw I think we can do way better just hasn't been the main research area initially

by g413n

2/25/2026 at 10:01:24 PM

Amazing!

by Obscura-

2/24/2026 at 2:22:20 PM

holy crap, this is so good. How did it get buried?

by 152334H

2/25/2026 at 10:28:40 PM

Too technical for HN

by yoyohello13

2/24/2026 at 5:46:52 PM

real

by nee1r

2/25/2026 at 11:48:00 PM

Are you guys affiliated with Meta’s ex-CTO in any way? I remember he famously implied that LLMs hyped. The demos are very impressive. Does this use an attention based mechanism too? Just trying to understand (as a layman) how these models handle context and if long contexts lead to weaker results. Could be catastrophic in the real world!

by sheepscreek

2/25/2026 at 11:50:39 PM

I think in the long run, we may need something like a batch job that compresses context from the last N conversations (in LLMs) and applies that as an update to weights. A looser form of delayed automated reinforcement learning.

Or make something like LoRA mainstream for everyone (probably scales better for general use models shared by everyone).

by sheepscreek

2/25/2026 at 10:50:39 PM

My tech-informed but ML-ignorant take: This will soon be the biggest thing since ChatGPT.

by akoboldfrying

2/26/2026 at 12:32:12 AM

[dead]

by aplomb1026

2/23/2026 at 6:18:47 PM

[dead]

by snowhale

2/23/2026 at 6:31:14 PM

no finetuning data for the blender task! we actually think its the opposite, there are a lot of video tutorials for complex tasks like onshape/blender/fusion360 but not as much of people idly browsing.

but also at the 11M hour scales it still sees a substantial amount of data

by nee1r