This Mac agent hits a goal for a fiftieth of a cent per step

Typesafe-computer-use drives a Mac toward a goal for 1/50th of a cent per step

A new open-source project, typesafe-computer-use, drives a Mac toward a plain-English goal for about $0.0002 per step by using OCR and a small classifier instead of sending screenshots to a frontier model. It matches Claude Opus 5's input tokens but costs 155x less per decision and runs 14–40x faster, with an honest caveat: reasoning the big model gets free must be rebuilt as deterministic state.

Frontier-model computer use is capable and expensive: every step ships a screenshot and waits several seconds for a plan. Most steps do not need a plan.
  1. jdkoeck

    I trailed off a few lines into the README. No human ever edited any of this. « LLM detected, project rejected ».

  2. mmastrac

    I'd be interested to see if using DiffusionGemma-as-Jev helps as you can feed the image directly into the model and it'll make decisions based on the image embeddings.

  3. jimmySixDOF

    not sure how this is innovative they show the System-1 model can play Doom right in the announcement [1] :

    >Doom

    >We love how this doomo doomonstrates real-time intelligence and what can be doone with code + AI. The engineer behind it was worried about making 10 queries a second (which ends up costing ~$7/hour), but the rest of us agreed that was lower than expected! This is so fun we intend to not only release an in-depth walkthrough, but also host some events to hack on this.

    [1] https://typesafe.ai/blog/introducing-system-one-models-and-j...

  4. john_minsk

    Super cool. Hope waitlist will move soon. I have a use case for it too.

    are you the author? If so - what are your notes on using Jev in this scenario?

  5. Zaraif13

    How does it do on OSWorld-verified? Recently read that even Fable 5 is just at 85% .

More from this day

2026-09-19