A 4B model that clicks the screen and also codes
Microsoft published Mage-VL-Cowork, a 4-billion-parameter agent that mixes on-screen actions with a terminal, under Apache 2.0. It reports 66.0% on OSWorld-Verified with a 50-step budget. That figure is theirs. We did not sit it at a laptop.
- Verdict
- Watch Promising, but something is missing: access, pricing, or proof.
- For
- researchers comparing small computer-use models
- Why it matters
- Microsoft published a 4-billion-parameter model under Apache 2.0. The 66.0% OSWorld figure is the company's.
- Next check
- Recheck before you treat that score as something measured on your own machine.
- Published
- Filed
- By
- The Loaded Tokens desk
- Topic
- Agents
- Sources
- Microsoft Mage
Sources & limits: This is our recap, not a copy of the original. The links above are the pages we read. We didn't test it ourselves.
Corrections: if a fact is wrong, email hello@loadedtokens.com.