wev
Local, open-weights decision model for browser agents: a drop-in for System One / Jev requests
Local System-One decision models: typed questions in, calibrated probabilities out. General decisions and browser-agent steps.
How it uses Jev
- 🌐 Browser steps on unseen websites: 76% right , where open decision models trained on general data reach at most 21%. - 🧠 General decisions stay strong. wev beats Laya on every general benchmark, and wev-8b is level with Kev-8B out of domain. - 🎓 As good as its teacher, locally. As the System One of an open browser agent, wev completes as many live tasks as the LLM it was distilled from. - ⚡ Fast. 10–37 ms for…
Highlights
- 🌐 Browser steps on unseen websites: 76% right , where open decision models trained on general data reach at
- 🧠 General decisions stay strong. wev beats Laya on every general benchmark, and wev-8b is level with Kev-8B out
- 🎓 As good as its teacher, locally. As the System One of an open browser agent, wev completes as many live
- ⚡ Fast. 10–37 ms for a general decision and 91–322 ms for a full browser page on one RTX 5090; 77 ms on a
- 📏 Calibrated. Probabilities you can threshold: raise the bar for DONE and early stops drop from 8.5% to 3.8%.
Getting started
m = wev.load("alanhuangya/wev-4b") downloads once from the Hugging Face Hub out = m.predict( state="Refund request: order 4411 arrived damaged, customer attached photos, first refund this year.", questions={ "action": {"type": "choice", "instructions": "What should support do?", "criteria": {"refund": "Refund the orde…
- Languages
- Python 53% · TeX 35% · BibTeX Style 7% · HTML 3%
- License
- Apache-2.0
- Last activity
- Sep 2026
- Added
- 2026-09-24
- Stars at snapshot
- 3
- Status
- Community listing
Source: community catalog. Details and metrics reflect the published community snapshot and are not a performance endorsement. Documentation details extracted from GitHub · the README.
