FULL STORY
Jev: Small-Model Actions, Big-Model Thinking for Faster Agents
Open-source project Jev sparked attention by letting small models handle browser actions while large models do the deep thinking, claiming 10x agent speedups. Real-world tests later showed only about 20% gains.
2026-09-17 ~ 2026-09-18 · 2 episodes · 4 posts
Episode 1 · Open-source Jev: small model acts, big model thinks (2026-09-17, 2 posts)
Open-source project Jev splits browser agent work: a small model handles fast actions like clicking and scrolling while the big model focuses on reasoning, enabling 200ms decisions via page accessibility controls.
- Browser agent decides each click in 200ms with Jev, a $0.00002-per-call model — TheMoonMidas · 2026-09-17
- Jev: an open-source action-picker that splits agent thinking from clicking — alexcovo_eth · 2026-09-17
Episode 2 · jev Claims 10x Agent Speedup, Real-World Tests Show 24% (2026-09-18, 2 posts)
jev claims four techniques—model routing, computer use, and auto review—make agents 10x faster and 50% cheaper, but independent testing shows a 24% speedup, revealing a gap between marketing and reality.
- Making agents 10x faster and cheaper with jev: routing, review, and a $0.0039 browser demo — gabriel1 · 2026-09-18
- jev claims 10x faster, cheaper agents via routing, auto review and subagent orchestration — gabriel1 · 2026-09-18