Someone built JevBench, the first benchmark for 'Jev-class models'; original Jev leads at 75.3

airesearch12 · x · 2026-09-19

Riding the hype around JEV, the tiny local computer-use model, someone launched JevBench — billed as the first benchmark for 'Jev-class models.' The original Jev by @typesafeai leads at 75.3, with SemIf close behind at 74.6. Largely a community in-joke, it playfully catalogues the wave of small, fast, screenshot-free local agent models.

Original post →

More from Fun

Fun channel →