Open-source 4B model tops JevBench on speed and cost, not intelligence

Open-source 4B model decider-4b v2 topped JevBench v1.4.2, beating Jev 1.13.0 and others. Its author clarified the win came from speed and cost, as Jev still leads on intelligence (53.1 vs 49.4); the benchmark will rotate hidden test sets and may raise the intelligence weight next version.

2026-09-25 ~ 2026-09-25 · 3 related posts

1 near-duplicate retellings: airesearch12