ImageJevBench v0.3 ranks 44 systems; Imajev-4B leads on capability
airesearch12 · x · 2026-10-07
ImageJevBench v0.3.0 is live with 2,441 items (713 real, 1,728 synthetic), including 300 public and 1,200 sealed questions; 46 systems measured, 44 ranked on the new scale. In the latest results Imajev-4B leads the capability-only ranking, while withy-1 (API) tops the composite score but is too slow to qualify for the Jev-class capability ranking — highlighting the speed-vs-capability tradeoff.
More from Models
- Running MiMo 2.6 Flash across an RTX 6000 and M5 laptop at 40 tokens/sec over 10 GbE — pcuenq · 2026-10-07
- Running MiMo 2.6 Flash across an RTX 6000 and M5 laptop at 40 tokens/sec over 10 GbE — pcuenq · 2026-10-07
- Hugging Face Introduces Falcon ASR Speech Recognition Model — Hugging Face Blog · 2026-10-07
- Strata's IQ3_S quant rated on par with K4_XL, suggesting near-lossless Q2.5 compression — lxfater · 2026-10-07
- Fixed token codes suffice: 1.7B LM trains without a trainable input embedding table — A. Bochkov · 2026-10-07
- Ruff author asks: could a frontier model actually design TypeScript, not just rewrite it? — charliermarsh · 2026-10-07