Jev 1.13 aces 153/153 real-length clinical notes, fastest top model at 345ms
MaziyarPanahi · x · 2026-10-13
On the 669-question Clinical System One benchmark, TypeSafe's Jev 1.13 scored a perfect 153/153 on real-length clinical notes — one of only two models to do so — tying #2 overall at 628/669 (93.9%) while being the fastest model above 90% at 345ms and $0.044/1K decisions. Leader pplx-decider v1.1 hit 96.1% at $0.025/1K; GPT-6 Luna 95.2%. The benchmark covers triage, notes, criteria, and ICD-10-CM, and tracks 'severe' errors like missed emergencies.
More from Models
- Meta's Muse Wonder model goes live on OpenRouter — jeff_weinstein · 2026-10-13
- Muse senior engineer benchmark results pulled over flawed run, retest underway — danshipper · 2026-10-13
- "Zero chance" a Meta model competes with Google, OpenAI or Anthropic right now — iruletheworldmo · 2026-10-13
- Dev turns EmbeddingGemma 2 into an on-device STT model, transcribing 1 hour of audio in 13 seconds — solyarisoftware · 2026-10-13
- Alexandr Wang hails Theo's model rankings: Opus 5.5 and Muse Wonder alone in S tier — alexandr_wang · 2026-10-13
- Reddit user rails against Gemini's decline: wrong bulb model, 13 failed script fixes — hotboxuzi · 2026-10-13