340M GLiNER2.5-Decide hits 12/12 zero-shot on clinical tasks, CPU-only
MaziyarPanahi · x · 2026-09-29
Maziyar Panahi reports that fastinoAI's newly released GLiNER2.5-Decide (340M params), run zero-shot on a CPU with no fine-tuning, scored a perfect 12/12 on clinical assertion status and allergy documentation tasks — matching much larger models — and 36/48 on the full set. He plans a clinical fine-tune (shorthand, med changes, contradictions across notes) validated on an untouched test set and is asking the community for tips.
Related event: 340M open-source model aces 12 clinical tasks zero-shot(2 posts)→
More from Models
- Vibe Check: Sonnet 5.5 Is a More Capable Partner Than Sonnet 5, If You Keep a Hand on the Wheel — kieranklaassen · 2026-09-29
- Third-party benchmark shows Claude Sonnet 5.5 crushing GPT-6 sol, nearly matching GPT-6 Astra — rickasaurus · 2026-09-29
- Anthropic says Sonnet 5.5 at low/medium effort beats Sonnet 5's best scores at ~1/10 the cost — airesearch12 · 2026-09-29
- Power user burns 170M+ tokens a week on ChatGPT, credits GPT 5.5 for winning him over — ChrisUniverse · 2026-09-29
- Every's vibe check: Claude Sonnet 5.5 runs 30%+ faster, up to 30% cheaper than Sonnet 5 — danshipper · 2026-09-29
- User: Claude Opus 5.5 beats Fable 5.1 on speed and quota, new Fable likely near — haider1 · 2026-09-29