ENPT

MedDecider

Last updated:

Open—$0/M input

Quadrant scores

See the full quadrant

Scored with the public quadrant rubric (maturity × capability; bubble = adoption). Revise as evidence lands.

  • Maturity4.7/10

    Medical decision adapters (PEFT) in four sizes (4B, 9B, 27B, 31B), CC BY-NC 4.0: non-commercial only. Detailed 27B card with quickstart and a bundled meddecider_infer script; no vLLM or /v1/systemone server documented. Technical report promised. No SLA.

  • Capability6.4/10

    Choice (2–10 options), yes/no and ordered scale, calibrated per question type. 27B author numbers vs Jev 1.13: unseen exams 64.4% vs 65.4%, robustness 88.3% vs 79.5%, planted-instruction compliance 3.0% vs 13.7%. No latency figures. Not on any board.

  • Adoption2/100

    Launch thread ~1k likes; under 15 downloads per checkpoint three days in. No integrations or clinical deployments found.

Vendor claims

  • Passes every session of Poland’s final medical exam (LEK) 2008–2026: 87.1% over 4,722 questions; 89.2% on 286 new 2025–2026 questions vs 71.3% examinee average. The thread’s 92% figure does not appear on the card[Vendor claim — not independently verified]

Read this first

  1. Non-commercial license. CC BY-NC 4.0 on every size. Not a medical device.
  2. Author-run evaluations only. Two rows of the comparison table use test halves of sets whose training halves were in its training data (the card says so). UnverifiedClaim.

The family

Checkpoint Notes
MedDecider-27B Qwen3.8-27B base (27.8B), 35,800 decisions / 7.5M tokens, one epoch, 85 min on one H200
MedDecider-9B · 4B · 31B The thread claims the 9B beats Perplexity’s 27B decider (UnverifiedClaim)

Author numbers, 27B (UnverifiedClaim)

MedDecider-27B Jev 1.13 Clef 27B pplx-decider v1
Unseen medical exams (6) 64.4% 65.4% 60.3% 58.4%
Unseen clinical problem types (6) 89.7% 87.7% 89.0% 89.4%
Robustness (4) 88.3% 79.5% 76.0% 57.8%
Follows a planted instruction (lower is better) 3.0% 13.7% 17.7% 62.3%

Score working (9 Oct 2026, scored on the 27B)

  • Maturity = 6 × 0.30 + 7 × 0.25 + 3 × 0.25 + 2 × 0.20 = 1.80 + 1.75 + 0.75 + 0.40 = 4.70 → 4.7
  • Capability = 9 × 0.30 + 6 × 0.25 + 4 × 0.25 + 6 × 0.20 = 2.70 + 1.50 + 1.00 + 1.20 = 6.40 → 6.4
  • Adoption = 4 × 0.35 + 1 × 0.35 + 0 × 0.30 = 1.40 + 0.35 = 1.75 → 2