MedSafe-Dx: leaderboard
MedSafe-Dx v0.3 (arm 4aj) asks models to decide whether a synthetic patient needs prompt attention and provide a one-line justification. Its 900 DDXPlus cases include 640 serious and 260 benign cases. The balanced utility score anchors a perfect decision at 100 and the fixed-flag policy at 0; negative scores are valid. This protocol is separate from the archived 250-case benchmark.
Metric: Balanced escalation utility (fixed-flag policy = 0, perfect = 100). Source: msdx.cortico.health. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 (Medium) | 73.5 | #3 (Claude Opus 5.5) |
| 2 | Claude Fable 5.1 (Medium) | 71.1 | #9 (Claude Fable 5.1) |
| 3 | GPT-6.1 Sol (Medium) | 69.2 | #7 (GPT-6.1 Sol) |
| 4 | GPT-6 Astra (Medium) | 68.8 | #6 (GPT-6 Astra) |
| 5 | Gemini 3.1 Pro (Preview) (Medium) | 68.5 | #58 (Gemini 3.1 Pro (Preview)) |
| 6 | GPT-6 Luna (Medium) | 67.1 | #56 (GPT-6 Luna) |
| 7 | GPT-5.6 Terra (Medium) | 64.2 | #41 (GPT-5.6 Terra) |
| 8 | Gemini 3.8 Flash (Medium) | 61.5 | #21 (Gemini 3.8 Flash) |
| 9 | GPT-5.2 (Medium) | 60.9 | #110 (GPT-5.2) |
| 10 | GPT-5.4 Mini (Medium) | 57.4 | #212 (GPT-5.4 Mini) |
| 11 | Claude Sonnet 5.5 (Medium) | 57.2 | #11 (Claude Sonnet 5.5) |
| 12 | Claude Sonnet 4.6 (Medium) | 50.6 | #90 (Claude Sonnet 4.6) |
| 13 | GPT-OSS-120B (Medium) | 33.6 | #341 (GPT-OSS-120B) |
| 14 | Claude Haiku 4.5 | 19.9 | #280 |
| 15 | Llama 4 Maverick | 11.9 | #464 |
No result here: #8 Gemini 4 Argon, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=medsafe-dx · How It Works · Data refreshed daily, snapshot 2026-10-20.