ClinSQL - Diagnostic Procedures: leaderboard
ClinSQL's Execution Score (a GPT-5 judge checks the executed result table against an expert rubric tree; percent, higher is better) restricted to the test tasks of its diagnostic procedures scenario: temporal sequencing of procedures and interventions within a stay. Queries come from chain-of-thought prompting with up to two execution-error refinements; Yale University and University of Washington, 2026.
Metric: Execution Score (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini (2025-08-07) | 75.64 | #341 |
| 2 | GPT-4.1 | 75.44 | #492 |
| 3 | Gemini 2.5 Pro | 72.36 | #262 |
| 4 | DeepSeek V3.1 | 71.46 | #500 |
| 5 | DeepSeek R1 0528 | 70.16 | #425 |
| 6 | Qwen 3 235B A22B 2507 Instruct | 69.55 | #619 |
| 7 | Gemini 2.5 Flash | 67.1 | #452 |
| 8 | O4 Mini (2025-04-16) | 66.04 | #362 |
| 9 | Qwen 3 Coder 480B A35B Instruct | 63.86 | #641 |
| 10 | Grok 4 Fast (Reasoning) | 63.81 | #438 (Grok 4 Fast) |
| 11 | GPT-5 Nano | 58.29 | #888 |
| 12 | Qwen 3 Next 80B A3B Instruct | 57.37 | #763 |
| 13 | Llama 4 Maverick Instruct | 51.65 | #1125 |
| 14 | Qwen 3 235B A22B 2507 FP8 (Thinking) | 47.17 | #431 |
| 15 | Mistral Medium 3 | 45.59 | #1079 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=clinsql-diagnostic-procedures · How It Works · Data refreshed daily, snapshot 2026-11-12.