ClinSQL - SQL Score: leaderboard
ClinSQL's second metric: the same test-split queries (chain-of-thought prompting, up to two execution-error refinements) scored by a GPT-5 judge against an expert rubric tree for the SQL text itself, covering cohort construction, clinical concept and code mapping, joins and aggregation with critical-first aggregation. Percent, higher is better; Yale University and University of Washington, 2026.
Metric: Rubric SQL Score (%). Source: arxiv.org. Saturation forecast: Around June 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 47.48 | #452 |
| 2 | Gemini 2.5 Pro | 47.28 | #262 |
| 3 | GPT-4.1 | 46.23 | #492 |
| 4 | DeepSeek R1 0528 | 44.91 | #425 |
| 5 | DeepSeek V3.1 | 43.19 | #500 |
| 6 | Grok 4 Fast (Reasoning) | 42.78 | #438 (Grok 4 Fast) |
| 7 | GPT-5 Mini (2025-08-07) | 42.72 | #341 |
| 8 | O4 Mini (2025-04-16) | 41.23 | #362 |
| 9 | Grok 4 Fast (Non-reasoning) | 39.85 | #438 (Grok 4 Fast) |
| 10 | Qwen 3 Coder 480B A35B Instruct | 39.05 | #641 |
| 11 | Qwen 3 235B A22B 2507 FP8 (Thinking) | 38.2 | #431 |
| 12 | Qwen 3 Next 80B A3B (Thinking) | 38.13 | #591 |
| 13 | Qwen 3 235B A22B 2507 Instruct | 36.36 | #619 |
| 14 | GPT-5 Nano | 36.13 | #888 |
| 15 | Mistral Medium 3 | 35.33 | #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-sql-score · How It Works · Data refreshed daily, snapshot 2026-11-12.