PSEBench - Evidence Citation: leaderboard
A PSEBench patient safety triage dimension: the micro F1 of the legal clause and guidance identifiers the model cites on correctly triaged cases, in percent; higher is better. PSEBench (2026) builds 5,074 verifiable cases from clause cards of the Minnesota 29 reportable adverse health events, and models may query a simulated information provider before answering.
Metric: Evidence citation micro F1 (%, M3) over canonical clause and guidance identifiers, on correctly triaged cases with a gold legal basis, on the 5,074-case MN29 benchmark (3,455 complete, 1,362 missing-information and 257 uncertain cases built from clause cards of the Minnesota 29 reportable adverse health events), agentic environment where the model may ASK a GPT-5.2 information provider within 10 turns and then answers with a verdict, clause, legal evidence and rationale; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 74.3 | #49 |
| 2 | Claude Opus 4.7 | 67.4 | #79 |
| 3 | GPT-5 | 66.4 | #167 |
| 4 | Claude Sonnet 4.6 | 65.8 | #161 |
| 5 | Gemini 3.1 Pro (Preview) | 65.5 | #94 |
| 6 | Gemini 2.5 Flash | 64.9 | #453 |
| 7 | Qwen 3 235B A22B 2507 Instruct | 60.6 | #619 |
| 8 | DeepSeek R1 | 56.3 | #473 |
| 9 | Mistral Small 3.2 | 49.4 | #1591 |
| 10 | GPT-OSS-120B | 45.4 | #700 |
| 11 | GPT-5 Nano | 41.1 | #889 |
| 12 | Llama 3.1 8B Instruct | 30.7 | #4246 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=psebench-evidence-citation · How It Works · Data refreshed daily, snapshot 2026-11-13.