PSEBench - Boundary Condition Hit Rate: leaderboard
A PSEBench patient safety triage dimension: the share of the boundary conditions of each case that the rationale of the model invokes consistently, judged by GPT-5.2, 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: Boundary condition hit rate (%, M4): share of the case boundary conditions that the rationale invokes consistently, judged by GPT-5.2, pooled over correctly triaged cases, 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 | 88 | #49 |
| 2 | Claude Sonnet 4.6 | 85.8 | #161 |
| 3 | GPT-5 | 83.9 | #167 |
| 4 | Qwen 3 235B A22B 2507 Instruct | 81.6 | #619 |
| 5 | Gemini 3.1 Pro (Preview) | 78.6 | #94 |
| 6 | Claude Opus 4.7 | 78.5 | #79 |
| 7 | Gemini 2.5 Flash | 77.5 | #453 |
| 8 | DeepSeek R1 | 76.5 | #473 |
| 9 | GPT-5 Nano | 72.5 | #889 |
| 10 | Mistral Small 3.2 | 71.4 | #1591 |
| 11 | GPT-OSS-120B | 68.5 | #700 |
| 12 | Llama 3.1 8B Instruct | 62.1 | #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-boundary-condition-hit-rate · How It Works · Data refreshed daily, snapshot 2026-11-13.