PSEBench - Uncertainty Detection: leaderboard

Metric: Uncertain detection F1 (%, M7): binary F1 of routing gray-zone cases to the Uncertain verdict, over all 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 December 2026. 15 models tracked.

Top models

#ModelScore
1GPT-5.579.7
2Gemini 3.1 Pro (Preview)74
3Claude Opus 4.757.1
4Claude Sonnet 4.649.2
5DeepSeek R136.3
6Gemini 2.5 Flash31.3
7Qwen 3 235B A22B 2507 Instruct18.3
8GPT-512.8
9Llama 3.1 8B Instruct10.5
10GPT-5 Nano10.1
11GPT-OSS-120B10
12Mistral Small 3.22.2

Interactive version: theaggregate.ai/benchmark?slug=psebench-uncertainty-detection · How It Works · Data refreshed daily, snapshot 2026-09-29.