PSEBench - Missing Information Detection: leaderboard

Metric: Missing-case detection F1 (%, M5): binary F1 of asking at least once, positive class the missing-information cases, 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
1Gemini 3.1 Pro (Preview)88.2
2Claude Opus 4.783.4
3GPT-5.582.2
4GPT-577.7
5Gemini 2.5 Flash70.5
6Claude Sonnet 4.670.1
7DeepSeek R166.5
8Qwen 3 235B A22B 2507 Instruct54.7
9Llama 3.1 8B Instruct46.7
10GPT-OSS-120B36.5
11GPT-5 Nano35.3
12Mistral Small 3.227.7

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