ESF-Bench - Unexpected User Behavior: leaderboard

Metric: Tagged slot accuracy (%; slots tagged with a UUB (Unexpected User Behavior) scenario, uncertain, hypothetical, sarcastic, misspelled or irrelevant user statements; 810 synthetic multi-turn enterprise slot-filling samples (6,530 slots, 8 domains) whose prompts combine a schema, a user-assistant conversation and auxiliary sources; varied output formats mapped to a standard form before exact comparison with the ground truth; 32k maximum tokens, reasoning setting as printed per model). Source: arxiv.org. Saturation forecast: Around May 2028. 10 models tracked.

Top models

#ModelScore
1Gemini 2.5 Flash74.5
2Gemini 2.5 Flash (Non-reasoning)66.8
3GPT-5.1 (High)64.9
4GPT-5.1 (Non-reasoning)62.5
5Ministral-3-14B-Reasoning-251260.1
6GPT-OSS-120B (High)59.5
7Ministral-3-14B-Instruct-251258.2
8Qwen 3 32B (Thinking)57.9
9GPT-OSS-120B (Low)57.6
10Qwen 3 32B (Non-reasoning)50.9

Interactive version: theaggregate.ai/benchmark?slug=esf-bench-unexpected-user-behavior · How It Works · Data refreshed daily, snapshot 2026-09-29.