ESF-Bench - Unexpected User Behavior: leaderboard
ESF-Bench (ServiceNow, 2026) slot accuracy on slots tagged with the Unexpected User Behavior scenario category: uncertain, hypothetical, sarcastic, misspelled or irrelevant user statements. In percent, higher is better.
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 June 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 74.5 | #453 |
| 2 | Gemini 2.5 Flash (Non-reasoning) | 66.8 | #453 (Gemini 2.5 Flash) |
| 3 | GPT-5.1 (High) | 64.9 | #229 (GPT-5.1) |
| 4 | GPT-5.1 (Non-reasoning) | 62.5 | #229 (GPT-5.1) |
| 5 | Ministral-3-14B-Reasoning-2512 | 60.1 | #1396 |
| 6 | GPT-OSS-120B (High) | 59.5 | #700 (GPT-OSS-120B) |
| 7 | Ministral-3-14B-Instruct-2512 | 58.2 | #1593 |
| 8 | Qwen 3 32B (Thinking) | 57.9 | #960 (Qwen 3 32B) |
| 9 | GPT-OSS-120B (Low) | 57.6 | #700 (GPT-OSS-120B) |
| 10 | Qwen 3 32B (Non-reasoning) | 50.9 | #960 (Qwen 3 32B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=esf-bench-unexpected-user-behavior · How It Works · Data refreshed daily, snapshot 2026-11-13.