ESF-Bench - Corrections: leaderboard

Metric: Tagged slot accuracy (%; slots tagged with a COR (Corrections) scenario, user corrections and relative or conditional updates of earlier values; 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 December 2026. 10 models tracked.

Top models

#ModelScore
1Gemini 2.5 Flash84.1
2Qwen 3 32B (Thinking)83.4
3GPT-5.1 (High)78.3
4Ministral-3-14B-Reasoning-251276.8
5GPT-OSS-120B (High)72.6
6GPT-OSS-120B (Low)70.7
7Gemini 2.5 Flash (Non-reasoning)68.1
8GPT-5.1 (Non-reasoning)67.8
9Qwen 3 32B (Non-reasoning)64.6
10Ministral-3-14B-Instruct-251259.6

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