ESF-Bench - Multi-Source: leaderboard

Metric: Tagged slot accuracy (%; slots tagged with a MS (Multi-Source) scenario, multi-hop inference across persona, knowledge base, session and form sources, including conflicts; 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
1Qwen 3 32B (Thinking)87.3
2Gemini 2.5 Flash86.9
3GPT-5.1 (High)84.1
4GPT-OSS-120B (High)80.8
5Ministral-3-14B-Reasoning-251278.4
6GPT-OSS-120B (Low)75.9
7GPT-5.1 (Non-reasoning)73.1
8Gemini 2.5 Flash (Non-reasoning)71.9
9Ministral-3-14B-Instruct-251266.9
10Qwen 3 32B (Non-reasoning)66.1

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