HoosierHelp - Tool Exact Match: leaderboard
Part of HoosierHelp (Li et al., 2026): the percentage of 240 simulated social-service conversations in which the agent issues search calls whose normalized constraints exactly match the expected ones for every need. Higher is better; GPT-4.1-mini plays the users.
Metric: Tool exact match (%; conversations in which every expected need has a matching normalized search_resources call with the full structured constraints; 240 simulated social-service conversations over 3,971 Indiana public resources, balanced over single and composite needs, direct-match and fallback-required constraints and five simulated-user behaviours; GPT-4.1-mini simulated users, at most eight interaction turns). Source: arxiv.org. Saturation forecast: Around April 2028. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 38.75 | #123 |
| 2 | Qwen 3.6 27B | 36.67 | #341 |
| 3 | Qwen 3.7 Max | 36.67 | #145 |
| 4 | GPT-OSS-120B | 34.58 | #700 |
| 5 | DeepSeek V4 Flash | 30 | #258 |
| 6 | GPT-4.1 Mini | 25 | #779 |
| 7 | Qwen 3.6 35B A3B | 23.75 | #462 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=hoosierhelp-tool-exact-match · How It Works · Data refreshed daily, snapshot 2026-11-13.