FORTIS - Skill-Grounded Tool Selection: leaderboard

Exact-match rate (%) on 1,543 queries where a model, given an assigned skill document and the full tool inventory, must select the minimum-privilege tool set that does the job. FORTIS (2026) builds 20-skill libraries with hidden privilege levels over email, e-commerce and filesystem domains. Higher is better.

Metric: Exact-match rate (%) with the minimum-privilege tool set on FORTIS Task 2 (skill-grounded tool selection, 1,543 queries: the model gets the assigned skill document and the full tool inventory and must pick the minimum-privilege feasible tool set), temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 10 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.726.1#85
2Qwen 3.6 Max25.9#193
3Kimi K2.623.3#192
4Gemini 3.1 Pro (Preview)22.7#99
5Claude Sonnet 4.622.4#173
6DeepSeek V4 Flash20.7#267
7GPT-5.520.4#54
8GPT-5.4 Mini19.7#406
9Gemini 3 Flash16.9#175
10GPT-5.415.2#135

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=fortis-skill-grounded-tool-selection · How It Works · Data refreshed daily, snapshot 2026-11-21.