ATRBench - Rule Guidance: leaderboard
Metric: ATRBench test-session accuracy TSAcc (%, 0-100): the share of user-offline rule-bound test sessions whose first relevant tool decision matches the gold action of the bound standing rule (a deterministic check on the trajectory), macro-averaged over 20 synthetic personas (284 rule and test-session pairs in six personal-assistant tool domains); learning sessions with a simulated user run first and their transcript is frozen as context for the test sessions; single trial, 20-turn cap, provider reasoning modes enabled; atr variant: the agent prompt adds a definition of a standing rule and a soft permission to ask about one, with a cost-benefit reminder, without disclosing any rule; higher is better. Source: arxiv.org. Saturation forecast: Around 2031. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 32 |
| 2 | Claude Opus 4.7 (Thinking) | 28.7 |
| 3 | DeepSeek V4 Flash (High) | 23.2 |
| 4 | MiniMax-M2.7 | 21.3 |
| 5 | DeepSeek V4 Pro (High) | 21.2 |
| 6 | Qwen 3.6 Plus (Thinking) | 19.4 |
| 7 | GPT-5.4 (High) | 13.3 |
Interactive version: theaggregate.ai/benchmark?slug=atrbench-rule-guidance · How It Works · Data refreshed daily, snapshot 2026-10-07.