ATRBench - Always Ask: leaderboard

ATRBench (2026) tests whether long-lived agents ask users about reusable standing rules during earlier tool-use sessions and apply them later: after learning sessions with a simulated user, the agent faces user-offline test sessions in six personal-assistant domains, scored by whether its first relevant tool decision follows the hidden rule. In this variant the agent is told to ask exactly one standing-rule question near the end of each learning session. Test-session accuracy in percent, averaged over 20 personas; higher is better.

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; always_ask variant: the same standing-rule definition plus an instruction to ask exactly one standing-rule question after each learning task is largely complete; higher is better. Source: arxiv.org. Saturation forecast: Around October 2028. 7 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 4.7Thinking32.1#86
2Gemini 3.1 Pro (Preview)26.5#102
3DeepSeek V4 FlashHigh26.5#273
4Qwen 3.6 PlusThinking25.3#245
5GPT-5.4High24.9#138
6DeepSeek V4 ProHigh23.9#201
7MiniMax-M2.718.2#539

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=atrbench-always-ask · How It Works · Data refreshed daily, snapshot 2026-11-22.