SMH-Bench (Direct Reasoning) - Ambiguous Intent: leaderboard
SMH-Bench (2026) category on vague complaints and underspecified requests, where the agent must act sensibly or ask for clarification, scored in the executable HomeEnv smart-home simulator. In the Direct Reasoning setting the model sees the full serialized home state and answers once with structured actions that are replayed in the simulator. The board is the success rate in percent; higher is better.
Metric: Success rate (%) on Ambiguous Intent (TC3, 150 subjective complaints, scene-level goals and slot-missing requests, acting or asking for clarification as appropriate); Direct Reasoning: the full serialized home state (rooms, devices, current values, service schemas) and the request in one prompt, one structured JSON answer replayed in the HomeEnv simulator; final home state checked against the goal with untouched devices preserved; clarification and query answers judged by GPT-5; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 78 | #123 |
| 2 | Qwen 3.5 397B A17B | 76.7 | #246 |
| 3 | Gemini 3.1 Pro (Preview) | 72.7 | #94 |
| 4 | Claude Haiku 4.5 | 72.7 | #568 |
| 5 | DeepSeek V3.2 (Thinking) | 65.3 | #406 (DeepSeek V3.2) |
| 6 | Claude Sonnet 4.6 | 61.3 | #161 |
| 7 | GPT-5.4 Mini | 61.3 | #389 |
| 8 | MiniMax-M2.7 | 60.7 | #518 |
| 9 | Qwen 3.5 Plus | 51.3 | #232 |
| 10 | GLM-5 | 49.3 | #260 |
| 11 | Qwen 3.5 9B | 47.3 | #803 |
| 12 | Qwen 3.5 4B | 44.7 | #1175 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=smh-bench-direct-reasoning-ambiguous-intent · How It Works · Data refreshed daily, snapshot 2026-11-13.