SMH-Bench (Direct Reasoning) - Compositional Control: leaderboard
SMH-Bench (2026) category on multi-device operations that depend on sensor states, rooms or selections, 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 Compositional Control (TC2, 250 multi-device, state-dependent, room-dependent and selection-based operations); 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 | Gemini 3.1 Pro (Preview) | 81.2 | #94 |
| 2 | GPT-5.4 | 78.8 | #123 |
| 3 | Claude Sonnet 4.6 | 73.2 | #161 |
| 4 | Claude Haiku 4.5 | 73.2 | #568 |
| 5 | Qwen 3.5 397B A17B | 72 | #246 |
| 6 | GPT-5.4 Mini | 66 | #389 |
| 7 | DeepSeek V3.2 (Thinking) | 60.8 | #406 (DeepSeek V3.2) |
| 8 | MiniMax-M2.7 | 55.6 | #518 |
| 9 | Qwen 3.5 Plus | 51.6 | #232 |
| 10 | GLM-5 | 50 | #260 |
| 11 | Qwen 3.5 9B | 38 | #803 |
| 12 | Qwen 3.5 4B | 35.6 | #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-compositional-control · How It Works · Data refreshed daily, snapshot 2026-11-13.