SMH-Bench (Direct Reasoning) - State Query: leaderboard
Metric: Success rate (%) on Environment-Grounded Query (TC7, 100 non-mutating inspection and aggregation questions, GPT-5 judged); 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: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 | 95 |
| 2 | Gemini 3.1 Pro (Preview) | 94 |
| 3 | GPT-5.4 Mini | 89 |
| 4 | DeepSeek V3.2 (Thinking) | 89 |
| 5 | Claude Sonnet 4.6 | 85 |
| 6 | Qwen 3.5 397B A17B | 84 |
| 7 | Claude Haiku 4.5 | 79 |
| 8 | GLM-5 | 78 |
| 9 | Qwen 3.5 9B | 77 |
| 10 | Qwen 3.5 4B | 75 |
| 11 | Qwen 3.5 Plus | 69 |
| 12 | MiniMax-M2.7 | 63 |
Interactive version: theaggregate.ai/benchmark?slug=smh-bench-direct-reasoning-state-query · How It Works · Data refreshed daily, snapshot 2026-09-29.