SMH-Bench (Interactive Agent) - State Query: leaderboard

Metric: Success rate (%) on Environment-Grounded Query (TC7, 100 non-mutating inspection and aggregation questions, GPT-5 judged); Environment-Interactive Agent: a ReAct loop that starts from partial observability and queries and controls the HomeEnv simulator through tools under a call budget; 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

#ModelScore
1DeepSeek V3.2 (Thinking)94
2Claude Sonnet 4.692
3Qwen 3.5 397B A17B92
4GPT-5.491
5MiniMax-M2.790
6Gemini 3.1 Pro (Preview)88
7Claude Haiku 4.587
8GLM-587
9Qwen 3.5 Plus87
10GPT-5.4 Mini86
11Qwen 3.5 9B81
12Qwen 3.5 4B76

Interactive version: theaggregate.ai/benchmark?slug=smh-bench-interactive-agent-state-query · How It Works · Data refreshed daily, snapshot 2026-09-29.