SMH-Bench (Interactive Agent) - Compositional Control: leaderboard

Metric: Success rate (%) on Compositional Control (TC2, 250 multi-device, state-dependent, room-dependent and selection-based operations); 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: Around December 2026. 13 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)84.4
2Claude Sonnet 4.681.2
3GPT-5.478
4Claude Haiku 4.573.6
5Qwen 3.5 397B A17B68.8
6GPT-5.4 Mini68.8
7GLM-565.6
8Qwen 3.5 Plus57.2
9Qwen 3.5 9B56
10DeepSeek V3.2 (Thinking)52.8
11Qwen 3.5 4B45.6
12MiniMax-M2.723.6

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