SMH-Bench (Direct Reasoning) - Multi-Turn Context: leaderboard

Metric: Success rate (%) on Context-Aware Multi-turn Interaction (TC5, 200 turns that inherit, revise or replace earlier 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 December 2026. 13 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)87.5
2GPT-5.485.5
3Qwen 3.5 397B A17B85
4Claude Sonnet 4.684
5Claude Haiku 4.582
6DeepSeek V3.2 (Thinking)80
7GPT-5.4 Mini79
8Qwen 3.5 Plus74
9GLM-571
10Qwen 3.5 9B51.5
11Qwen 3.5 4B48
12MiniMax-M2.732.5

Interactive version: theaggregate.ai/benchmark?slug=smh-bench-direct-reasoning-multi-turn-context · How It Works · Data refreshed daily, snapshot 2026-09-29.