SMH-Bench (Interactive Agent) - Personalized Memory: leaderboard

Metric: Success rate (%) on Personalized Memory (TC6, 100 requests that need short- or long-term user preferences); 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)86
2Claude Sonnet 4.680
3Qwen 3.5 397B A17B77
4GPT-5.474
5DeepSeek V3.2 (Thinking)74
6Claude Haiku 4.573
7GPT-5.4 Mini69
8Qwen 3.5 9B65
9Qwen 3.5 Plus65
10Qwen 3.5 4B57
11GLM-551
12MiniMax-M2.727

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