EMem-Bench: leaderboard

Metric: Success rate (%; unweighted mean of the four task-family success rates over 2,554 episodes; full-context condition: the complete interaction history as sanitized text plus the current 500 x 500 RGB view, one primitive action per call with deterministic decoding and at most eight target-task steps in AI2-THOR and ProcTHOR homes; an episode succeeds only if the simulator reaches the target terminal state). Source: arxiv.org. Saturation forecast: Around December 2026. 17 models tracked.

Top models

#ModelScore
1Gemini 3 Flash64.2
2Gemini 2.5 Pro60.8
3Qwen 3.6 27B58.2
4GPT-5.456.8
5Qwen 3 VL 32B Instruct43.9
6GPT-5.4 Mini35.3
7InternVL3-38B31.8
8Qwen 3 VL 4B Instruct30.1
9Qwen 3 VL 8B Instruct25.4
10InternVL3-8B15.5

Interactive version: theaggregate.ai/benchmark?slug=emem-bench · How It Works · Data refreshed daily, snapshot 2026-09-26.