EMem-Bench - Experience Generalization: leaderboard

Metric: Success rate (%; the 203 Experience Generalization episodes, applying a regularity learned from earlier corrections to a new object; 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 Flash75.4
2Gemini 2.5 Pro72.9
3GPT-5.469
4Qwen 3.6 27B68
5Qwen 3 VL 32B Instruct63.1
6GPT-5.4 Mini41.4
7InternVL3-38B33
8Qwen 3 VL 8B Instruct21.7
9Qwen 3 VL 4B Instruct15.8
10InternVL3-8B4.4

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