EMem-Bench - Interaction Failure: leaderboard

Metric: Success rate (%; the 263 Interaction Failure episodes, avoiding a target that an earlier failed action showed to be unusable; 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 Flash60.5
2Gemini 2.5 Pro55.1
3Qwen 3.6 27B54.4
4GPT-5.451
5Qwen 3 VL 32B Instruct45.3
6InternVL3-38B29.3
7Qwen 3 VL 4B Instruct20.9
8GPT-5.4 Mini19
9Qwen 3 VL 8B Instruct11.8
10InternVL3-8B5.3

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