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
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash | 64.2 |
| 2 | Gemini 2.5 Pro | 60.8 |
| 3 | Qwen 3.6 27B | 58.2 |
| 4 | GPT-5.4 | 56.8 |
| 5 | Qwen 3 VL 32B Instruct | 43.9 |
| 6 | GPT-5.4 Mini | 35.3 |
| 7 | InternVL3-38B | 31.8 |
| 8 | Qwen 3 VL 4B Instruct | 30.1 |
| 9 | Qwen 3 VL 8B Instruct | 25.4 |
| 10 | InternVL3-8B | 15.5 |
Interactive version: theaggregate.ai/benchmark?slug=emem-bench · How It Works · Data refreshed daily, snapshot 2026-09-26.