MHG-Mem Study - MMA - Mem-Gallery Judge: leaderboard
MHG-Mem Study:fixed MMA memory system, three backbones, same environment, mean of 3 runs, temperature 0 and 65,536 token limit; GPT-5.4-mini judge, Qwen served through vLLM on four A800 GPUs. MemEye MCQ exact match, open-answer BLEU/Judge, other benchmarks F1. WorldMemArena-S is a subset; memory systems remain separate protocols. Score (%).
Metric: Score (%). Source: arxiv.org. Saturation forecast: Around January 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.8 27B | 69.32 | #118 |
| 2 | GPT-4.1 Mini | 60.87 | #351 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mhg-mem-study-mma-mem-gallery-judge · How It Works · Data refreshed daily, snapshot 2026-10-23.