EvolMem - Inference: leaderboard
Long-context LLM backbones on multi-session conversational memory, paper v1 Table 2. Seven memory abilities are scored independently; the memory-system ablation table is a separate protocol. Mistral-Large has no pinned endpoint in the paper or released evaluator and retains an explicitly unspecified identity. This series reports Inference (%). Source: https://arxiv.org/html/2601.03543v1.
Metric: Inference (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 92.05 | #82 |
| 2 | GPT-5.1 | 72.38 | #145 |
| 3 | DeepSeek V3.2 | 60.84 | #210 |
| 4 | MiniMax-M2 | 59.38 | #327 |
| 5 | Llama 4 Maverick | 47.24 | #486 |
| 6 | Kimi K2 | 43.36 | #252 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=evolmem-inference · How It Works · Data refreshed daily, snapshot 2026-11-05.