DeepSeek MRCR 1M (BenchLM): leaderboard
Retrieval from a million-token context, as reported in DeepSeek’s technical evaluation and mirrored by BenchLM. Uses the source’s MMR score and remains distinct from MRCR v2 task configurations. These are published results, not BenchLM ability estimates.
Metric: MMR (%). Source: benchlm.ai. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Pro (0813) | 83.5 | #58 |
| 2 | DeepSeek V4 Flash (0731) | 78.7 | #84 |
| 3 | Muse Spark 1.1 | 54.1 | #55 |
| 4 | Gemini 3.5 Flash | 26.6 | #66 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=deepseek-mrcr-1m-benchlm · How It Works · Data refreshed daily, snapshot 2026-11-05.