MRCR 1M
We show this table for reference; we do not rank on it.
A million-token MRCR long-context retrieval benchmark reported in DeepSeek-V4 model evaluations.
Benchmark score on MRCR 1M — October 29, 2026
We compile the MRCR 1M rows from provider self-reports. DeepSeek V4 Pro 0813 leads the table at 83.5%, followed by DeepSeek V4 Pro (High) (83.3%) and DeepSeek V4 Flash 0731 (78.7%). We do not use these results to rank models overall.
DeepSeek V4 Pro 0813
DeepSeek
DeepSeek V4 Pro (High)
DeepSeek
DeepSeek V4 Flash 0731
DeepSeek
8 modelsReasoningCurrentDisplay onlyUpdated October 29, 2026
Benchmark score table (8 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | DeepSeek V4 Pro 0813DeepSeek | 83.5% | Not reported | Open |
| 2 | DeepSeek V4 Pro (High)DeepSeek | 83.3% | Not reported | Open |
| 3 | DeepSeek V4 Flash 0731DeepSeek | 78.7% | Not reported | Open |
| 4 | DeepSeek V4 Flash (High)DeepSeek | 76.9% | Not reported | Open |
| 5 | Muse Spark 1.1Meta | 54.1% | Not reported | Closed |
| 6 | DeepSeek V4 ProDeepSeek | 44.7% | Not reported | Open |
| 7 | DeepSeek V4 FlashDeepSeek | 37.5% | Not reported | Open |
| 8 | Gemini 3.5 FlashGoogle | 26.6% | Not reported | Closed |
Among the reported MRCR 1M rows, DeepSeek V4 Pro 0813 is first at 83.5%. The third row is 4.8 points behind. The broader top-10 range is 56.9 points, so the table still separates the published systems.
8 models have been evaluated on MRCR 1M. The benchmark falls in the Reasoning category. MRCR 1M is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About MRCR 1M
Year
2026
Tasks
Million-token retrieval
Format
Long-context retrieval MMR
Difficulty
Million-token long context
BenchLM stores this DeepSeek-reported MRCR 1M value as a display-only row distinct from the existing MRCRv2 keys.
Freshness and provenance
Version
MRCR 1M 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does MRCR 1M measure?
A million-token MRCR long-context retrieval benchmark reported in DeepSeek-V4 model evaluations.
Which model scores highest on MRCR 1M?
DeepSeek V4 Pro 0813 by DeepSeek currently leads with a score of 83.5% on MRCR 1M.
How many models are evaluated on MRCR 1M?
8 AI models have published results on MRCR 1M in the BenchLM catalog.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.