Reasoning · Benchmark profile
A million-token MRCR long-context retrieval benchmark reported in DeepSeek-V4 model evaluations.
Data verified 21 Jul 2026 · Methodology 1.6.0
Benchmark score on MRCR 1M
DeepSeek V4 Pro (Max) leads at 83.5%, followed by DeepSeek V4 Pro (High) (83.3%) and DeepSeek V4 Flash (Max) (78.7%).
Visual analysis
Switch between model placement, score distribution and descriptive provider averages. Every view uses the same sourced leaderboard.
8 ranked models · higher is better
| Model | Rank | Provider | Score |
|---|---|---|---|
| DeepSeek V4 Pro (Max) | #1 | DeepSeek | 83.5% |
| DeepSeek V4 Pro (High) | #2 | DeepSeek | 83.3% |
| DeepSeek V4 Flash (Max) | #3 | DeepSeek | 78.7% |
| DeepSeek V4 Flash (High) | #4 | DeepSeek | 76.9% |
| Muse Spark 1.1 | #5 | Meta | 54.1% |
| DeepSeek V4 Pro | #6 | DeepSeek | 44.7% |
| DeepSeek V4 Flash | #7 | DeepSeek | 37.5% |
| Gemini 3.5 Flash | #8 | 26.6% |
One best score per model · higher is better
| Rank | Model | Provider | License | Evidence use | Score |
|---|---|---|---|---|---|
| #1 | DeepSeek V4 Pro (Max) deepseek-v4-pro-max | DeepSeek | open | Estimated reference | 83.5% |
| #2 | DeepSeek V4 Pro (High) deepseek-v4-pro-high | DeepSeek | open | Estimated reference | 83.3% |
| #3 | DeepSeek V4 Flash (Max) deepseek-v4-flash-max | DeepSeek | open | Estimated reference | 78.7% |
| #4 | DeepSeek V4 Flash (High) deepseek-v4-flash-high | DeepSeek | open | Estimated reference | 76.9% |
| #5 | Muse Spark 1.1 muse-spark-1-1 | Meta | closed | Reference only | 54.1% |
| #6 | DeepSeek V4 Pro deepseek-v4-pro | DeepSeek | open | Reference only | 44.7% |
| #7 | DeepSeek V4 Flash deepseek-v4-flash | DeepSeek | open | Estimated reference | 37.5% |
| #8 | Gemini 3.5 Flash gemini-3.5-flash | closed | Estimated reference | 26.6% |
The top of this snapshot is led by DeepSeek V4 Pro (Max) at 83.5%; third place is 4.8 points behind. The top-8 spread is 56.9 points.
About MRCR 1M
A million-token MRCR long-context retrieval benchmark reported in DeepSeek-V4 model evaluations. Results stay tied to the exact model variant and evaluation system. Multiple systems for the same model use the best published score on this page; overall Lumina scoring uses the median of ranking-eligible rows.
Open benchmark source ↗FAQ
A million-token MRCR long-context retrieval benchmark reported in DeepSeek-V4 model evaluations.
DeepSeek V4 Pro (Max) by DeepSeek currently leads with 83.5%.
8 models in the LuminaBench cohort have a qualifying score on this benchmark.
Related
