Skip to content

MRCR 1M

A million-token MRCR long-context retrieval benchmark reported in DeepSeek-V4 model evaluations.

4 models

Higher is better.

  1. 1DeepSeek V4 Pro 0813DeepSeek83.5
  2. 2DeepSeek V4 Flash 0731DeepSeek78.7
  3. 3Muse Spark 1.1Meta54.1
  4. 4Gemini 3.5 FlashGoogle26.6
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Long-context retrieval MMR
Tasks
Million-token retrieval
Authors
DeepSeek-AI (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.