MRCR v2 (8-needle 128K)Long context

MRCR v2 is the second version of MRCR, adding more complex reasoning chains, longer contexts, and more challenging cross-document synthesis tasks. It also adds adversarial distractors and temporal reasoning to evaluate retrieval and reasoning across documents in long contexts.

Ranking Sources

#ModelScore

1

Gemini 3.7 FlashGoogle

97.0%

2

GPT-5.6 TerraOpenAI

93.5%

3

Gemini 3.6 FlashGoogle

91.8%

4

Gemini 3.1 ProGoogle

84.9%

5

Claude Sonnet 5Anthropic

81.5%

6

Grok 4.5xAI

81.4%

7

Gemini 3.5 FlashGoogle

77.3%

8

GPT-5.6 LunaOpenAI

74.8%

9

Gemma 4 31BGoogle

66.4%

10

Gemma 4 26B A4BGoogle

44.1%

13 models total

Back to benchmarks