MRCR v2 (8-needle, Context Arena) - 1M pointwise
General QA
Context Arena's mean MRCR v2 8-needle string similarity over the 310 tests in its 1M bin (bin key 1,048,576 tokens), the length of Google's 1M pointwise MRCR v2 metric. Context Arena reports it only for models whose window holds these prompts, mostly Gemini and Grok, so it stays a component of the 128K AUC family.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3.7 Flash | 63.5 |
| Gemini 3.6 Flash | 58.6 |
| Gemini 3.5 Flash | 26.0 |
| Gemini 3.1 Pro Preview | 25.9 |
| Gemini 3 Flash Preview | 25.6 |
| Gemini 3.5 Flash-Lite | 22.3 |
| Gemini 3.1 Flash-Lite Preview | 12.4 |
| Grok 4.20 | 9.3 |