Context Arena MRCR (4-needle): leaderboard
OpenAI MRCR benchmark: finds and distinguishes between 4 identical pieces of information in long conversations (8K–1M tokens). Harder variant requiring finer positional discrimination.
Metric: AUC@1M (%). Source: contextarena.ai. Status: saturation imminent. 50 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash (Preview) (High) | 68 |
| 2 | Gemini 3 Flash (Preview) (Medium) | 64.5 |
| 3 | Gemini 3.1 Pro (Preview) (Low) | 63.6 |
| 4 | Gemini 3.1 Pro (Preview) (Medium) | 62.8 |
| 5 | Gemini 3.1 Pro (Preview) (High) | 61 |
| 6 | Gemini 3 Pro (Preview) | 57.3 |
| 7 | Gemini 3 Flash (Preview) | 50.4 |
| 8 | Gemini 2.5 Flash (Preview 05-20) | 49.5 |
| 9 | nova-2-lite-v1 | 47.5 |
| 10 | GPT-4.1 | 32.8 |
| 11 | GPT-4.1 Mini | 27.8 |
| 12 | Grok 4.1 Fast | 25.9 |
| 13 | GPT-5.2 (xHigh) | 22.6 |
| 14 | GPT-5.2 | 21.8 |
| 15 | Gemini 2.0 Flash (001) | 19.3 |
Interactive version: theaggregate.ai/benchmark?slug=context-arena-mrcr-4-needle · How It Works · Data refreshed daily, snapshot 2026-09-05.