Context Arena MRCR (8-needle): leaderboard
OpenAI MRCR benchmark: finds and distinguishes between 8 identical pieces of information in long conversations (8K–1M tokens). Hardest variant, top models score under 50% AUC@1M.
Metric: AUC@1M (%). Source: contextarena.ai. Saturation forecast: Around November 2027. 52 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) (High) | 49.4 | #82 (Gemini 3 Flash (Preview)) |
| 2 | Gemini 3.1 Pro (Preview) (Medium) | 48.5 | #58 (Gemini 3.1 Pro (Preview)) |
| 3 | Gemini 3.1 Pro (Preview) (High) | 47.5 | #58 (Gemini 3.1 Pro (Preview)) |
| 4 | Gemini 3.1 Pro (Preview) (Low) | 47 | #58 (Gemini 3.1 Pro (Preview)) |
| 5 | Gemini 3 Flash (Preview) (Medium) | 45.9 | #82 (Gemini 3 Flash (Preview)) |
| 6 | Gemini 3 Flash (Preview) (Low) | 40.9 | #82 (Gemini 3 Flash (Preview)) |
| 7 | Gemini 3 Pro (Preview) | 39 | #66 |
| 8 | Gemini 3 Flash (Preview) | 29.3 | #82 |
| 9 | Gemini 2.5 Pro (Preview 06-05) | 28 | #106 |
| 10 | Gemini 2.5 Flash (Preview 05-20) | 27 | #216 |
| 11 | Nova 2.0 Lite (Non-reasoning) | 25.3 | #417 (Nova 2.0 Lite) |
| 12 | GPT-5.2 (xHigh) | 19 | #110 (GPT-5.2) |
| 13 | GPT-4.1 | 17.5 | #248 |
| 14 | GPT-4.1 Mini | 16.7 | #355 |
| 15 | Grok 4.1 Fast | 16.5 | #217 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=context-arena-mrcr-8-needle · How It Works · Data refreshed daily, snapshot 2026-10-20.