MRCR-v2 128k: leaderboard
MRCR-v2 long-context retrieval subset using a 128K context window with 8 needles, as reported in Qwen's Qwen3.7-Max launch post.
Metric: Score (self-reported). Source: github.com. Source: benchmarklist.com. Saturation forecast: Around August 2027. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.7 Flash | 97 | #23 |
| 2 | GPT-5.6 Terra | 93.5 | #41 |
| 3 | Gemini 3.6 Flash | 91.8 | #51 |
| 4 | Qwen 3.7 Plus | 91.7 | #92 |
| 5 | Qwen 3.7 Max | 90.4 | #75 |
| 6 | Qwen 3.6 Plus | 85.9 | #120 |
| 7 | Gemini 3.1 Pro (Preview) | 84.9 | #58 |
| 8 | Claude Opus 4.6 (Max) | 84 | #63 (Claude Opus 4.6) |
| 9 | Claude Sonnet 5 | 81.5 | #57 |
| 10 | Grok 4.5 | 81.4 | #48 |
| 11 | Gemini 3.5 Flash | 77.3 | #60 |
| 12 | GPT-5.6 Luna | 74.8 | #68 |
| 13 | DeepSeek V4 Pro (Max) | 74.4 | #102 (DeepSeek V4 Pro) |
| 14 | Gemma 4 31B | 66.4 | #200 |
| 15 | Kimi K2.6 (Thinking) | 63.1 | #104 (Kimi K2.6) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mrcr-v2-128k · How It Works · Data refreshed daily, snapshot 2026-10-20.