LLM Stats (MRCR v2 (8-needle)): leaderboard
MRCR hides several identical writing requests inside a long synthetic conversation and asks the model to return one specific instance, such as the second poem on a topic in a named style; the v2 8-needle variant uses eight such requests and adds the style to the key. The score is an average text similarity to the target. LLM Stats republishes provider figures here, higher is better. Source: arxiv.org/abs/2409.12640.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Spark 1.3 | 98.5 | #24 |
| 2 | Gemini 3.7 Flash | 97 | #23 |
| 3 | Qwen 3.8 Max | 92.9 | #35 |
| 4 | GPT-5.6 Sol | 91.5 | #19 |
| 5 | GPT-5.6 Terra | 89.6 | #41 |
| 6 | Claude Opus 4.6 | 76 | #63 |
| 7 | GPT-5.5 | 74 | #31 |
| 8 | Gemma 4 31B | 66.4 | #200 |
| 9 | Gemini 3.1 Flash Lite | 60.1 | #191 |
| 10 | Gemini 3.6 Flash | 54 | #51 |
| 11 | Seed 2.0 Mini | 51.4 | #276 |
| 12 | Gemma 4 26B A4B | 44.1 | #358 |
| 13 | Gemma 4 12B | 43.4 | #535 |
| 14 | GPT-5.6 Luna | 41.3 | #68 |
| 15 | GPT-5.4 Mini | 33.6 | #212 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-mrcr-v2-8-needle · How It Works · Data refreshed daily, snapshot 2026-10-20.