AA-LCR Extended (Mercor held-out)
Search
Mercor's held-out Extended twin of Artificial Analysis Long Context Reasoning: 89 Mercor long-context questions that require reasoning across several documents, scored by answer accuracy. The public twin is AA-LCR v1.1 (aa-lcr-1-1-artificial-analysis), which Mercor also reruns. Mercor does not publish the held-out tasks or say how they were drawn or written, so this is its own benchmark, not a component of the public one.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol Pro | 79.5 |
| GLM-5.3 | 76.8 |
| Gemini 4 Argon | 74.7 |
| GPT-5.6 Terra | 74.4 |
| Claude Opus 5.5 | 74.2 |
| Claude Opus 5 | 73.9 |
| Claude Sonnet 5.5 | 73.8 |
| Claude Fable 5.1 | 73.0 |
| GPT-5.6 Luna | 72.2 |
| Claude Fable 5 | 71.1 |
| DeepSeek V4.1 Flash | 70.8 |
| GPT-5.5 | 70.5 |
| GPT-6 Sol | 70.2 |
| Kimi K3 | 69.1 |
| GPT-5.4 | 68.3 |
| GPT-6 Luna | 68.3 |
| Grok 4.7 | 68.2 |
| GLM-5.3 Flash | 66.9 |
| Muse Spark 1.3 | 66.9 |
| Grok 4.6 | 65.2 |
| Claude Haiku 5.5 | 63.8 |
| Opus 4.8 | 63.2 |
| DeepSeek V4 Pro 0813 | 62.9 |
| Sonnet 5 | 61.8 |
| Muse Spark 1.2 | 61.5 |
| Grok 4.5 | 60.4 |
| Beam 1.0 501B A23B | 60.1 |
| Opus 4.7 | 59.8 |
| Sonnet 4.6 | 58.1 |
| Gemini 3.8 Flash | 55.6 |
| Inkling | 50.6 |
| Kimi K2.7 Code | 49.2 |
| Gemini 3.7 Flash | 48.9 |
| Gemini 3.6 Flash | 46.3 |
| MiniMax M3 | 45.5 |
| GLM-5.2 | 42.7 |
| Nemotron 3 Ultra 550B A55B (NVFP4) | 41.6 |
| Gemini 3.5 Flash | 39.9 |
| GLM-5.1 | 37.9 |
| Gemini 3.1 Pro Preview | 35.4 |