MLCR-AA Accuracy (Judged Responses)
Professional Work
MLCR-AA accuracy among nonempty responses meeting the conciseness limit and sent to the three-judge panel. This conditional denominator excludes overlong and empty responses.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 98.3 |
| GPT-6 Sol | 96.7 |
| Claude Fable 5.1 | 94.8 |
| GPT-6 Luna | 93.8 |
| GPT-5.6 Terra | 93.7 |
| GPT-5.6 Sol | 92.5 |
| Claude Fable 5 | 90.1 |
| Qwen3.8-Max | 89.9 |
| Qwen3.8 27B | 89.3 |
| Opus 4.8 | 89.0 |
| Claude Opus 5 | 88.6 |
| GPT-5.5 | 87.4 |
| Qwen3.8 Max (0902) | 86.6 |
| Kimi K3 | 85.0 |
| Grok 4.7 | 84.7 |
| Sonnet 5 | 84.1 |
| DeepSeek V4.1 Flash | 83.3 |
| GPT-5.6 Luna | 82.8 |
| Step 5 Preview | 82.8 |
| DeepSeek V4 Pro 0813 | 81.4 |
| GLM-5.3 | 81.3 |
| Muse Spark 1.3 | 81.1 |
| DeepSeek-V4-Pro | 80.7 |
| DeepSeek-V4-Flash-0731 | 80.6 |
| DeepSeek-V4-Flash | 80.6 |
| Grok 4.6 | 77.7 |
| GLM-5.3 Flash | 76.4 |
| GPT-5.4 Mini | 75.4 |
| GPT-5.4 Nano | 74.7 |
| Sonnet 4.6 | 73.7 |
| Gemini 3.1 Pro Preview | 72.1 |
| Gemini 3.8 Flash | 69.7 |
| Grok 4.5 | 69.5 |
| Gemini 3.6 Flash | 68.4 |
| Gemini 3.7 Flash | 67.8 |
| Muse Spark 1.2 | 66.9 |
| Gemini 2.5 Flash | 65.5 |
| Opus 4.7 | 64.2 |
| Kimi K2.7 Code | 64.2 |
| GLM-5.2 | 62.2 |