MLCR-AA Completeness (October 2026 methodology)
Professional Work
Completeness under Artificial Analysis' October 2026 MLCR-AA methodology: the share of responses that a majority of the three judges find answer every part of the question, agree with the ground-truth conclusion and name each documentation gap on the way to it, averaged over three runs of 60 questions. It is a second reading of the run behind the MLCR-AA pass rate and stays diagnostic.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Sonnet 5.5 | 64.8 |
| Claude Opus 5.5 | 63.3 |
| Claude Fable 5.1 | 61.1 |
| Claude Haiku 5.5 | 51.7 |
| Claude Opus 5 | 45.0 |
| Claude Fable 5 | 44.4 |
| GPT-6 Astra | 42.2 |
| Sonnet 5 | 40.0 |
| GPT-6.1 Sol | 40.0 |
| Opus 4.8 | 37.8 |
| GPT-5.6 Sol | 37.8 |
| Sonnet 4.6 | 37.2 |
| Kimi K3 | 36.7 |
| GLM-5.3 | 36.1 |
| GPT-5.6 Terra | 36.1 |
| GPT-5.5 Instant (2026-06-25 hosted snapshot) | 35.6 |
| GPT-6 Sol | 35.6 |
| DeepSeek V4.1 Flash | 34.4 |
| GLM-5.3 Flash | 33.9 |
| Qwen3.8-Max | 32.8 |
| Qwen3.8 2.4T A95B | 32.2 |
| Gemini 4 Argon | 31.7 |
| GPT-5.5 | 31.7 |
| Grok 4.7 | 31.1 |
| MiMo-V2.6-Pro | 31.1 |
| Ling 3.1 Flash | 30.6 |
| Muse Spark 1.3 | 30.0 |
| Qwen3.8 27B | 30.0 |
| Step 5 Preview | 29.4 |
| GLM-5.2 | 28.9 |
| Qwen3.8-Flash-Next | 28.9 |
| Grok 4.6 | 28.3 |
| Gemini 3.8 Flash | 27.8 |
| GPT-5.6 Luna | 27.8 |
| GPT-6 Luna | 27.8 |
| MiMo-V2.6-Flash | 26.7 |
| Inkling | 26.1 |
| Qwen3.8 Max (0902) | 25.6 |
| Grok 4.5 | 23.9 |
| Quasar 438B | 23.9 |