GDP.pdf (Artificial Analysis, Mean Criteria)
Professional Work
Task-macro mean atomic-criterion pass rate for Artificial Analysis' five-attempt GDP.pdf implementation, with GPT-5.6 Luna Medium judging across 100 tasks. Distinct from the stricter all-pass metric.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 82.3 |
| GPT-6 Astra | 82.0 |
| Claude Fable 5.1 | 80.2 |
| Muse Spark 1.3 | 78.1 |
| GPT-5.6 Sol | 77.6 |
| GPT-6 Sol | 77.5 |
| Claude Fable 5 | 77.2 |
| Qwen3.8 Max (0902) | 76.0 |
| GPT-5.6 Terra | 75.9 |
| Gemini 3.8 Flash | 75.9 |
| Claude Opus 5 | 75.9 |
| Kimi K3 | 75.7 |
| GPT-5.6 Luna | 75.5 |
| Muse Spark 1.2 | 75.3 |
| Gemini 3.7 Flash | 75.2 |
| GPT-5.5 | 75.0 |
| Opus 4.8 | 74.9 |
| Qwen3.8-Max | 74.6 |
| Muse Spark 1.1 | 74.2 |
| GLM-5.3 Flash | 71.5 |
| Grok 4.7 | 71.2 |
| Grok 4.6 | 71.1 |
| Gemini 3.6 Flash | 70.8 |
| GPT-5.5 Instant (2026-06-25 hosted snapshot) | 70.5 |
| Gemini 3.5 Flash | 70.4 |
| Grok 4.5 | 70.3 |
| GPT-6 Luna | 69.8 |
| Sonnet 4.6 | 68.9 |
| Gemini 3.1 Pro Preview | 68.7 |
| MiMo-V2.6-Pro | 68.1 |
| DeepSeek V4.1 Flash | 67.5 |
| Step 5 Preview | 67.2 |
| Qwen3.8 27B | 67.1 |
| Sonnet 5 | 66.6 |
| Kimi K2.7 Code | 66.3 |
| Qwen3.8-Flash-Next | 65.4 |
| Qwen3.8 2.4T A95B | 64.6 |
| Inkling | 64.5 |
| Muse Glimmer 30B | 64.3 |
| Gemini 3.5 Flash-Lite | 64.0 |