GDPval (Mercor 213-task rubric score)
Professional Work
Mercor's rubric-graded mean score on 213 GDPval knowledge-work tasks. GDPval itself grades deliverables by blind expert pairwise comparison, and Mercor does not document its rubric, so this is a diagnostic.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 88.3 |
| DeepSeek V4.1 Flash | 86.7 |
| GPT-6 Astra | 86.7 |
| Claude Fable 5.1 | 86.6 |
| Claude Sonnet 5.5 | 86.5 |
| Grok 4.6 | 86.0 |
| Claude Opus 5 | 85.6 |
| Claude Haiku 5.5 | 85.5 |
| Gemini 3.7 Flash | 85.2 |
| GPT-5.5 | 84.6 |
| Opus 4.8 | 84.5 |
| Kimi K3 | 84.2 |
| Sonnet 5 | 84.1 |
| Claude Fable 5 | 83.8 |
| Gemini 3.6 Flash | 83.8 |
| Muse Spark 1.2 | 83.8 |
| Gemini 3.8 Flash | 83.5 |
| GPT-6 Sol | 83.5 |
| GLM-5.2 | 82.9 |
| Opus 4.7 | 82.6 |
| Sonnet 4.6 | 82.5 |
| Muse Spark 1.1 | 82.4 |
| Grok 4.5 | 81.7 |
| GPT-5.6 Terra | 80.7 |
| GPT-5.6 Sol Pro | 80.1 |
| GPT-6 Luna | 80.1 |
| MiniMax M3 | 79.0 |
| GPT-5.6 Luna | 78.9 |
| MiniMax M2.7 | 76.2 |
| Kimi K2.7 Code | 75.9 |
| Gemini 3.1 Pro Preview | 75.3 |
| Gemini 3.5 Flash | 74.4 |
| Inkling | 73.7 |
| GPT-5.4 | 73.0 |
| DeepSeek-V3.2 | 71.9 |
| Qwen3.5 397B A17B | 71.0 |
| Kimi K2 Thinking | 66.2 |
| Nemotron 3 Ultra 550B A55B (NVFP4) | 63.5 |
| gpt-oss-120b | 58.9 |