GDPval-AA v2 (normalized)
Professional Work · 2026-04-18
GDPval-AA v2 is a normalized benchmark for real-world professional tasks across 44 occupations and 9 industries. Models use shell and web access to produce documents, slides, diagrams, and spreadsheets; higher scores mean better judged task performance.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 67.9 |
| Claude Fable 5 | 62.2 |
| GPT-5.6 Sol | 61.6 |
| Kimi K3 | 59.4 |
| Sonnet 5 | 55.0 |
| Opus 4.8 | 54.5 |
| GPT-5.6 Terra | 54.1 |
| GPT-5.6 Luna | 54.1 |
| DeepSeek-V4-Flash-0731 | 52.9 |
| Grok 4.5 | 51.4 |
| GLM-5.2 | 50.5 |
| Opus 4.7 | 49.5 |
| GPT-5.5 | 49.5 |
| Gemini 3.6 Flash | 46.2 |
| GPT-5.4 | 44.6 |
| MiniMax M3 | 44.5 |
| Sonnet 4.6 | 43.8 |
| Muse Spark 1.1 | 43.8 |
| Gemini 3.5 Flash | 42.1 |
| DeepSeek-V4-Pro | 40.2 |
| Qwen3.7-Max | 38.5 |
| Inkling-Small | 38.4 |
| JT-4.1 Flash 236B A21B | 38.3 |
| MiMo-V2.5-Pro | 38.2 |
| Motif-3-Beta | 37.9 |
| GLM-5.1 | 37.8 |
| Nex-N2-Pro | 37.5 |
| Inkling | 36.8 |
| Hy3 | 35.8 |
| Grok Build 0.1 | 35.7 |
| DeepSeek-V4-Flash | 34.5 |
| Kimi K2.7 Code | 34.4 |
| Kimi K2.6 | 34.4 |
| Agnes 2.5 Pro Alpha | 33.6 |
| GPT-5.4 Mini | 33.5 |
| GLM-4.7 | 33.2 |
| Nemotron 3 Ultra 550B A55B | 33.1 |
| MiniMax M2.7 | 32.9 |
| MiMo-V2.5 | 32.2 |
| Muse Spark | 32.2 |