GDPval-AA v2
Professional Work · 2026-06-16
Artificial Analysis GDPval-AA v2 evaluation using 220 GDPval gold tasks across 44 occupations. V2 introduced a new sandbox and dependency environment, a three-judge panel, longer agent runs, and an Elo scale rebaselined to a human expert at 1000.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 1861 |
| Claude Fable 5.1 | 1764 |
| Muse Spark 1.3 | 1703 |
| GPT-5.6 Sol | 1702 |
| Qwen3.8 Max (0902) | 1689 |
| Kimi K3 | 1668 |
| Grok 4.6 | 1663 |
| GLM-5.3 Flash | 1655 |
| Claude Fable 5 | 1648 |
| Qwen3.8-Flash-Next | 1647 |
| GLM-5.3 | 1636 |
| DeepSeek V4.1 Flash | 1632 |
| Qwen3.8-Max | 1630 |
| Qwen3.8 2.4T A95B | 1628 |
| Sonnet 5 | 1618 |
| GPT-5.6 Terra | 1593 |
| GPT-6 Astra | 1580 |
| DeepSeek V4 Flash Vision Exp | 1577 |
| Agnes 3.0 Flash (hosted) | 1573 |
| GPT-5.6 Luna | 1539 |
| Grok 4.5 | 1535 |
| Opus 4.8 | 1529 |
| Muse Spark 1.2 | 1523 |
| GLM-5.2 | 1514 |
| DeepSeek V4 Pro 0813 | 1493 |
| GPT-5.5 | 1471 |
| Gemini 3.8 Flash | 1464 |
| Qwen3.8 27B | 1463 |
| DeepSeek-V4-Flash-0731 | 1442 |
| Gemini 3.7 Flash | 1433 |
| Gemini 3.6 Flash | 1421 |
| K2 Horizon 375B A23B | 1400 |
| Opus 4.7 | 1396 |
| Agnes 2.5 Pro Beta | 1376 |
| Gemini 3.5 Flash | 1349 |
| GPT-5.4 | 1307 |
| MiniMax M3 | 1304 |
| Sonnet 4.6 | 1295 |
| Muse Spark 1.1 | 1294 |
| Apodex 1.1 | 1273 |