Benchmark · agentic
GDPval
OpenAI's real-world knowledge-work eval: win/tie rate against industry professionals on 1,320 deliverables across 44 occupations, reported as a percentage. NOT the Artificial Analysis rerun, which is Elo-scored — file that under gdpval-aa.
| Data | Modelo | Pontuação | Fonte |
|---|---|---|---|
| 2026-08-13 | Gemini 3.7 Flash | 34.0% | Google apresenta o Gemini 3.7 Flash com melhorias em codificação e menor custo |