SemiAnalysis has introduced GDPx, a benchmark that tests AI agents on 206 work tasks using professional software and simulated work environments. Some tasks allow up to eight hours to complete. Claude Opus 5.5 posted the top score at 12%; Gemini 4 Carbon scored 10.9%, and GPT-6 Astra scored 9.6%. The methodology has drawn criticism over closed tasks, the effects of safety filters and differences in the conditions agents work under.
Source: the-decoder.com
Gemini 4 Carbon comes close to Opus 5.5 but costs only about a third as much. At the low end, GPT-6 Luna and DeepSeek 4.1 Flash hit roughly five percent for about a dollar per task. Muse Wonder Full is missing because it has no public price. | Image: SemiAnalysis
Source: the-decoder.com
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X