New GDPx work benchmark: no AI agent scores above 12 percent
SemiAnalysis tests agents on 206 tasks across eleven domains, working for up to eight hours in professional software. As reported by The Decoder, Claude Opus 5.5 leads at 12 percent, Gemini 4 Carbon follows at 10.9 percent for $16.29 per task, about a third of the price, and GPT-6 Astra scores 9.6 percent. Critics question safety-filter effects and secret tasks.
Why it matters: It is a much harder test of real work than most leaderboards, though it is new and unreplicated.
| Benchmark | Score | Type | Measured or published by |
|---|---|---|---|
| GDPx overall | Claude Opus 5.5: 12% | Cited | SemiAnalysis, via The Decoder |
| GDPx overall | Gemini 4 Carbon: 10.9% | Cited | SemiAnalysis, via The Decoder |
| GDPx overall | GPT-6 Astra: 9.6% | Cited | SemiAnalysis, via The Decoder |
Sources: the-decoder.com