SemiAnalysis launches GDPx, an agentic capability benchmark measuring 11 domains, including chip design, cybersecurity, and RSI; Opus 5.5 tops with a 12% score (SemiAnalysis)
Safety filters affect the leaderboard: SemiAnalysis says they depress scores for Opus 5.5 and GPT-6 Astra.
GDPx tests agentic capabilities across 11 domains, with specialist data partners contributing tasks and SemiAnalysis handling final quality checks. Opus 5.5 leads at 12%, followed by Google's upcoming Gemini Carbon at 10.9%. SemiAnalysis also presents results excluding cybersecurity, biology and alignment, where safety classifiers sometimes intervene. Commenters question the binary scoring, saying models can pass more than 60% of a task's verifiers and still receive zero.
Techmeme's note
GDPx tests agentic capabilities across 11 domains, with specialist data partners contributing tasks and SemiAnalysis handling final quality checks. Opus 5.5 leads at 12%, followed by Google's upcoming Gemini Carbon at 10.9%. SemiAnalysis also presents results excluding cybersecurity, biology and alignment, where safety classifiers sometimes intervene. Commenters question the binary scoring, saying models can pass more than 60% of a task's verifiers and still receive zero.
Techmeme's note
score 6