No AI agent cracks 12% on SemiAnalysis' real-world work benchmark
semianalysis releases gdpx benchmark, testing ai agents in realistic work environments, with claude opus 5.5 leading at 12 percent, highlighting challenges in practical ai applications.
semianalysis has released its gdpx benchmark, which tests ai agents in simulated work environments with 206 tasks across eleven domains, including chip design, biology, and cybersecurity. claude opus 5.5 leads the field with 12 percent, closely followed by gemini 4 carbon at 10.9 percent and gpt-6 astra at 9.6 percent. the benchmark's results are impacted by safety filters, which affect rankings, and the secrecy surrounding test tasks has drawn scrutiny. the benchmark aims to assess ai agents' performance in realistic work environments, highlighting challenges in practical applications.