#

Key takeaways

• SemiAnalysis published GDPx on November 2, 2026.

• The benchmark covers eleven domains of frontier-model work.

• It uses interactive tasks and a pass-at-one scoring method.

• The authors say the best overall model scores about twelve percent.

• SemiAnalysis also notes that safety classifiers can affect measured results.

What GDPx is

SemiAnalysis published GDPx as a benchmark for frontier-model capabilities across eleven domains. The authors are Max Kan, Andrew Megalaa, and Dylan Patel. The benchmark focuses on demanding agentic work rather than a narrow prompt-and-response setup.

The eleven domains are chip design, cybersecurity, biology, finance, legal work, quantitative trading, recursive self-improvement, hardware design, materials science, alignment, and software engineering. SemiAnalysis presents this as a benchmark-design choice. It is not described as a forecast of the economy.

How the benchmark is built

SemiAnalysis says it collaborated with fourteen specialized data partners to create the tasks. The tasks require domain expertise. They also use interactive work environments, not only a self-contained prompt and attached reference document.

The methodology includes a default eight-hour task limit. Some tasks use shorter limits. Models run in their native harness when available. Otherwise, SemiAnalysis uses OpenCode.

The headline result is pass-at-one across three attempts. Domains receive equal weight. Partners within each domain are also equally weighted. That design means the benchmark does not weight tasks by each occupation's share of the economy.

What the results show

The authors report that the best overall model scores about twelve percent on GDPx. They also discuss differences between domains. In addition, they note the influence of safety classifiers on measured results.

These observations should be attributed to SemiAnalysis. They are benchmark results and interpretation, not universal claims about model quality. The benchmark does not measure a country's GDP growth, actual investment returns, or all professions.

Why the design matters

GDPx is built around work that SemiAnalysis believes could produce early economic value from advanced AI. That is the authors' judgment about benchmark design. It is not presented as an established forecast.

The benchmark's structure matters because it tests models in settings that are closer to sustained work. The use of interactive environments suggests a focus on longer tasks and more complex workflows. The equal-weighting approach also keeps the benchmark centered on the selected domains rather than on broad labor-market shares.

Governance and measurement considerations

The source material highlights a few measurement issues. First, the benchmark uses a specific scoring method, pass-at-one across three attempts. Second, the results can be influenced by safety classifiers. Third, the benchmark's domain choices reflect editorial and technical judgment from the authors.

Those points matter when reading the results. A benchmark can be useful without being a complete measure of model capability. GDPx is a targeted evaluation of frontier-model performance in selected domains. It is not a universal score for every kind of AI work.

Bottom line

GDPx is a new SemiAnalysis benchmark for frontier models in eleven demanding domains. It uses interactive tasks, equal weighting, and a pass-at-one scoring method. SemiAnalysis reports that the best overall model reaches about twelve percent, while also warning that safety classifiers can affect the measured outcome.