newsletter.semianalysis.com web signal

SemiAnalysis Launches GDPx; Best Frontier Model Hits 12%

TL;DR

  • SemiAnalysis published GDPx on November 2, grading frontier models across eleven high-impact domains including chip design, biology, finance, legal and quant trading.
  • The best-performing model scored just 12% overall, with finance and legal returning 0% under binary grading across every model tested.
  • Opus 5.5 won quant trading, hardware design and chip design; the authors argue safety-classifier refusals suppressed its cyber and biology scores.

"Even the best model scores just 12%." That is the finding across GDPx, the agentic benchmark SemiAnalysis's Max Kan, Andrew Megalaa and Dylan Patel published on November 2. GDPx spans eleven domains the authors picked for where frontier models will have the most economic impact: chip design, cybersecurity, biology, finance, legal, quant trading, recursive self-improvement, hardware design, materials science, alignment and software engineering. "Opus 5.5 is the best all-around model and wins in quant trading, hardware design, and chip design," they write. Gemini 4 Carbon leads biology and software engineering. GPT-6 Astra tops recursive self-improvement and materials science.

Finance and legal came back empty. Every model, binary grading, zero.

The authors qualify their own headline. "Opus 5.5 and GPT-6 Astra have sandbagged overall scores due to their safety classifiers," they write, pointing at the cyber, biology and alignment splits where models refuse rather than attempt. Muse Wonder Full, with fewer refusals, ranks top on cyber.

The pitch against OpenAI's GDPval is pointed. "Every GDPval task is precisely specified and non-interactive, with all the necessary context directly spoon-fed to the model in the prompt," the authors write. GDPx builds environments instead: the finance split, assembled with Living Context, drops the model into a deal room of roughly 2,000 files from a real PE transaction representing over $100 billion in enterprise value, including a $30 billion media take-private and an $880m-plus-earnout acquisition of Haverly Snack Company. Quant trading comes via Third Moment, led by Jeffrey Yan, previously Head of Semi-systematic Options at Hudson River Trading.

On task quality: "when we run this same QA process on popular benchmarks like Terminal Bench 4.0, it flags nearly 50% of tasks as broken."

Anthropic has been in heavy rotation elsewhere this quarter: our tracker logged 357 Anthropic stories in the past 90 days, including Karpathy shipping HackerTube in roughly two hours with Opus 5.5 just before GDPx landed.