SemiAnalysis Releases GDPx Benchmark; Top Model Scores Just 12% on Real Professional Work
AccBalanced · x · 2026-11-02
SemiAnalysis has launched GDPx, a frontier-level agentic benchmark measuring how well AI agents perform economically valuable professional work rather than generic knowledge tasks.
- Structure: 206 tasks across 11 domains, built with 14 specialized partners, covering chip design, financial analysis, legal navigation, and other high-value work.
- First results: the best model, Claude Opus 5.5, scores only 12.0%, exposing a wide gap between conversational competence and dependable autonomous work.
- Takeaway: GDPx targets whether agents can actually complete real work, offering a much stricter yardstick for frontier agentic capability.
More from AGI Musings
- DeepMind: why AI keeps breaking records while the economy barely budges — JMateosGarcia · 2026-11-02
- AI agents must fit 'organisational mosaics' to unlock unbounded economic impact — JMateosGarcia · 2026-11-02
- DeepMind essay: coordination, not capability, is the real barrier to AGI's economic impact — JMateosGarcia · 2026-11-02
- Aerospace insider explains how Mach tuck may have downed a Learjet 60 in Florida — bingxu_ · 2026-11-02
- Warning: AI-enabled cyber attacks are outpacing hand-crafted defenses — uwukko · 2026-11-02
- Survey of 8,000 adults: 81% satisfied at work, UBI can't replace meaning of jobs — mikeflache · 2026-11-02