Best AI Scores Only 12% on Real Jobs

The best AI finished only 12 percent of real jobs.

SemiAnalysis released GDPx on November 2, 2026: 206 real professional tasks across 11 domains, each with an eight-hour limit, run inside simulated workplaces — not quiz questions. Claude Opus 5.5 led at 12.0 percent. Google’s Gemini 4 Carbon scored 10.9 percent, and OpenAI’s GPT-6 Astra scored 9.6 percent. Under binary grading, nearly every model scored zero on finance and legal tasks.

Storyline

• Hook: The best AI finished only 12 percent of real jobs.
• Context: SemiAnalysis gave thirteen models 206 real tasks. Eight hours each, inside simulated workplaces.
• Stakes: Finance and legal tasks scored zero. Quiz scores flatter AI models. Real jobs punish one missed file or broken rule.
• Payoff: Use agents for drafts, never unsupervised decisions. Check every output yourself first. Send this to the one person who trusts AI blindly.

Watch next: Wall Street hedges AI compute prices (late-afternoon edition, November 2, 2026).

How we verified this: Facts checked against SemiAnalysis GDPx reporting (November 2, 2026) via multiple independent write-ups: 206 tasks, 11 domains, 8-hour limit, 13 models ranked, Opus 5.5 at 12.0%, Gemini 4 Carbon at 10.9%, GPT-6 Astra at 9.6%, and finance/legal at zero under binary grading.

Watch the Instagram reel: https://www.instagram.com/p/DfQ4tIUg87F/

SemiAnalysis GDPx leaderboard: gdpx.semianalysis.com

Follow Simple AI Explains on Instagram: https://www.instagram.com/simpleaiexplains/

Support Simple AI Explains: https://buymeacoffee.com/simpleaiexplains