SemiAnalysis Launches GDPx: Even the Best Model Scores 12%

What happened

Positioned as the successor to OpenAI's GDPval from September 2025, the benchmark was built with 14 specialist data partners and spans 11 professional domains and 206 tasks. On the debut leaderboard, Anthropic's Claude Opus 5.5 leads at 12.0%, followed by Google's Gemini 4 Carbon at 10.9% and OpenAI's GPT-6 Astra at 9.6%; the publisher is research firm SemiAnalysis and the release date is Nov 2.

Unlike question-answering tests, GDPx asks models to actually deliver work inside a simulated workplace: each task has a default limit of eight hours, and grading is pass-or-fail wherever possible, so a near-complete but undelivered result does not pass. The headline score averages each model's first-attempt pass rate over three attempts, with all 11 domains weighted equally.

By domain, Opus 5.5 leads quant trading, hardware design and chip design, Gemini 4 Carbon leads biology and software engineering, and GPT-6 Astra leads recursive self-improvement and materials science; finance and legal tasks score near zero for almost every model under binary grading. SemiAnalysis notes the totals for Opus 5.5 and GPT-6 Astra are depressed by safety classifiers refusing some cybersecurity, biology and alignment tasks.

Key facts

1. Publisher and version: SemiAnalysis, Nov 2, 2026, GDPx v1.0, authored by Max Kan, Andrew Megalaa and Dylan Patel
2. Scale: 206 tasks across 11 domains with 14 specialist data partners; domains include chip design, cybersecurity, biology, finance, legal, quant trading, recursive self-improvement, hardware design, materials science, alignment and software engineering
3. Scoring: eight-hour default limit per task; the headline score averages first-attempt pass rates over three attempts, with all domains weighted equally
4. Top three: Claude Opus 5.5 at 12.0% ($51.41 per task), Gemini 4 Carbon at 10.9% ($16.29), GPT-6 Astra at 9.6% ($53.99)
5. Next ranks: Meta Muse Wonder Full 9.0%, Claude Fable 5.1 8.5%, GPT-6.1 Sol 8.4%, Claude Sonnet 5.5 8.2%, Gemini 4 Argon 7.3%
6. Harness: each model runs in its native agent harness — Claude Code for Opus 5.5, Codex for GPT-6 Astra, Antigravity for Gemini 4 Carbon, OpenCode as fallback
7. Update plan: SemiAnalysis describes GDPx as a living benchmark with new versions every 3–6 months