Skip to main content

SemiAnalysis releases GDPx, a benchmark for agentic AI

2 NOVEMBER 2026·2 MIN READ·5 SOURCES

SemiAnalysis released GDPx on November 2, 2026, presenting it as a benchmark for measuring general agentic capabilities. It covers 11 domains, with 14 partners and 206 tasks.

SemiAnalysis releases GDPx, a benchmark for agentic AI

Key takeaways · 4

  • 01

    GDPx spans 11 domains and comprises 14 partners and 206 tasks.

  • 02

    Claude Opus 5.5 leads the leaderboard, but its average task cost is $51.41; Gemini 4 Carbon scores 10.9% at $16.29 per task.

  • 03

    Finance and legal tasks produced zero scores for every model under their intended binary grading, so SemiAnalysis reported average partial scores for those domains.

  • 04

    GDPx tasks have a default time limit of eight hours, and each partner open-sourced the rollout for one task.

A wide-ranging benchmark

SemiAnalysis says GDPx aims to be the world’s best benchmark for measuring general agentic capabilities.[2] Its 11 domains include chip design, cybersecurity, biology, finance, legal, quant trading, recursive self-improvement, hardware design, materials science, alignment and software engineering.[2] The benchmark comprises 14 partners and 206 tasks.[3] SemiAnalysis says it intentionally left out industries including government, real estate and insurance.[2] The company argues that AI’s first trillion dollars of revenue will concentrate in sectors where frontier intelligence can be deployed for high returns.[2]

Scores and costs differ

The leaderboard ranks 13 models.[1] Claude Opus 5.5 leads at 12.0%, with an average cost of $51.41 per task and 548,000 output tokens.[1] Gemini 4 Carbon is second at 10.9% and $16.29 per task, while GPT-6 Astra is third at 9.6% and $53.99 per task.[1] Meta’s Muse Wonder Full ranks fourth at 9.0%, with no public price listed.[1] Z.ai’s GLM 5.3 ranks 13th at 2.7%.[1]

How to read the results

SemiAnalysis says the headline score averages pass@1 across three attempts, weights all domains equally and uses each task’s intended binary or continuous grading method.[1] Every model scored zero on the finance and legal tasks under their intended binary grading, so SemiAnalysis reported average partial scores for those domains.[1] GDPx tasks have a default time limit of eight hours.[1] These scoring choices matter when comparing a model’s overall rank with its performance on a particular task or domain.

Domain leaders and access

SemiAnalysis reports Claude Opus 5.5 as the best all-around model and the winner in quant trading, hardware design and chip design.[2] It says Gemini 4 Carbon leads in biology and narrowly wins software engineering, while GPT-6 Astra wins in recursive self-improvement and materials science.[2] SemiAnalysis also says safety classifiers depress the overall scores of Opus 5.5 and GPT-6 Astra.[2] Each GDPx partner open-sourced the rollout for one task; the remaining rollouts stay private to prevent hill climbing.[4] SemiAnalysis plans to update GDPx every three to six months.[5]

For teams assessing agentic systems, GDPx offers results across multiple technical and professional domains, but its overall ranking is only one input to a decision. Compare domain performance, task costs and scoring rules against the work you need a system to do.

Why it matters
Story quiz

Bring this to your team — one session a day, built for your industry.

kju.ai is built for teams: daily sessions, team analytics, and rollout support for your organization.

For teams

How this developed

  1. 2 November 2026

    SemiAnalysis releases GDPx, a benchmark for agentic AI

Sources

AI fluency, one session a day, built for your work.