Top AI Agent Completes Only 12% of Tasks
A new SemiAnalysis benchmark across 11 professional domains shows frontier models still falling far short on agentic work.
Updated on Nov. 2, 2026 in Artificial Intelligence
Live Poll
Would you trust an AI system to handle important professional work for you today?
SemiAnalysis released GDPx, a benchmark testing frontier AI models on agentic tasks across 11 domains, on November 2, 2026. Anthropic's Claude Opus 5.5 led the GDPx v1.0 leaderboard with a 12.0% overall success rate; the benchmark is a research-stage evaluation rather than a deployed product.
Why it matters
GDPx selects domains by where SemiAnalysis expects frontier models to generate the highest return on investment rather than by GDP share or occupational taxonomy, reflecting a view that early AI revenue will concentrate in high-ROI sectors. Its low absolute scores suggest that agentic task completion across varied professional domains remains far from solved.
GDPx scores are overall success rates on agentic tasks, with Claude Opus 5.5 at 12.0%, Gemini 4 Carbon at 10.9%, and GPT-6 Astra at 9.6% on the v1.0 leaderboard. The extraction does not report per-domain breakdowns beyond the leaders, and no model is shown to solve tasks across all domains.
The players
SemiAnalysis
A semiconductor and AI research firm that built GDPx with 14 specialized data and reinforcement learning startups.
Anthropic
An AI developer whose Claude Opus 5.5 model topped the GDPx v1.0 leaderboard at 12.0%.
The developer of Gemini 4 Carbon, which scored 10.9% on GDPx and led in biology and software engineering.
OpenAI
The developer of GPT-6 Astra, which scored 9.6% on GDPx and led in recursive self-improvement and materials science.
The details
GDPx is built from tasks that 14 specialized data and reinforcement learning startups generated within specific domains, then used to test frontier models on multi-step work. The benchmark covers chip design, biology, and finance among its 11 domains, and it excludes government and insurance because SemiAnalysis views those sectors as offering lower return potential for AI applications. Gemini 4 Carbon led in biology and software engineering, while GPT-6 Astra led in recursive self-improvement and materials science.
Timeline
November 2, 2026: SemiAnalysis unveils the GDPx benchmark.
Every 3 to 6 months: planned cadence for GDPx updates.
The Tech Race
GDPx departs from the SOC/O*NET occupational taxonomy by choosing domains according to where SemiAnalysis expects frontier models to generate the most return. The benchmark positions competing frontier models from Anthropic, Google, and OpenAI against one another on the same agentic tasks.
GDPx is a research benchmark, not a consumer product, so its scores do not change how people use these models today. The results signal which professional domains may see AI deployment first, though the extraction gives no availability or pricing details for the tested models.
The takeaway
Watch the next GDPx update, due within 3 to 6 months, for whether the top score of 12.0% rises as partners are added. Track whether any model posts results across all 11 domains, which the extraction says no current model does.
What happens next
SemiAnalysis plans to update the GDPx benchmark every 3 to 6 months, with future versions expected to incorporate feedback and add partners.
Further reading
For broader context on how these systems are measured, see Artificial Intelligence.
Source note: This article includes information reported by Crypto Briefing.
Live Poll
Would you trust an AI system to handle important professional work for you today?
