SemiAnalysis released GDPx on 2 November, a benchmark it says measures agentic capabilities across the most valuable domains for frontier models. In the research firm’s leaderboard, Anthropic’s Claude Opus 5.5 scores 12.0%, the highest of the 13 models shown, and SemiAnalysis wrote that “even the best model scores just 12%”.
Google’s Gemini 4 Carbon is second at 10.9%. By our arithmetic, that is about 91% of Opus 5.5’s score at about 32% of its cost per task: $16.29 against $51.41 in SemiAnalysis’s table.
Business Pill · Distribution Advantage
A short explainer on distribution advantage. It is a general explainer and is not about this story.
The key insight: SemiAnalysis’s new GDPx benchmark spans 206 tasks in 11 domains, and its top model, Claude Opus 5.5, scores 12.0%; Gemini 4 Carbon reaches about 91% of that score at about 32% of the cost per task, by our arithmetic from SemiAnalysis’s table.
The benchmark has 206 tasks across 11 domains, built with 14 data partners, according to SemiAnalysis. Chip design has the most tasks, 34, followed by finance with 24 and materials science with 23. The other domains are hardware design, legal, cybersecurity, alignment, software engineering, biology, recursive self-improvement and quant trading.
The leaderboard says the score “averages each model across all 11 domains, each weighted equally, as is each partner within a domain”. Each task is graded on the average of 3 attempts, according to the leaderboard’s footnote.
SemiAnalysis wrote that weighting a general-purpose benchmark by share of GDP or by an occupational taxonomy “is therefore a mistake”, and that it chose “chip design, cyber, and bio over domains like government, real estate, and insurance”.
GDPx score across all 11 domains against cost per task for the 12 models with a listed cost in SemiAnalysis’s 2 November leaderboard. The 91% and 32% comparisons for Gemini 4 Carbon against Claude Opus 5.5 are our arithmetic from that table.
What Was Posted on X
Today weโre releasing GDPx, a benchmark for measuring agentic capabilities across the most valuable domains.
GDPx is a true frontier-level benchmark: even the best model scores just 12% (1/7)๐งต
SemiAnalysis’s 2 November post announcing GDPx, which it describes as a benchmark for agentic capabilities across the most valuable domains, and in which it writes that even the best model scores just 12%. The leaderboard image in the post is the source of the scores and costs in this piece.
The Ranking
After Opus 5.5 and Gemini 4 Carbon come OpenAI’s GPT-6 Astra at 9.6%, Meta’s Muse Wonder Full at 9.0%, Claude Fable 5.1 at 8.5% and GPT 6.1 Sol at 8.3%. The lowest scores in the table are Grok 4.7 at 3.9% and Z.ai’s open model GLM 5.3 at 2.7%.
SemiAnalysis said some models were held back by their own safety systems: “certain models like Opus 5.5 and GPT-6 Astra have sandbagged overall scores due to their safety classifiers”. Leaving out cybersecurity, biology and alignment, GPT-6 Astra leads at 11.8% and Opus 5.5 scores 11.7%, in its second table.
The newsletter adds that “Our finance and legal tasks are so challenging that basically every model scores exactly 0”, and that “Muse Wonder Full is actually the best available cyber model today by a meaningful margin”.
The Business Lever: Cost per Point
In SemiAnalysis’s table, DeepSeek 4.1 Flash costs $0.94 per task for 5.4% and GPT-6 Luna $0.92 for 5.1%, while Claude Fable 5.1 costs $67.16 for 8.5%. The newsletter describes the cost-performance frontier this way: “Luna and DeepSeek fight for the low end, Opus wins the high end, and OpenAI claims the middle”.
As we read it, the low ceiling is by design: SemiAnalysis says it chose the domains where frontier models will be most impactful rather than weighting by share of GDP.
As we read the two tables, the leader depends on which domains count: Opus 5.5 leads on all 11, while GPT-6 Astra leads once cybersecurity, biology and alignment are left out.
As we read the cost column, the spread is wide: per-task costs in SemiAnalysis’s table run from $0.92 for GPT-6 Luna to $67.16 for Claude Fable 5.1.
SemiAnalysis, 2 November 2026
“We believe the first trillion dollars of AI revenue will be concentrated in sectors that can deploy frontier intelligence at high ROI, not broadly diffused throughout the whole economy”
Three Implications
FOR MODEL BUYERS SemiAnalysis’s table lists cost per task next to each score, from $0.92 for GPT-6 Luna to $67.16 for Claude Fable 5.1.
FOR LABS SemiAnalysis says safety classifiers held back the overall scores of Opus 5.5 and GPT-6 Astra in cybersecurity, biology and alignment.
FOR BENCHMARK MAKERS SemiAnalysis says it built GDPx with 14 specialised data partners and positions it as a successor to OpenAI’s GDPval, released on 25 September 2025.
The framework puts it this way: “Orchestration, memory, tool use, and workflow execution have become the new sources of competitive advantage.”
As we read it, GDPx measures that layer directly: SemiAnalysis describes it as a benchmark of agentic capabilities rather than single prompts. The framework does not discuss GDPx.
The scores, costs and task counts are SemiAnalysis’s own results; we have not rerun any task. The leaderboard shows 13 of 20 models. Gemini 4 Carbon is not yet generally available, according to SemiAnalysis, which also notes that results for models with safety classifiers depend on which domains are counted. The 91% and 32% comparisons are our arithmetic from the published table.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
SemiAnalysis’s new GDPx benchmark spans 206 tasks in 11 domains, and its top model, Claude Opus 5.5, scores 12.0%. Gemini 4 Carbon comes within about 91% of that score at roughly a third of the cost per task, by our arithmetic from SemiAnalysis’s table.
A note on sourcing. Scores, costs, task counts and quotations come from SemiAnalysis’s 2 November X thread, its GDPx leaderboard images and the public portion of its newsletter. The 91% and 32% comparisons are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
