SemiAnalysis launches GDPx, a tougher agentic AI benchmark
New benchmark testing frontier models on high-value domains finds even the top model scores just 12%.
What to know
- GDPx targets high-ROI economic domains rather than broad GDP- or occupation-weighted coverage, under the premise that AI revenue will concentrate in a few sectors.
- Even the top-scoring model reaches only about 12% overall, underscoring the benchmark's difficulty relative to prior tests like GDPval.
- Built-in safety classifiers in Opus 5.5 and GPT-6 Astra suppress scores on cyber, biology, and alignment tasks, meaning raw rankings understate true capability in those areas.
- SemiAnalysis positions GDPx as a needed successor to OpenAI's September 2025 GDPval, which it argues was designed before agentic models existed and is now inadequate.
SemiAnalysis Benchmark creator and publisherOpenAI Developer of GPT-6 AstraAnthropic Developer of Opus 5.5Google Developer of Gemini 4 Carbon
How it unfolded 2 developments, newest first · click a bar or a number to jump posts
-
2
SemiAnalysis argues GDPval is outdated for agentic models
The newsletter argued that OpenAI's GDPval, released September 2025 and still widely cited on model release cards, predates the agentic era and is poorly suited to evaluating current frontier models, motivating GDPx's design.
“GDPval may have been marketed as a “knowledge work” benchmark, but agents weren’t really even a thing yet.”
— SemiAnalysis -
Super interesting Benchmark. GDPx measures how well AI agents handle 206 challenging tasks across 11 economically valuable fields, including chip design, software engineering, finance, and biology. But here's the interesting part: Opus 5.5 wins with 12%, closely followed by Gemini 4 Carbon with 10...
2 more of the top 3 · 3 posts in this stretch
-
SemiAnalysis just released GDPx, a new benchmark for measuring agentic capabilities across some of the most economically valuable domains. Crazy to see Opus 5.5 at the top with just 12.0%, followed by Google's upcoming Gemini Carbon at 10.9%.
-
Today we're 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)🧵
-
-
background
Report finds safety classifiers suppress scores on cyber, bio, alignment — SemiAnalysis found that models including Opus 5.5 and GPT-6 Astra underperform on overall GDPx scores because built-in safety classifiers throttle outputs in cyber, biology, and alignment tasks, and that excluding those domains reorders the rankings.
-
background
SemiAnalysis details GDPx methodology and model rankings — A companion newsletter explained that GDPx focuses on high-ROI domains rather than broad occupational coverage, and published per-domain results showing Opus 5.5, Gemini 4 Carbon, and GPT-6 Astra each leading different categories.
-
1
SemiAnalysis unveils GDPx benchmark on X
SemiAnalysis announced GDPx, a new benchmark for measuring agentic AI capabilities across what it considers the most economically valuable domains, noting the best model scores only 12% (1/7).
“GDPx is a true frontier-level benchmark: even the best model scores just 12% (1/7)…”
— @SemiAnalysis_ -
Today we’re 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)🧵
-