Compare AI task purchasing power by country and model
Wage Against the Machine is an interactive tool that compares AI task purchasing power across countries and models. It answers the question: for one hour of work, how many AI tasks can that income buy? The site pairs average gross hourly wages from ILOSTAT/OECD with published benchmark costs per task and converts them into tasks per hour worked for a range of leading AI models.
The comparison is organized around 10 countries (USA, Switzerland, Germany, Brazil, India, China, Japan, Canada, UK, Sweden) and 16 model configurations, including Opus 5 Max, Sol Max, GPT-5.5 Xhigh, and Gemini 3.6 High. Users can add countries and models to explore more combinations and view results as both a chart and a numeric table.
It is built for developers, AI buyers, researchers, and global employers who want a concrete, data-backed view of how far an hour of wages goes when buying AI benchmark tasks. The data is sourced from DeepSWE v1.1, CursorBench 3.2, FrontierCode 1.1, and ILOSTAT/OECD, and was updated July 25, 2026.
Compares tasks-per-hour across 10 countries using average gross hourly wages from ILOSTAT/OECD, from India to Switzerland.
Choose among 16 model variants such as Opus 5 Max, Sol Max, GPT-5.5 Xhigh, and Gemini 3.6 High to see tasks per hour worked.
Uses published benchmark cost per task from DeepSWE v1.1, CursorBench 3.2, and FrontierCode 1.1.
Tasks/hour equals average gross hourly wage divided by published benchmark cost/task, so results are easy to understand and verify.
Displays results as horizontal bars and numeric task counts for quick cross-country and cross-model comparison.
Lets you add additional countries and models to expand the comparison beyond the defaults.
Includes a Share to X button to share a comparison snapshot.
Tasks per hour worked equals the average gross hourly wage for a country divided by the published benchmark cost per task. Wages come from ILOSTAT/OECD, and benchmark costs come from sources such as DeepSWE v1.1.
The tool uses DeepSWE v1.1, CursorBench 3.2, and FrontierCode 1.1 benchmark costs per task. You can toggle among these benchmarks in the interface.
Default countries are USA, Switzerland, Germany, Brazil, India, China, Japan, Canada, UK, and Sweden. You can add more countries using the '+ Add country' control.
The page lists 16 model configurations, including Opus 5 Max, Sol Max, Fable 5 Max, Terra Max, Kimi K3 Max, Luna Max, GPT-5.5 Xhigh, Opus 4.8 Max, Sonnet 5 Max, Grok 4.5 High, Muse Spark, Gemini 3.6 High, GLM-5.2 Max, Kimi K2.7 Code, Sonnet 4.6 High, and Gemini 3.1 Pro.
The site states it was updated July 25, 2026. Wage data is sourced from ILOSTAT/OECD, and the DeepSWE v1.1 source is linked.
Yes, the page has a 'Share to X' button that lets you share the current comparison on X.