What this tool does
It shows what people in a given role actually report earning, country by country, so you can sanity-check a job offer or a freelance rate. Each source appears on its own card with its sample size and licence. Sources are never blended: a survey median and a government median are different measures, so they stay side by side.
For AI training work, the figures come from listings with stated hourly pay on our jobs board, published as the open AI training pay dataset. Those cards cover remote listings from all countries together.
Where the numbers come from
Three sources, each with its own method:
- Stack Overflow Developer Survey 2026 (database licensed ODbL 1.0): annual total compensation converted to US dollars by Stack Overflow. We keep employed people and freelancers, drop students and retirees, and keep pay between $1,000 and $1,000,000 a year, leaving 12,114 responses. Median and quartiles use linear interpolation.
- US Bureau of Labor Statistics OEWS, May 2025 (public domain): national 25th percentile, median and 75th percentile annual wages for wage-and-salary employees. US only, and it excludes the self-employed.
- Our job board: every hourly listing with stated pay counts once, at the midpoint of its range. Categories are model evaluation, coding, research, writing and feedback, and data annotation.
Worked example
Software developer, India: 291 survey responses, a median of $15,669 a year and a middle half from $7,312 to $30,286. The all-countries median is $80,000 from 7,756 responses.
Software developer, United States: the survey median is $150,000 from 1,789 responses. BLS gives software developers a median of $135,980, with the 25th to 75th percentile running from $105,210 to $171,980. They differ because they measure different groups.
AI trainer (model evaluation): 577 listings posted between 18 January and 15 November 2026 had a median of $77.50 an hour, with the middle half from $50 to $105. Coding listings had a median of $90 (82 listings) and data annotation $40 (152). To turn an hourly figure into a yearly one, use the salary converter.
Limitations
Survey cells with fewer than 30 responses show "not enough data" instead of a figure. Data scientists in India, with 5 responses, are an example.
The survey reflects the people who answered it, mostly developers, and is not limited to remote jobs. Stated pay on a listing is not the same as what workers earn after fees and gaps between projects. Medians in US dollars also ignore local prices; compare buying power with the PPP salary calculator.