Earn $1,000 Per Task Building AI Coding Benchmarks

If you're an experienced developer who knows a codebase well, Australian AI research company Askable has an unusual contract gig worth knowing about. The work involves turning real code changes from repositories you're already familiar with into structured tasks designed to challenge leading AI coding models. There's no cap on how many tasks you can submit, and the role is fully remote with no location restrictions.

The $1,000 per-task rate applies once a task clears their review process. Askable asks for at least one completed task per week, which they estimate takes around 10 hours. The tasks you build are used to evaluate and train AI models, so the quality bar is specific: the task needs to genuinely trip up strong AI models while remaining clearly solvable for a competent human engineer.

The process has a few distinct stages. After an initial screener, you propose a candidate change from a repo you know. If approved, you set up the repository in its pre-change state, write the problem brief, supply the correct solution, and build automated checks to verify it. Then you deliberately test bad or shortcut answers against those checks to make sure they don't slip through. Finally, you run the task against real AI models to confirm the difficulty balance is right before a human reviewer signs off.

It's a niche role, but well-suited to senior engineers who think carefully about code quality and have deep familiarity with at least one substantial open-source or professional codebase. The flexible, task-based structure also makes it stackable alongside other freelance or part-time work.

Inspired by reporting from We Work Remotely: Remote jobs in design, programming, marketing and more.