AI Training Jobs for Software Engineers: Pay by Language

What a coding AI trainer does

AI labs need people who can tell working code from code that only looks right. A coding AI trainer writes problems a model should solve, writes the correct solution, builds tests that catch wrong answers and grades the model's attempt. None of it needs machine learning experience; it needs the judgment you use in code review.

The task types differ by listing. micro1's Software Engineer role asks you to "Build challenging coding problems from real repositories, with reproducible environments and verifiers" (micro1 listing). Alignerr's Python role asks you to review and evaluate AI-generated Python code for correctness and efficiency (Alignerr listing). DataAnnotation says "most tasks use Python, C++, Rust, Go, or JavaScript" (DataAnnotation).

So the list of freelance work that suits software engineers now includes this alongside client work and open source: it pays by the hour or task, needs no sales, and ends when the project ends.

AI training jobs for programmers, by language and area

Every program below had at least one live listing for the area on 14 November 2026. Pay is as each program publishes it; ranges are per listing and "up to" figures are ceilings. Programs are in alphabetical order within each row.

Sources, all read 14 November 2026: Alignerr Python, Rust, TypeScript, C++, code review and agentic coding listings; DataAnnotation backend, frontend and ML engineer roles; G2i job board; Handshake AI software engineer, senior (US), web developing and machine learning listings; Invisible Python listing; Mercor backend, frontend, full-stack, DevOps, MLOps, ML engineer, open-source maintainer and SWE-Bench auditor listings; micro1 software engineer, senior, ML engineer, open source and Codeforces listings.

Also recruiting coders, without a public rate by language: CrowdGen lists software engineers at up to $150 an hour (CrowdGen); Outlier says it depends on "subject matter experts in coding, STEM, languages, and more" (Outlier); Snorkel's coding community is "Open to eligible U.S.-based contractors; residents of California are not currently eligible" and pays by task (Snorkel); Turing lists roles for software engineers and mentions only "competitive pay" (Turing).

Hours, pay model and what "part-time" really means

Read the hours line before the rate. Three patterns show up on the listings:

• Flexible, no minimum. Handshake AI's machine learning listing says most contributors work approximately 5-20 hours per week on an active project (Handshake AI), and DataAnnotation lists "Flexible hours" with weekly payouts (DataAnnotation).
• A steady part-time block. Mercor's engineering networks state "typical commitments ranging from 15-30 hours per week" (Mercor); Alignerr's coding roles say 10-40 hours/week (Alignerr); micro1's Software Engineer role says about 15 hours a week, "paid per approved task" (micro1).
• Close to full-time. Mercor's open-source maintainer role "requires 30-40 hours per week" (Mercor), its MLOps role wants at least 40 hours a week on weekdays (Mercor), and Handshake AI's senior US contract pays "$6,000/week · 40 hours a week" (Handshake AI). These sit badly next to a full-time job.

Paid work for open-source maintainers and competitive programmers

Two niches have their own listings. Mercor's open-source role asks for a maintainer "(named in MAINTAINERS or CODEOWNERS, or 10+ PR reviews in the last 12 months)" or a core contributor of a mature, widely used project, pays $150 an hour and is open in the US, Canada and Europe (Mercor). micro1's Open Source Contributions role pays $50-100 an hour for feature, bug-fix and refactoring work in existing open-source codebases (micro1).

For competitive programmers, micro1's Codeforces Problem Setter role asks you to "Create original competitive programming problems with reference solutions and robust test suites", problems that do not exist on any online judge (micro1). Keep your own contest problems and unpublished setter work out of it unless you hold the rights to them.

Before you start: invention assignment, moonlighting, non-competes and NDAs

Most software employment contracts include an invention-assignment clause and an outside-work policy. Read both, because the risk for engineers is less about time and more about who owns what you write.

• Invention assignment. In California, an assignment clause does not cover an invention "developed entirely on his or her own time without using the employer's equipment, supplies, facilities, or trade secret information", but that carve-out excludes work that relates to the employer's business or its research and development (California Labor Code 2870). If your employer builds AI models or developer tools, a coding task may relate to its business. Other states and countries differ; ask HR or a lawyer if in doubt.
• Non-competes. California voids "every contract by which anyone is restrained from engaging in a lawful profession, trade, or business of any kind" except as its law provides (California B&P Code 16600). Elsewhere a non-compete or exclusivity clause may restrict paid work for companies in your employer's market, so check yours.
• Confidential code. Never use your employer's or a client's code, repositories, tickets or data in a task. Mercor says "Your work at Mercor will not involve access to confidential or proprietary information from any employer, client, or institution" (Mercor); hold every program to that.
• NDAs run both ways. Programs ask you to keep tasks, model outputs and client names confidential, so you usually cannot show this work in a portfolio or on GitHub.

For the engineering code of ethics on outside work, see AI training jobs for engineers. For tax on contractor income, see AI training jobs and taxes.

Which option fits your profile

A few patterns from the listings:

• Senior backend or full-stack engineers have the most choice: Mercor, DataAnnotation, Handshake AI (US) and micro1 all list general software roles.
• Python, Rust and TypeScript specialists can apply to Alignerr's language-specific reviewer roles as well as the general pools.
• ML and MLOps engineers: Mercor's ML network has the highest published ceiling; Handshake AI and micro1 ask for a PhD or master's on their ML roles.
• Students: Handshake AI's listings say F-1 students eligible for CPT or OPT may be eligible, and "STEM OPT is not supported" (Handshake AI). Talk to your school's designated official first.
• Outside the US: the Alignerr, G2i and micro1 listings in the table say remote and name no country (G2i notes that "eligibility may vary depending on your location"), Mercor works from a country list (Mercor supported countries), and Invisible's Python role is for India. DataAnnotation's own pages publish no country list; third-party reviews report that it accepts applicants only from the US, Canada, the UK, Ireland, Australia and New Zealand, so check before you apply.

Next step

Compare published rates across all fields in AI trainer pay rates compared, check a program in Are AI training jobs legit?, and if DataAnnotation is on your list read the DataAnnotation assessment first, since you get one attempt.

Tell us what you do. We'll send you the paid AI work that fits your field, wherever you live. It's free.

Origin Data Partners is independent of the programs on this page. A program may pay Origin when Origin introduces someone who joins it, which never changes what you are paid (how Origin is paid).