Engineers tend to ask what the tasks actually are before anything else, because “AI training” sounds like labelling images and the rates do not match that.
There are 48 open engineering roles here carrying 1,680 openings, from micro1 (21), Mercor (18), AfterQuery (7) and Terac (2). Hourly rates span $25–280, with a median ceiling of $110 and an upper quartile of $150.
The work is review and evaluation, not feature work
The listings are for people who can judge whether code and system behaviour is correct, not for people to build a product.
Mercor lists a Software Engineer role for agentic search systems at $80 to $150, and a C-language role described as codebase question answering at $130. AfterQuery lists an ML Research Engineer role covering inference and GPU kernels at $100 to $170, and a Security Engineer role covering cryptanalysis and reverse engineering at the same band.
In practice the tasks are some mix of: reading model-generated code and deciding whether it is correct, ranking two solutions against each other, writing the reference solution with the reasoning made explicit, constructing test cases that expose a failure, and probing an agentic system until it does something it should not.
It is much closer to being a demanding code reviewer than to shipping features, and unlike a LeetCode screen there is rarely one right answer to produce quickly. The judgement is the deliverable.
Where the rates sit
The top of the range is seniority plus specialism. micro1 lists a Senior Software Engineer role at $245 to $280, and an AI Software Engineering Domain Expert role at $100 to $200. Mercor’s Expert Senior SWE listing runs $150 to $210.
The broad middle is Mercor’s talent network listings, which cover frontend, backend, full-stack and DevOps or platform engineering, all at $70 to $150. These take the most people in the category, and the rate inside that band depends on the project you are matched to.
micro1’s standard Software Engineer and Software Engineering Specialist roles sit at $90 to $140.
Check the pay unit before you accept
4 of the 48 engineering roles pay per completed task rather than by the hour. That distinction matters more in engineering than elsewhere, because the work is exactly the kind that gets sent back.
On a task-based contract you are paid for accepted output, so a review round that asks for changes is unpaid rework. Contractors have raised the structural problem this creates: reviewers are usually paid hourly while the task worker is not. The Mercor guide covers it in more detail.
At a similar headline rate, an hourly contract is worth more than a task-based one, because the hours you spend on revisions are the ones you cannot predict.
How does the screening work?
Every platform screens with an assessment, an AI interview, or both, and for engineering roles the assessment is usually technical.
The platforms match you to projects afterwards using the profile that screening builds, so the specialism you demonstrate matters as much as passing. An engineer who demonstrates systems and infrastructure depth surfaces for different work than one who demonstrates frontend fluency, and the rates attached to those differ.
Where to start
Browse the open engineering roles to see what is listed now with stated rates, or read whether this can be a full-time income for what the rates do not tell you about available hours.