Pay explorer
What does training AI actually pay?
The honest answer: it depends enormously on the work and your skills. The same hour can be worth a few dollars labeling images or a few hundred writing formal proofs. Here is the whole map.
- Amazon Mechanical Turk
- Sapien
- Stellar AI
- AfterQuery
- AITraining.jobs projects
- Aligned Labs
- Alignerr
- Anthropic
- Anyone AI
- Appen
- Babel Audio
- Braintrust
- Claru
- CleverX
- Clickworker
- Cohere
- DataAnnotation
- Datacurve (Shipd)
- DATAmundi
- Deccan AI (Soul AI)
- DiagnosUs
- dscout
- Edison Scientific
- Elsevier Health
- EvolutionIQ
- Fleet AI
- Handshake AI
- Hashlist
- Headway
- Hive Micro
- iMerit
- Innodata
- Invisible
- Inworld AI
- JazzX AI
- Kastle
- Lifted (Upwork)
- Limbic
- Mercor
- micro1
- Mindrift
- Mistral AI
- Neevo
- OneForma
- OpenTrain AI
- Outlier (Scale AI)
- Pareto AI
- Pasiflora AI
- Perle Labs
- Pila8
- PointClickCare
- Prime Intellect
- Prolific
- PublicAI
- RealTrainAI
- RemoExperts
- Remotasks
- Respondent
- Rise Data Labs
- rustLabs
- RWS TrainAI Community
- Sahara AI
- Sama
- Scale AI Human Frontier Collective
- Shaip (Collaborate by Shaip)
- Sigma AI
- Snorkel AI Expert Community
- Sovrano AI
- SuperAnnotate (SME)
- Surge AI
- TELUS Digital AI
- Terac
- Toloka
- TransPerfect DataForce
- Trellis
- Turing
- Twine AI
- Uber AI Solutions
- User Interviews
- Vals AI
- Vetto
- Volga Partners
- Weekday
- Welo Data
- Wirestock
- xAI (SpaceXAI)
Ordered by what applying is worth, lifted by a referral boost for the platforms that pay us a referral fee; it never changes what a platform pays you.
“Typical” is the middle of the market: a compiled benchmark across platforms and experience levels. Live per-role medians replace it here as soon as enough open roles state an hourly rate. Other figures are compiled from public sources, each labeled by how we know it. Data last refreshed 25 Jul 2026. How we compile pay. For the findings written up as a report (highest-paying work, the pay gap, what share of roles publish a rate), read The State of AI Training Pay, or see the highest-paying platforms ranked. To turn these rates into your own number, use the earnings calculator.
By type of work
12 kinds of AI training work.
Pick the work that fits your background. Each links to a full breakdown of who hires for it and what they pay.
New to AI training work? Start with the learning guides to understand what each kind of work actually involves, who it is for, and how to begin, then come back here to compare the pay.
AI Evaluation & Rating (RLHF)
Rate, rank, and compare model responses so a model learns which answers are better. The backbone of modern alignment work, and an increasingly tiered one: routine rating is commoditizing while the premium moves to credentialed expert evaluation and reinforcement-learning reward signals.
Compiled across platforms and experience levels
Coding & Software Evaluation
Review, write, and grade code so models can program: bug-finding, code-generation review, agentic-coding evaluation, and worked engineering examples. One of the best-paid categories for people with real software experience.
Compiled across platforms and experience levels
Math & STEM
Author and verify advanced math, physics, chemistry, and biology problems, including formal proofs. Rare expertise (formal-proof mathematics, physics PhDs) sits at the very top of the pay scale on several platforms.
Compiled across platforms and experience levels
Law, Medicine & Finance
Credentialed professionals (physicians, attorneys, CPAs, financial analysts, consultants) evaluate model reasoning in their field. The highest-ceiling work in the industry: in 2026 the expert-data premium climbed steeply, with medical, legal, and finance specialists commanding $250–$450/hr and the rarest consulting and executive engagements reaching $500–$1,000/hr for a verifiable license and years of experience.
Compiled across platforms and experience levels
Agentic Tasks & RL Environments
Build and grade the interactive tasks and reinforcement-learning environments that agentic models train against: author multi-step tool-use scenarios, annotate agent trajectories and reasoning traces, and design verifiable-reward graders. This is the 2026 frontier of AI-training work and one of the best-paid, because it needs real domain judgment rather than volume.
Compiled across platforms and experience levels
Data Annotation & Labeling
Annotate and label images, video, LiDAR point clouds, and text so models can perceive and categorize the world. Data annotation is the most accessible entry point into AI training, though the simplest labeling is increasingly automated by AI pre-labeling, so pay stays geography-priced with a real floor while budgets shift toward verifiable expert work.
Compiled across platforms and experience levels
Robotics & Physical-AI Data
Generate and annotate the data that teaches robots and embodied AI to act in the physical world: label robot and egocentric video, annotate vision-language-action (VLA) demonstrations, collect teleoperation and manipulation data, and quality-check robot-perception datasets. This is the newest frontier of AI-training work, growing fast as humanoid and physical-AI labs race to close the real-world data gap that text and images cannot fill.
Compiled across platforms and experience levels
Multilingual & Localization
Provide native-language data, translation, and dialect expertise, especially for the underserved and low-resource languages frontier labs are racing to cover as they close the multilingual reasoning gap. Native fluency in an in-demand language is a genuine premium skill, not just an entry ramp for non-US workers.
Compiled across platforms and experience levels
Transcription, Audio & Speech
Transcribe, edit, and annotate recorded speech so AI models can hear and speak, or record your own voice for speech datasets. It spans accessible per-task transcription and voice recording and a fast-growing premium tier: native-language transcription, audio annotation, voice and audio evaluation, and AI-trainer sound engineering now command the top rates.
Compiled across platforms and experience levels
Writing & Content
Write prompts, craft reference answers, and create or edit content that teaches models to write well. A strong fit for professional writers and educated generalists.
Compiled across platforms and experience levels
Red-Teaming & AI Safety
Adversarially probe models to find unsafe, biased, or wrong behavior before release. Blends security, domain, and creative skills; pays a premium over generalist rating.
Compiled across platforms and experience levels
Research Studies & Surveys
Take part in surveys, experiments, and human-preference studies that generate research and evaluation data. Two tiers: low-barrier consumer surveys are flexible supplemental income, while professional and expert panels (B2B interviews, paid expert consultations) pay materially more for verified expertise.
Compiled across platforms and experience levels
Estimate your rate
A rough guide based on compiled ranges, not a quote. Real pay is set per project.
Rough estimate
$50–$110/hrFrom compiled ranges
Pay by platform
Rows marked Live from feed show the typical pay (the 25th percentile of what roles pay (the middle of each advertised range) to the 90th-percentile advertised ceiling) measured from that platform’s open roles, as of 2026-07-25, and open them by pay. They refresh every hour. Every other figure is a platform’s researched rate card, compiled from public sources and dated on its platform’s pages by the day it was last checked (today from 31 Oct 2026 to 8 Nov 2026; how we research them).
Highest pay ranks the 0 platforms whose live boards state hourly rates by what a typical role there pays, as the pay ranking does. The others state no hourly rate to rank and follow them. Ordered by what applying is worth, lifted by a referral boost for the platforms that pay us a referral fee; it never changes what a platform pays you.
Some links on AITraining.jobs are referral links. If you sign up through one, AITraining.jobs may earn a commission at no extra cost to you. It never changes what a platform pays you.