If most of your Scale work was everyday annotation and transcription, Zilo AI’s managed team can take it over. For the expert data that trains and tests large models, Scale AI’s closest competitors in 2026 are Surge AI, Mercor, Turing, Invisible Technologies and Handshake AI, with Labelbox and Snorkel AI for training environments and evaluations, and Appen and iMerit for large labeling programs.
The question matters more since Meta, one of Scale’s biggest customers, bought a non-voting stake in the company, reported by CNBC as 49% for $14.3 billion, and founder Alexandr Wang left to work on Meta’s AI efforts. If your lab competes with Meta, that changes who you want holding your training data. On July 30, 2026, Scale also named Francis deSouza as its new chief executive.
Below are 10 Scale AI alternatives, what each one sells and who it suits, then what changed at Scale and how to choose. First, our pick for teams whose Scale work was labeling rather than frontier research.
Scale AI competitors at a glance
| # | Company | What it sells | Best for |
|---|---|---|---|
| 1 | Zilo AI | Managed annotation and transcription | Everyday labeling in many languages |
| 2 | Surge AI | RLHF data, grading rubrics, RL environments | Fine-tuning data for AI labs |
| 3 | Mercor | Vetted professionals, APEX benchmarks | Tasks written and graded by professionals |
| 4 | Turing | Datasets, RL environments, experts | Coding and STEM reasoning data |
| 5 | Invisible Technologies | Training data plus enterprise AI | Data and deployment from one partner |
| 6 | Handshake AI | Graduate experts sourced from universities | PhD-level reviewers |
| 7 | Labelbox | RL environments, Alignerr experts, agent software | Environments and evaluations |
| 8 | Snorkel AI | Expert datasets, evaluations, environments | Specialized domains where models fail |
| 9 | Appen | Human data across 500+ locales | Many languages and data types at once |
| 10 | iMerit | Fine-tuning, evaluation and domain data | Medical, mobility and robotics data |
Details come from each company’s own site or announcements, checked in September 2026.
1. Zilo AI
If most of what you sent Scale was labeling, such as tagging text, drawing boxes on images or transcribing calls, you do not need a vendor built for frontier labs. Zilo’s team does text annotation, image annotation (2D and 3D bounding boxes, polygons, polylines, landmark and semantic segmentation) and voice annotation, plus audio, video and multilingual transcription.
Zilo says its team has more than 1,600 trained annotation and speech recognition experts and has annotated over 10 million data points, for industries from retail to financial services and healthcare. Its languages include German, French, Spanish, Arabic, Mandarin, Korean and Vietnamese, among many others. From Bangalore, it also provides IT staffing for roles from data engineers to generative AI engineers.
Best for: teams that want text, image or voice data labeled and transcribed by a managed team, in many languages. See Zilo’s services.
2. Surge AI
Surge AI builds the human data that AI labs use to fine-tune and test models: RLHF (reinforcement learning from human feedback) preference data, rubrics that grade model answers, reinforcement learning (RL) environments where AI agents practice tasks and get scored, and human evaluation. It works in more than 70 languages, recruits experts such as doctors, lawyers and mathematicians, and names OpenAI, Anthropic, Meta and Google among the companies it works with.
Surge says it has been profitable from day one without raising venture funding. Labs in its Trusted Program can train on its off-the-shelf datasets first and pay only if their metrics improve.
Best for: AI labs that need expert preference data, grading rubrics and RL environments.
3. Mercor
Mercor, founded in 2023, recruits experts in more than 300 professional fields, vets them with AI-run interviews, and puts them to work training and evaluating frontier models. Its APEX benchmarks test whether models can do economically valuable work in professional services, medicine and software engineering. It also builds custom AI agents for enterprises, and pays companies to license their anonymized workflow data, such as messages, documents and CRM records, to AI labs.
Best for: training and evaluation tasks that need working professionals, such as physicians, lawyers or engineers.
4. Turing
San Francisco-based Turing sells frontier labs custom and off-the-shelf datasets, RL environments and vetted experts in software engineering, enterprise knowledge work and STEM subjects such as biology, chemistry, math and physics. Its agent environments include MCP servers, computer use and terminals, and its data spans audio, images, video and LiDAR. It also works with enterprises deploying AI.
Best for: coding, agent and STEM reasoning data for fine-tuning and evaluation.
5. Invisible Technologies
Invisible, founded in 2015, has the business model closest to Scale’s: training data for model builders, plus AI systems for enterprises. For labs it offers annotation, RLHF, evaluation, red-teaming and RL environments, staffed from its Meridial expert network, and it trains and evaluates models in more than 80 languages. For enterprises it builds agents, evaluation tools and data platforms. Invisible says it has trained foundation models for more than 80% of the world’s leading AI providers.
Best for: companies that want training data and help deploying AI from the same partner.
6. Handshake AI
Handshake made its name as a careers network for college students, and has since launched Handshake AI to supply AI labs with graduate-level experts. It sources them directly from universities, so credentials are verified rather than self-reported, and the experts write domain-specific prompts, assess model answers and judge output quality in fields from law and finance to virology. Handshake runs sourcing, training and quality checks in-house on its own annotation platform, and in January 2026 it bought Cleanlab, an MIT-founded company focused on automated data quality.
Best for: labs that need PhD-level reviewers in academic subjects.
7. Labelbox
Labelbox started in 2018 as software that machine learning teams used to build training data. Today it builds Horizon, the RL environments and evaluations frontier labs train in, and runs Alignerr, a network of experts whose judgments become training signals for models. It also collects robotics data and sells Recursion, a platform for running fleets of AI agents on enterprise work. Labelbox says it is profitable.
Best for: labs that want environments, evaluations and expert feedback from one supplier.
8. Snorkel AI
Snorkel AI grew out of Stanford AI Lab research on building models without hand-labeling every example, and became a company in 2019. It now calls itself a frontier AI data lab, building expert-curated datasets, evaluation frameworks and runnable environments for labs, and offering data development as a service. It also publishes its own benchmarks for AI agents.
Best for: specialized domains where models break and generic data does not close the gap.
9. Appen
Appen has built human-labeled AI data for 30 years, starting with speech recognition. Its range now runs from reasoning traces, RLHF and red-teaming for large models to agent trajectories, speech data across more than 500 locales, video annotation, LiDAR and robotics data, and audits for hallucination and bias. It is SOC 2 and ISO 27001 certified. If Appen is also on your list to replace, see our guide to companies like Appen.
Best for: large programs that need many languages or data types from one supplier.
10. iMerit
iMerit became part of EXL when the acquisition closed on August 3, 2026. It calls itself a leader in fine-tuning, evaluation and reinforcement learning, and supplies data for autonomous mobility, healthcare AI and robotics. Its Scholars network of physicians, scientists, engineers and linguists handles reasoning evaluation and human feedback, working with customers on iMerit’s Ango Hub platform.
Best for: medical, driving and robotics data, especially if you already work with EXL.
What happened to Scale AI?
Scale is still a major supplier. It sells training data, RLHF and model evaluations to AI labs, builds AI applications for enterprises and governments, and has added customers such as Mayo Clinic, BP and Allianz. In January 2026 it said its data business had just had its strongest half-year and is now profitable. Three changes still send buyers looking around:
- Ownership. Meta’s stake carries no voting power, and the deal valued Scale at over $29 billion. Scale says it remains independent and told customers Meta would not get a board seat, but Wang, now at Meta, remains a director on Scale’s board. OpenAI told CNBC it had been winding down its work with Scale for six to 12 months and that the Meta deal did not influence that decision. Forbes reported in May 2026 that OpenAI was the most high-profile loss, while Google dropped Scale after the deal and returned a few months later.
- Focus. Under interim CEO Jason Droege, Scale shifted investment away from data labeling toward AI applications, though labeling was still the vast majority of its revenue, Forbes reported. It also laid off 14% of its staff, largely in data labeling.
- Leadership. Francis deSouza, formerly chief operating officer of Google Cloud and CEO of Illumina, took over as CEO on August 10, 2026.
None of this makes Scale the wrong choice, but ask what a new chief executive’s priorities mean for your contract.
How to choose a Scale AI alternative
Start with what you actually bought from Scale:
- Everyday labeling and transcription in many languages: Zilo AI, with a managed team for text, image and voice annotation.
- RLHF, rubrics and RL environments for fine-tuning: Surge AI, Turing, Labelbox or Snorkel AI.
- Tasks written or graded by credentialed experts: Mercor for working professionals, Handshake AI for graduate-level academics.
- Training data plus help deploying AI in your business: Invisible, or iMerit now that it sits inside EXL.
- Very large programs across many data types: Appen.
- A wider shortlist of general labeling vendors: our roundup of data annotation service providers.
Then test before you move everything. Run a paid pilot on a batch you already have trusted answers for, compare the cost per usable label rather than the price per task, and ask each vendor in writing how your data is kept separate. That question is not only about Scale: Surge lists Meta among the companies it works with, and Labelbox built a reasoning benchmark for Meta. What sets Scale apart is that Meta owns part of it. If your work is annotation or transcription, talk to Zilo about a pilot batch.
Companies like Scale AI that pay experts
If you meant companies like Scale AI to work for, Scale’s own contributor platform is Outlier. Mercor, Surge AI, Turing, Snorkel AI, Appen, Labelbox (through Alignerr) and Invisible (through Meridial) all recruit contributors, and the Handshake AI Fellowship takes people based in the US with valid work authorization. Our guide to remote data annotation jobs lists more places to apply.
Frequently asked questions
Who are Scale AI’s biggest competitors?
For everyday annotation and transcription, our pick is Zilo AI. For frontier data, Forbes names Surge AI, Mercor, Handshake and Invisible as the rivals that moved in after the Meta deal. Turing, Labelbox, Snorkel AI, Appen and iMerit also compete with Scale for AI lab and enterprise data work.
Did Meta buy Scale AI?
No. Meta bought a non-voting minority stake, reported as 49% for $14.3 billion. Scale says it remains independent, though founder Alexandr Wang, who joined Meta, remains a director on its board.
Who is the CEO of Scale AI?
Francis deSouza, since August 10, 2026. He succeeded Jason Droege, who had been interim CEO since Alexandr Wang left for Meta.
Why did companies leave Scale AI?
Mainly the Meta stake, because labs that compete with Meta were wary of sharing training data with a company it part-owns. OpenAI said its decision had nothing to do with Meta, and Forbes reported that Google left and later returned.
Is Scale AI still a good choice in 2026?
It can be, particularly for large programs that combine training data with evaluation or government work. Scale says its data business is profitable, but if you compete with Meta, get its data separation terms in writing before you sign.