Scale AI
Human-labeled data, RLHF and model evals — now 49% owned by Meta
Overview
Data labeling, RLHF, data curation and model evaluation infrastructure used by AI labs and enterprises to train and fine-tune models. Since Meta's June 2025 investment for a 49% economic stake, the company has pivoted toward its enterprise applications business under interim CEO Jason Droege.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 13 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
What you can produce with Scale AI
- Data labeling across text, image, video, LiDAR and 3D sensor data
- RLHF and expert human-preference data pipelines
- Model evaluation, red-teaming and SEAL benchmarks
- GenAI Platform for enterprise fine-tuning and deployment
- Donovan and other government/defense AI applications
- Self-serve labeling plus enterprise custom-quoted contracts
ASEAN Perspective
Scale AI in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Scale built the unglamorous plumbing behind a lot of frontier models — labeled data, human-preference (RLHF) pipelines and evaluation/red-teaming — and its SEAL evals earned real respect. The story in 2026 is the Meta entanglement. In June 2025 Meta paid about $14.3B for a 49% non-voting stake (valuing Scale ~$29B) and pulled founder Alexandr Wang into its Superintelligence Labs; interim CEO Jason Droege now runs the company. The fallout matters if you're a competing lab: OpenAI and xAI paused work over data-confidentiality worries, and Google briefly cut ties before resuming months later. Scale responded by leaning into its enterprise apps business (Donovan, GenAI Platform) rather than pure labeling. Still the default for enterprises and governments that need reliable labeled data and evals at scale — but weigh the Meta stake carefully if your training data is competitively sensitive.
What people say
Coverage of Scale in 2026 is less about product quality — which is broadly respected — and more about the aftershocks of Meta's investment.
The deal itself: in June 2025 Meta paid roughly $14.3B for a 49% non-voting economic stake, valuing Scale at about $29B (a valuation reported as frozen since). Founder and CEO Alexandr Wang left to lead Meta's newly created Superintelligence Labs as Chief AI Officer while keeping a board seat, and Jason Droege — former chief strategy officer and an ex-Uber Eats executive — took over as interim CEO.
The customer fallout is the most-reported consequence. Major clients including Google, OpenAI and xAI reduced or paused engagements over data-confidentiality concerns, since Meta is a direct competitor to the labs Scale serves. OpenAI was the highest-profile loss — notable because it was the lab that first gave Scale its generative-AI start, back when Scale helped train GPT-3. Google initially dropped Scale as a labeling vendor after the announcement but resumed work a few months later.
Internally, the company restructured hard. In July 2025 Scale laid off about 200 employees (~14% of staff) plus roughly 500 contractors and consolidated 16 GenAI teams down to 5. Under Droege the strategy shifted from a roughly 70/30 split favoring data labeling toward the enterprise applications business, which reached about $200M in annualized revenue by year-end. Despite the turmoil, Scale reported its biggest revenue year ever in 2025, citing over $1B in new business across data and applications.
On reputation, Scale's SEAL evaluation work and expert human-data pipelines still carry weight with enterprises and government buyers (its Donovan defense product is frequently cited). The knock from the labeling world remains the human cost — reporting over the years has scrutinized contractor pay and conditions across its global annotation workforce.
The practical read: technically capable and still the default for enterprises needing labeled data, RLHF and evals at scale, but the Meta stake is a genuine strategic consideration for any AI lab whose training data would be sensitive to a competitor sitting on the cap table.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Scale AI's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Scale AI unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Scale AI directly →
Spotted something out of date? Suggest an update →
More in Research & Data
