- Verified by BizGlocal
Data Engine, GenAI Platform And Government AI With Assigned Engagement Managers
- 1,000+
- 2016
- San Francisco, CA
- 8+
About Scale AI
Industry leader in connected fleet operations
Market Position
One of the best-known names in AI training data, with a customer base that has historically included frontier labs, autonomous vehicle makers and the US government. Its position changed after the 2025 Meta deal, and the two lab customers press reports said were pulling back were OpenAI and Google.
Innovation Focus
A move beyond labeling toward applications: the GenAI Platform for agents on enterprise data, Donovan for defense and national security, and Scale Evaluation for testing frontier models.
Core Strengths
Expert and generalist human labeling combined with ML checks, a track record on frontier model post-training, government-grade authorizations on the GenAI Platform and Donovan, and Engagement Managers assigned to Pro and Nucleus accounts.
Use Cases
RLHF and preference data, red teaming and safety evaluations, autonomous vehicle and robotics annotation, document and text processing, and building evaluated agents for healthcare, finance and public-sector work.
Core Products & Solutions
Scale Data Engine
The core data platform for collecting, curating and annotating data, then training and evaluating models in a loop. Scale lists text, image, video and 3D sensor fusion including LiDAR, with ML pre-labeling and human review driven by confidence scores.
Generative AI Data Engine
Prompt and response generation, RLHF, red teaming and evaluation for generative models. Scale presents this as the workflow used to improve large language models after pre-training.
Scale Evaluation
A platform to compare models across thousands of tests, break results down by language, topic or task, and cluster errors into recurring modes. Scale updated it in April 2025 and says its SEAL team built benchmarks such as MultiChallenge and MASK.
Scale GenAI Platform
A platform for building agents that reason over a company’s own data, with connectors for Confluence, SharePoint and S3, deployment in your own VPC on AWS, Azure or GCP, and support for models from OpenAI, Google, Meta and Mistral. Scale lists SOC 2 Type II, ISO 27001 and FedRAMP High on the page.
Scale Donovan
A no-code agent platform for public-sector and defense workflows with side-by-side model comparison and traceable reasoning. Scale’s page says its unclassified government cloud is DISA IL4 and FedRAMP High authorized, and that it runs on air-gapped networks.
Scale Pro And Nucleus
Pro is a labeling platform with API access and dedicated Engagement Managers who set up custom projects. Nucleus is a dataset management tool for reviewing annotations and measuring model performance.
Pricing Overview
Data Labeling & Data Engine
Teams that need annotation, RLHF or evaluation data delivered through managed projects
Custom
Pricing
Custom quote; the pricing page is a demo request with no rates
- Text, image, video and 3D sensor fusion labeling
- RLHF, red teaming and model evaluation data
- ML pre-labeling plus human review on confidence scores
- Pro and Nucleus accounts name an Engagement Manager or Field Engineer
- No free tier or self-serve checkout is listed
GenAI Platform
Enterprises that want evaluated agents running on their own data and cloud
Custom
Pricing
Custom quote; the product page says “Talk to our team”
- Data connectors and in-VPC deployment on AWS, Azure or GCP
- Agent orchestration, evaluation and monitoring
- Post-training of open-weight models
- Model choice across OpenAI, Google, Meta, Mistral and others
- A G2 reviewer says pricing can add up quickly when human review is needed
Donovan For Government
Defense and public-sector teams building mission agents on controlled networks
Custom
Pricing
Custom quote; scoped per agency and deployment
- Agent Factory no-code builder and prebuilt agents
- Model comparison on accuracy, hallucination rate and latency
- Government cloud listed as DISA IL4 and FedRAMP High authorized
- Air-gapped deployment described on the product page
- Contract vehicles and accreditation dates are not listed
Who Is Scale AI Best For?
Small Businesses
1-25 EmployeesRarely the right fit. There is no free tier or published price, a G2 reviewer at a small firm calls it built for huge enterprise budgets, and support is delivered through assigned managers on larger accounts. A small team with a modest labeling job should confirm minimums first.
Mid-Market Companies
26-100 EmployeesWorkable for a funded AI product team with a defined labeling or evaluation need. Expect a sales process, and plan for the learning curve a G2 reviewer describes when configuring custom labeling workflows.
Enterprise Companies
100+ EmployeesScale’s core audience, with customers such as BP, Mayo Clinic, EY and NVIDIA on its site, SOC 2 Type II, ISO 27001 and FedRAMP High listed for the GenAI Platform. Ask which authorizations cover the product you are buying.
AI Labs And Public-Sector Teams
Any SizeModel builders needing expert RLHF and evaluation data, and agencies needing accredited deployments, gain the most. Labs that compete with Meta should ask how their data is isolated.
Measurable Impact & Benefits
90%
Of Leading Model Builders
Scale says 90% of the world’s leading generative AI model builders are powered by Scale, a vendor claim on its homepage.
11
More Minutes Per Patient
Scale’s Mayo Clinic story says doctors have spent an average of 11 more minutes with each patient since launch.
25%
With Advanced Degrees
Scale says 25% of its data contributors have advanced degrees, a vendor-stated figure on its homepage.
10X
Throughput Ramp
Scale’s labeling page says Toyota Research Institute reported a 10X throughput ramp, attributed to the customer.
Days Or Weeks To Seconds
Time To Insights
Global Atlantic reports on Scale’s GenAI Platform page that time to insights fell from days or weeks to seconds.
4.5
G2 Rating
G2 shows Scale GenAI Platform at 4.5 out of 5 from 16 reviews, which does not cover data labeling.
9 To 5
Pacific Support Hours
Scale’s support page lists 9 AM to 5 PM Pacific, Monday to Friday, excluding major US holidays.
1
Assigned Contact
Pro and Nucleus customers are directed to an Engagement Manager or Field Engineer, per Scale’s support page.
Scale AI Real reviews
Our Rating:
4.4
- Verified Users
- Verified Users
Compare Alternatives
See how Scale AI stacks up against competitors
Get Your Free Checklist
Enter your email to download instantly
No spam. Instant access. Unsubscribe anytime.
Download
Spam
Forever
Scale AI’s Pros & Cons
Balanced evaluation to help you make an informed decision
Strengths
- Pro and Nucleus customers are assigned an Engagement Manager or Field Engineer who triages quality and technical issues, per Scale’s support page
- Managed labeling with ML pre-labeling, confidence-based human review and a worker evaluation pipeline, as described by Scale
- SOC 2 Type II, ISO 27001, DoD IL4 and FedRAMP High listed for the GenAI Platform, and IL4 and FedRAMP High for Donovan, per Scale
- Broad data coverage across text, image, video, LiDAR and RLHF, plus red teaming and evaluation
- Named customers including BP, Mayo Clinic, EY, NVIDIA and Toyota Research Institute, with a quantified Mayo Clinic result
- Gartner Service & Support subrating of 4.5, and a G2 rating of 4.5 for the GenAI Platform
- New CEO Francis deSouza, a former Google Cloud COO, from August 2026, with an enterprise and government focus
Limitations
- No published prices, free tier or trial, and G2 reviewers call it costly and built for large budgets
- Support hours are 9 AM to 5 PM Pacific on weekdays, and the page lists no response times, SLAs or escalation process
- Public review evidence is thin: 16 G2 reviews and 2 to 3 Gartner ratings, mostly for the GenAI Platform, with no Capterra reviews
- Meta’s 49% stake led to reports that OpenAI, Google and Microsoft were pulling back, and roughly 200 staff and 500 contractors were cut in July 2025
- Business Insider reported in June 2025 that Scale left confidential client files in publicly accessible Google Docs, and Scale said it disabled public sharing
- Contractor labor disputes, including 2024 and 2025 wage and misclassification suits, and a Department of Labor probe that was dropped in May 2025
Our Verdict
Scale AI earns its Best in Support placement on the model of support, not on published proof. Pro and Nucleus customers get an assigned Engagement Manager or Field Engineer, labeling runs as a managed service with human review, and the GenAI Platform and Donovan carry listed government authorizations and a 4.5 Gartner Service and Support subrating. Rival Labelbox shows that published terms are possible: its Enterprise plan lists dedicated technical support, a shared Slack channel and an SLA. Scale does not. The caveat is that the evidence is weak. Scale’s support page gives weekday office hours and no response times, G2 has no support score for it, one G2 reviewer says advanced customization needed extra support engagement, and the Gartner subrating rests on 2 reviews. Add the Meta stake, the 2025 layoffs and the exposed Google Docs, and buyers should require written support terms, named contacts, data isolation commitments and a termination right before signing.
Resources
Resources & Documentation
Detailed materials for deeper evaluation
Customer Support Documentation
Support hours and who Pro and Nucleus customers contact
Data Engine Product Page
Data types, generative AI services, quality claims and customer logos
Data Labeling Page
Pre-labeling, ML checks, human review and customer testimonials
GenAI Platform Page
Connectors, deployment options, models and listed security certifications
Donovan Product Page
Agent Factory, model comparison and government authorization claims
Scale Evaluation Announcement
April 2025 update on model comparison, error modes and SEAL benchmarks
Mayo Clinic Partnership Story
How Scale and Mayo Clinic built clinical AI, including the 11 minute result
New CEO Announcement
July 2026 notice of Francis deSouza as CEO and the company’s stated priorities
What to Know
What to Know Before Buying
-
Pro and Nucleus accounts have an Engagement Manager to help set up projects -
Custom domain workflows may need extra support engagement
-
Ask for pilot batches with agreed quality and turnaround targets
-
The GenAI Platform lists connectors for Confluence, SharePoint and S3 and deployment in your VPC on AWS, Azure or GCP. Data Engine work runs through API and project setup, so confirm formats and delivery methods.
-
Scale says the Data Engine scales from low-volume experiments to high-volume production. Costs grow with task volume and review passes, and a reviewer notes human review raises cost, so model spend as volume grows.
-
No rates are public. Ask about per-task pricing, rework and review charges, minimum commitments, platform fees, and whether Engagement Manager or Field Engineer coverage is included or extra.
-
Moving from another vendor or an in-house team means exporting labels, taxonomies, ontologies and guidelines. Ask for export formats, data deletion terms and what happens to your data if ownership of Scale changes.
-
Scale publishes no term, renewal, SLA or cancellation terms. Given Meta’s 49% stake, confirm data confidentiality, isolation from other customers, audit rights and exit terms in writing.
Alternatives
Compare Alternatives
See how Scale AI stacks up against competitors
Labelbox’s pricing page lists Free, Starter and Enterprise plans, with Enterprise listing dedicated technical support, a shared Slack channel and an SLA, and its homepage cites an Alignerr network of 2.6M+ experts and SOC 2 Type II. Scale publishes no plans or SLA, so compare written support terms.
Snorkel AI describes expert data development for frontier AI, with Expert Data-as-a-Service, curriculum-structured datasets, custom evals and benchmarks, and shows SOC 2 and HIPAA logos. Its site lists no prices or dedicated support offering, and it is narrower than Scale’s labeling plus GenAI and government range.
Get Started
Ready to Get Started?
Get personalized recommendations from top 3 vendors
By submitting, you agree to receive recommendations from verified vendors. No spam, unsubscribe anytime.
Are You a Software Vendor?
List your company and get in front of decision-makers actively searching for solutions.
FAQ’s
Frequently Asked Questions
Common questions about Scale AI
Scale publishes no prices, plans or free tier, and its pricing page is a demo request. Costs are quoted per engagement, and a G2 reviewer says pricing can add up quickly on larger labeling projects with human review. Request an itemized quote.
Scale’s support page lists 9 AM to 5 PM Pacific, Monday to Friday, excluding major US holidays, and says Pro and Nucleus customers contact an assigned Engagement Manager or Field Engineer. It lists no response times or SLAs, so get them in writing.
In June 2025 Meta took a 49% non-voting stake for about $14.3 billion, and Alexandr Wang joined Meta. Interim CEO Jason Droege said Scale stays independent and model agnostic. Reuters reported Google planned to cut ties, and OpenAI said it was winding down work with Scale, saying Meta was not the reason.
Scale’s GenAI Platform page lists SOC 2 Type II, ISO 27001, DoD IL4 and FedRAMP High. We could not load a trust page, so request the reports. Business Insider reported in June 2025 that Scale left client files in public Google Docs.
Yes. Former workers sued in December 2024 and January 2025, alleging misclassification and underpayment, and Scale disputes the claims and says it complies with the law. The Department of Labor probe was reported in March 2025 and dropped in May 2025, per TechCrunch.
Scale lists text, image, video and 3D sensor fusion including LiDAR, plus RLHF, red teaming and evaluation. Request sample batches to test quality.
Francis deSouza became CEO on August 10, 2026, per Scale. He was previously COO of Google Cloud and CEO of Illumina. Jason Droege, interim CEO since June 2025, is helping with the transition.
Common alternatives include Labelbox, Snorkel AI, Surge AI, Encord and SuperAnnotate.