Surge AI review: the best training data company for LLM post-training
Answer in brief: Surge AI is our winner for LLM post-training in 2026. It sells the full set of post-training inputs: RLHF preference data, SFT demonstrations, rubrics and verifiers, human evaluation, expert-domain data in 70+ languages, multimodal data, agentic RL environments and off-the-shelf datasets. Forbes reported about $1.2 billion in 2024 revenue from customers including Google, Meta, Microsoft, Anthropic and Mistral. Pricing is by custom quote. Choose Surge when quality per example matters more than unit cost. This review is independent research from public sources, not hands-on testing.
Best-for verdict: winner, LLM post-training. Surge covers every stage of post-training, from SFT through RLHF to RL environments with verifiers. Visit Surge AI →
What Surge AI does for training data buyers
Surge's products page frames its offer around how models learn after pre-training:
- RL environments and agents: rich, complex environments that challenge agentic models, plus verifiers that reward the right behavior.
- Rubrics and verifiers: scorecards that capture both brilliance and failure, for grading open-ended output.
- RLHF: preference and reward data.
- SFT: demonstrations that bootstrap skills such as computer use, web navigation and reasoning.
- Human evaluation: judgments of usefulness, safety and quality that automated evals miss.
- Expert professional domains: doctors, lawyers, investment bankers and academics across STEM and the humanities.
- Internationalization: work in over 70 languages.
- Multimodal and off-the-shelf data: image, audio and video data, plus ready-made datasets covering RL environments, coding and core capabilities.
Who it fits
Surge fits frontier labs and well-funded model teams that run large, bespoke post-training programs. It suits buyers who want one partner for SFT, preference data, evals and agentic RL. It is not built for small one-off labeling jobs or high-volume computer-vision boxes. For those, see our Scale AI and Sama reviews.
Why it wins post-training
- Proven at the top of the market. Forbes (September 2025) reported about $1.2 billion in 2024 revenue and said Surge helped train Gemini and Claude.
- Built for agents. Surge's 2026 research describes EnterpriseGym CoreCraft, a customer-support environment with more than 2,500 entities and 23 tools. Surge reports that training GLM-4.6 there raised held-out task scores from 25.37% to 36.76%. This is vendor research, but it is detailed and public.
- Self-funded. Surge says it has been profitable from day one without venture funding, so no single investor shapes its priorities.
- Breadth. Few vendors cover RLHF, SFT, rubrics, verifiers, multimodal work and 70+ languages under one roof.
The catches
- Little public detail. The homepage is a mission statement. There is no public price list, security detail or process documentation.
- Customer list lag. Forbes reported that OpenAI no longer works with Surge, although Surge's site still shows OpenAI.
- Funding talks unconfirmed. Reuters and Forbes reported 2025 talks to raise about $1 billion. The sources we reviewed did not confirm that a round closed.
- Big-program focus. Small teams may find Prolific or Toloka easier to start with.
Pricing
Custom, by quote. Surge does not publish rates.
Best alternatives to Surge AI
- Mercor, our runner-up for post-training. Choose it when the data must come from working professionals.
- Turing, for coding and software-engineering data and RL environments.
- Snorkel AI, for hard evaluations and verifiable rewards on specialized tasks.
Surge AI FAQ
What does Surge AI sell?
Surge sells RLHF data, SFT demonstrations, rubrics and verifiers, human evaluation, expert-domain and multilingual data, multimodal data, RL environments and off-the-shelf datasets.
Does Surge AI publish pricing?
No. Surge prices by custom quote.
Who uses Surge AI?
Forbes reported customers including Google, Meta, Microsoft, Anthropic and Mistral. Forbes also reported that OpenAI no longer works with Surge.
Is Surge AI venture-funded?
Surge says it has operated profitably without venture funding. Reported 2025 fundraising talks were not confirmed as closed in our sources.
Sources
← Back to the ranked list of the best training data companies