EPAM launches data and evaluation service for frontier AI models

EPAM Systems announced a new service for developing the data and evaluation capabilities frontier AI models need to handle complex enterprise workflows more reliably. The company said the work will include specialized data generation, model evaluation and custom reinforcement-learning environments. The offering is intended to help models move beyond general language and coding toward specialized business tasks.

The focus reflects changing demands as generative AI adoption grows. EPAM’s announcement says earlier models were trained on broad public data, while newer enterprise use cases involve multi-step agent execution and domain-specific reasoning. The company plans to apply its experience building and maintaining complex workflows to produce vetted domain knowledge for that work. It described the service as addressing a reliability bottleneck in using frontier models for specialized business processes.

A central part of the offering is the development of custom simulation environments for reinforcement learning. EPAM says these virtual settings will reproduce complex enterprise systems and workflows. Developers can use them to test multi-turn interactions, reasoning and tool use before production. The environments are also designed to provide secure, closed-loop feedback for reinforcement learning and ongoing model improvement. For AI builders, this means evaluation can include simulated enterprise-agent tasks before deployment, alongside broader measures of model capabilities.

EPAM said the service builds on relationships with Anthropic, OpenAI, Google and Microsoft. The company reported nearly 10,000 Claude-certified architects, 3,000+ OpenAI-certified forward-deployed engineers and 5,000+ Gemini-certified specialists. EPAM said its experience across those model ecosystems and enterprise deployments can support work with labs and help businesses move from pilots toward measurable outcomes.

The announcement cited an April 2026 Gartner report that projected wider adoption of simulation by agent platforms. The report forecast that 99% of agent platform providers would offer simulation environments by 2028, up from less than 25% in 2026. The cited forecast concerns platform-provider offerings, while EPAM’s launch focuses on services for developing and assessing models for specialized enterprise work.