Get early access to environment-free training (EFT)

Reinforcement learning (RL) has proven to be the most effective way to post-train large language models (LLM) on agentic tasks. But RL needs simulated environments, which are expensive and take time to build, hard to test, and sometimes impossible to model accurately.

To solve this, we are introducing environment-free training (EFT) in our Serverless RL offering, eliminating the need to build simulated environments altogether. EFT enables agents to learn through continual “on-the-job” training and improve reliability.