Woofun AI reports that Prime Intellect has released Verifiers 0.2.0, featuring a preview of the next-generation v1 architecture, alongside the open-sourcing of the prime-RL model training framework. Verifiers serves as an open-source system for presenting questions, executing tasks, and scoring AI agents, while prime-RL handles model training based on task outcomes.

The v1 architecture decouples tasks from agent operations into three distinct components: Taskset, Harness, and Runtime. This modular design allows developers to switch between agents like Codex and Kimi Code or change execution environments without rewriting rules. The update supports tracking sub-agent invocations and context compression for reinforcement learning, with future plans to support multi-agent environments and frameworks such as OpenEnv.