Prime Intellect had to invent Hierarchical GRPO because classic GRPO can't express role-aware credit.

The August 7 PRIME-RL update makes Agent and Env first-class, where Env.run(task, agents) programs arbitrary multi-agent control flow and every finished agent run auto-joins the Episode, letting you pick which roles learn (Prime Intellect). Four environments ship: Agentic Judging, User-Sim, Proposer-Solver, Kuhn Poker self-play. The credit-assignment problem is the transferable part, and anyone doing multi-agent RL will hit it: solver attempts must compare against attempts on the same proposed problem while proposer traces compare against proposals from the same seed, a hierarchy plain GRPO cannot represent. Proposer-Solver reward is calibrated so learnability peaks at a 50% solve rate.