Very real In a sense all these open models are still proofs of concept. It's not so much about compute (though it matters bigly) as about work-centuries that went into the data, the reward models, and the EVALS. "RL environments" is a cute new gimmick. Smaller in scale.