Prime Intellect Open Lab Platform: Agent post-training is made into a closed loop, billed by token rather than GPU hours

CryptoWorld News reports that Prime Intellect has announced that its Agent Post-Training Platform Lab has transitioned from beta to a full release. The platform integrates evaluation, reinforcement learning (RL) training, adapter deployment, and inference into a closed loop, allowing users to define tasks and scoring criteria. The platform automatically drives the model to repeatedly trial and error in tasks, collect reward signals, and train LoRA adapters. Training is billed per token rather than GPU hours, based on the company’s open-source Prime-RL framework. The first batch of supported models includes 14 models from NVIDIA, OpenAI, Meta, and Qwen, with parameters ranging from 1 billion to 70 billion, covering dense and MOE architectures. Prime Intellect was founded in 2023, with a total funding of over $70 million. Series A was led by Founders Fund, and Series B was led by Radical Ventures.