Prime Intellect Launches Prime Inference: Serverless and Reserved Serving for Frontier Open Models

תקציר מקורי באנגליתPrime Intellect has launched Prime Inference , a serving platform for frontier open-source models. It offers serverless endpoints and reserved capacity on Prime’s own GPUs across multiple datacenters. Before public release, it processed nearly a trillion tokens per day internally. That traffic came from RL rollouts, synthetic data generation, evaluations and long-running coding agents. What is Prime Inference? Prime Inference is the serving layer of Prime Intellect’s open training stack. The company already ships post-training tools such as prime-rl, verifiers and sandboxes. Serving closes that loop: deployed models generate production traces that can feed back into training. Prime reports its GLM-5.3 endpoint ranks among the fastest on OpenRouter. It also cites a near-zero too