Laminar Launches at YC With flow-1, a Trace-Analysis Model That Matches GPT-4o at 23x Lower Cost

YC S24 startup Laminar launched flow-1, a reinforcement-learning-trained model that reads agent traces to catch silent failures — cases where every tool call succeeds but the agent clicked the wrong button or cited the wrong document. The team says flow-1 matches GPT-4o detection quality at 23x lower cost, enabling full-trace analysis without sampling. The observability platform ships as Apache 2.0 and self-hostable, with managed cloud as the business model.