Laminar releases flow-1, an RL-trained model for finding errors in agent traces

Laminar has released flow-1, a model trained with reinforcement learning to find errors in AI agent traces, including silent failures.

The company says that on its own benchmark, flow-1 matches GPT-6-sol in detection quality at 23x lower cost. Laminar says the lower cost makes it practical to analyze every production trace instead of a sample. The benchmark results have not been independently verified.