Laminar releases flow-1 model for agent observability

Laminar has launched flow-1, an open-source model trained via reinforcement learning to detect errors in agent traces. The model matches GPT-6-sol detection quality on internal benchmarks while operating at 23x lower cost.