Introducing Beam: Reflection’s 501B open-weight model — Reflection

Reflection has introduced Beam, its first open-weight Mixture-of-Experts model featuring 501 billion total parameters and 23 billion active parameters. Built through extensive pretraining and massive reinforcement learning using 10.5K NVIDIA GB300 GPUs, Beam is optimized for coding, reasoning, and agentic workloads. The model delivers competitive performance comparable to larger open models while offering exceptional inference compute efficiency.