Reflection unveils Beam, a 501-billion-parameter open-weight model that it says rivals GLM-5.2 with less compute

Reflection AI, a two-year-old startup based in Brooklyn, has officially unveiled Beam, its first frontier open-weight model. The announcement confirms weekend reports that a launch was close. Beam is a text-only mixture-of-experts model with 501 billion total parameters and 23 billion active. It was pretrained on 23.8 trillion tokens and has a 1 million token context window. Reflection says it trained the model with high-compute reinforcement learning for reasoning, coding and agentic tasks, and it plans to publish the weights under an Apache 2.0 license later this month. The company says Beam matches Z.ai's GLM-5.2 on advanced reasoning benchmarks while using three to four times less inference compute. These claims have not been independently verified. Help Net Security notes that the compute figure comes from a formula, twice the active parameters times the average number of generated tokens, which leaves out prompt processing, attention costs and serving overhead. On Terminal Bench v2.1, Beam scored 80.1, compared with 81.0 for GLM 5.2, 88.3 for Kimi K3 and 90.6 for DeepSeek V4.1 Flash. On DeepSWE v1.1 it scored 44.4, just above GLM 5.2 and 29.8 points behind DeepSeek V4.1 Flash. Reflection pitches Beam as a workhorse model for enterprises, the public sector and developers.