Reflection put the weights of Beam on Hugging Face under Apache 2.0, eleven days after announcing it. Beam is a mixture of experts with 501 billion parameters, about 23 billion active per token, 52 layers, 256 experts and a 1M token context. Reflection claims comparable answers to leading open models with roughly three times fewer reasoning tokens. Independent scoring is less flattering: Artificial Analysis puts it 19th of 120 models. The cheapest hosted route is $0.25 per million input tokens. This was missed in the 20 October intake.