Reflection AI introduced Beam on October 5 as its first open-weight model, built for coding, reasoning and agentic workloads. The company says the sparse mixture-of-experts system has 501 billion total parameters, with 23 billion active per token, and was pretrained on 23.8 trillion tokens.
Early access now, weights promised later this month
Beam is available only to a select early-access group while final red-teaming and evaluations continue. Reflection says it plans to release the weights under Apache 2.0, along with a technical report, model card and developer artifacts, later in October.
Reflection reports training Beam across 6,144 NVIDIA GB300 GPUs and running high-compute reinforcement learning on 10,500 GB300 GPUs. Its benchmark and efficiency comparisons are company-run claims; Reuters corroborated the announcement, but independent reproduction was not located in this review.