AI news ·
Reflection releases Beam 501B open weights under Apache 2.0
Reflection released open weights for a 501B-parameter AI model under Apache 2.0, activating 23B parameters for coding, reasoning, and agentic tasks.

Reflection has released the open weights for Beam 501B under an Apache 2.0 license. The model activates 23 billion of its 501 billion total parameters and is built to handle coding, reasoning and agentic workloads.
Making the weights openly available broadens the scope for experimentation across research, product development and enterprise evaluation. Teams can download and run the model directly, which removes certain access barriers but introduces practical requirements around infrastructure planning. Independent evaluations and appropriate safeguards remain necessary when deploying models of this size and capability.