Reflection Beam: Get Agent-Ready Before the Weights Drop

Reflection AI released Beam on October 5: a 501B-parameter sparse mixture-of-experts model with 23B active parameters per token, aimed at coding and agentic work. It is the first big American open-weight model in a while, and the announcement says the weights ship under Apache 2.0 later this month. Developer chatter has been loud, mostly around one question: is this the US answer to GLM and Kimi?

Here is the catch that most summaries skip. As of the announcement, Beam is available only through a waitlisted beta API, and the weights are not downloadable yet. So the useful thing to do today is not to benchmark it. It is to get your agent code ready, so that when access or weights arrive you change one environment variable and go.