Reflection AI releases Beam model achieving high ranking on open index

Reflection AI has released its first artificial intelligence model, called Beam, which reached second place among open-weight models from United States laboratories on the Artificial Analysis Intelligence Index. OfficeChai reported that the New York-based laboratory was founded by two former Google DeepMind researchers. [1]

The Beam model scored 30 on the Intelligence Index managed by Artificial Analysis, OfficeChai reported. The index is a composite evaluation that measures model capabilities across ten areas including agentic tasks, coding, science, and general knowledge. OfficeChai noted that Beam ranks behind only Meta's Muse Spark Open among open-weight models built in the United States. [1]

As FTMQ SI reported earlier, Beam is a 501-billion parameter sparse Mixture-of-Experts model that uses 23 billion active parameters per token. The model matches GLM-5.2 reasoning performance while requiring three to four times less inference compute. Reflection AI trained Beam on rented Nvidia GB300 NVL72 hardware and released the open weights under an Apache 2.0 license on Hugging Face. [10][11]

The release comes as international laboratories continue to deploy sparse architecture models. MarkTechPost reported that Chinese firm StepFun released open weights for its Step 5 Preview model, a 600-billion parameter Mixture-of-Experts architecture with 27 billion active parameters per token and a context window of up to 1 million tokens. [4]