Reflection's Beam becomes the most capable open-weight model built outside China

Reflection has released Beam, its first open-weight model. The mixture-of-experts system activates just 23 billion of its 501 billion parameters per token and aims to match GLM 5.2 on coding and reasoning while using three to four times less compute. The strategy against Chinese open-weight rivals like Deepseek and Qwen is efficiency over raw performance. The article Reflection's Beam becomes the…
This is a summary curated by AIFuture. Read the complete article at the original source:
Read the full story on The Decoder