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

Reflection released Beam, its first open-weight model, a mixture-of-experts system that activates 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. Reflection positions the efficiency-over-raw-performance strategy against Chinese open-weight rivals Deepseek and Qwen, making Beam 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 most capable open-weight model built outside China https://the-decoder.com/reflections-beam-becomes-the-most-capable-open-weight-model-built-outside-china/ appeared first on The Decoder https://the-decoder.com .