Reflection

Latest news

1. 6 Oct
   Reflection debuts Beam, a 501B open-weight model focused on efficient coding and agents

   Reflection launched Beam, a sparse mixture-of-experts open-weight model with 501 billion total parameters that activates 23 billion per token to reduce inference compute. The Nvidia-backed startup says Beam is aimed at coding, reasoning and agentic workloads and is competitive with some Chinese open models while using less compute.

   Models · 5 sources

   

Articles

No articles about Reflection yet.