Models · Updated 6 Oct, 05:41 pm IST
Reflection debuts Beam, a 501B open-weight model focused on efficient coding and agents
Why it matters for readers: Beam is a new large AI model that aims to give good coding and agent behavior while using less computing power than some rivals.
- Beam is a sparse mixture-of-experts model with 501 billion total parameters and 23 billion active parameters per token to lower runtime compute.135
- Reflection positions Beam for coding, logical reasoning and agentic tasks and targets enterprises building private, customized AI systems.125
- Reflection says Beam matches or approaches performance of Chinese open models such as GLM‑5.2 and Qwen3.8‑Max on coding and agentic benchmarks while using less compute.135
- The startup is backed by Nvidia and was founded in 2024 by former DeepMind researchers Misha Laskin and Ioannis Antonoglou.3
- Reflection plans to release Beam’s weights, technical report, model card, and developer artifacts shortly after final red-teaming and evaluations.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get startedHow this story developed
6 Oct, 12:46 am · Hacker News (100+ points)
Beam: Reflection's 501B open-weight model
6 Oct, 01:03 am · TechCrunch AI
Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost
6 Oct, 08:47 am · ET Tech
Nvidia-backed Reflection unveils first AI model to take on Chinese open models
6 Oct, 09:14 am · Indian Express Technology
Nvidia-backed Reflection unveils first AI model to take on Chinese open models
6 Oct, 05:41 pm · The Decoder
Reflection's Beam becomes the most capable open-weight model built outside China