AI Image
Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost
TechCrunch • • business
Key Points:
- Reflection AI has officially launched Beam, a 501-billion-parameter open-weight AI model with 23 billion active parameters, claiming performance on par with leading Chinese models like Z.ai’s GLM-5.2 but at significantly lower computational costs.
- Beam is a text-only mixture-of-experts model trained on 23.8 trillion tokens, featuring a 1 million token context window, and is designed for advanced reasoning, coding, and agentic tasks with 3-4 times less inference compute than rivals.
- The Brooklyn-based startup, founded in 2024 by former Google DeepMind researchers, has raised about $4.7 billion and secured over $7 billion in compute deals with SpaceX and Nebius to access Nvidia’s GB300 chips through 2029.
- Reflection aims to serve enterprises and sovereign nations by enabling "AI factories" that allow customized, local AI systems trained on proprietary data, with early partnerships including South Korea’s Shinsegae Group.
- Beam’s full technical details and model weights are set to be released this month, with distribution planned through hyperscalers, neoclouds, and integration with open-source libraries.
