Reflection AI launches Beam to challenge Chinese open-weight models

Beam operates as a mixture-of-experts model featuring 23 billion active parameters and a 1-million-token context window. While the company’s internal benchmarks suggest it achieves parity with top-tier Chinese models and outperforms leading Western open-weight alternatives, these performance claims await independent verification. Reflection intends to market the system as a "workhorse" for enterprises and the public sector, specifically targeting those seeking to build localized "AI factories" using proprietary data.

The venture, founded in 2024 by former Google DeepMind researchers, has secured $4.7 billion in funding from backers including Nvidia, Sequoia Capital, and Lightspeed Venture Partners. To support its infrastructure, the firm has finalized long-term compute deals worth over $7 billion with SpaceX and Nebius, ensuring access to Nvidia GB300 chips through 2029. Testing for sovereign AI partnerships is already underway, notably with South Korea’s Shinsegae Group. Reflection plans to release the model’s weights and technical documentation later this month via major hyperscalers and open-source libraries.