Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost
techcrunch ·
Reflection AI has officially launched Beam, its first frontier open-weight model designed to compete with leading Chinese alternatives like DeepSeek and Qwen. The startup claims that Beam matches the performance of Z.ai’s GLM-5.2 on advanced reasoning benchmarks while utilizing three to four times less inference compute. This 501-billion-parameter mixture-of-experts model is trained specifically for reasoning, coding, and agentic tasks at a fraction of the token cost of rivals. Although these performance claims have not been independently verified, Reflection positions Beam as a cost-effective alternative for enterprises and developers. The company aims to challenge both closed labs like OpenAI and other Western open-source players such as Mistral and Meta. To support this ambitious goal, Reflection secured over $7 billion in compute deals with SpaceX and Nebius through 2029. The startup plans to release Beam's weights and technical details later this month via hyperscalers and open-source libraries.
techcrunch