Reflection AI unveils an open-source Western answer to Chinese labs

Reflection AI, founded by former DeepMind researchers, has released Beam, an open-weight model positioned as a Western alternative to recent Chinese open models. Beam is presented as a workhorse especially strong at coding and agentic tasks, and claims to use three to four times less compute for reasoning than comparable Western open models, making it faster and cheaper to run. Its capability level is said to be similar to Z.ai’s GLM-5.2 and approaching Alibaba’s Qwen 3.8-Max, though it trails top closed US models from OpenAI and Anthropic. Reflection is already training a successor that it says will be substantially more powerful than Beam.

Reflection is pitching Beam to businesses, governments, and other organizations that want sovereign, controllable AI without relying on Chinese models, arguing that current Western options are limited. The company has heavyweight investors including Nvidia, Sequoia, and Citigroup and confirmed a $25 billion pre-money valuation in June, joining other Western open-weight entrants like Thinking Machines Lab and Mistral. Its launch comes amid heightened regulatory pressure and security concerns about open models enabling exploits; Reflection is engaging with U.S. and U.K. safety bodies and seeks to amplify open-model perspectives in policy discussions.