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

Reflection AI, a startup positioning itself as America's answer to open-source Chinese AI, has unveiled its first model called Beam. Founded in 2024 by former DeepMind researchers, the company aims to create a "workhorse" model that outperforms other Western open models. Co-founder and CEO Misha Laskin stated that Beam excels at coding and agentic tasks, with its computational efficiency being three to four times lower than comparable open models.

Beam's performance is on par with China's Z.ai's GLM-5.2, released in June, and is approaching the capabilities of Alibaba's most advanced model, Qwen 3.8-Max. While open models enable cost reductions for repetitive tasks, they also pose security concerns due to their accessibility. However, Reflection sees a significant market opportunity in businesses, governments, and entities seeking to build sovereign AI systems that aren't reliant on Chinese models.

The startup, backed by Nvidia, Sequoia, and Citigroup, recently closed a funding round worth $25 billion. Beam's release comes amid growing calls for AI regulation in the US, following alarming security incidents. Reflection's CEO believes that the model will enable builders of open models to contribute to the ongoing discourse on AI regulation in DC, fostering collaboration between open and closed model proponents.