Reflection AI CEO on Beam, Open-Source Models, and the U.S.-China AI Race

Reflection AI CEO Misha Laskin discusses the release of Beam, an open-source model positioned as a U.S. alternative to Chinese open-weight models such as DeepSeek. The interview covers model capabilities, training approach, competitive dynamics with OpenAI and Anthropic, and implications for enterprise AI systems.

■ Beam and the U.S.-China AI race
- Release of open-source model Beam and positioning relative to Chinese open-weight frontier
- Focus on reasoning efficiency and coding/agentic capabilities

■ Development and competition
- Trained from scratch with reinforcement learning; rocket-ship analogy for faster iterations
- Distillation attacks, Chinese competitors, and U.S. open-source ecosystem examples like Meta's Muse Spark

■ Enterprise AI systems
- AI factory and end-to-end systems including hardware, cluster management, inference software
- Margins across the stack and demand for agentic inference

This interview suits viewers tracking AI industry competition, open-source model development, and enterprise AI adoption. It provides an overview of how one CEO frames the U.S. open-source position relative to China and the commercial landscape. Watch to understand the strategic context rather than detailed benchmarks or pricing.