NY-based Reflection unveils Beam, an open model it claims rivals GLM 5.2 on reasoning while using 3x-4x less compute and approaches Qwen3.8-Max on agentic tasks

The cost to run advanced AI reasoning tasks using open models drops significantly, making powerful AI accessible for more applications.

New York-based startup Reflection AI has launched its first model, Beam, an open-source alternative to Chinese AI models. Founded by former DeepMind researchers, Reflection AI positions Beam as a high-performance "workhorse" model, particularly excelling in coding and agentic tasks. Beam reportedly requires 3x-4x less computing power for reasoning compared to similar open models, making it faster and more cost-effective to operate. Its performance is claimed to rival GLM-5.2 from China's Z.ai and approach that of Alibaba's Qwen 3.8-Max, though still trailing top closed models from OpenAI and Anthropic. The company aims to serve businesses and governments seeking to develop sovereign AI systems without relying on Chinese models, addressing security concerns associated with open-source AI. Beam's development is supported by investors including Nvidia, Sequoia, and Citigroup, and Reflection AI is already working on a more powerful successor model.

Beam's efficiency in compute requirements suggests a potential shift towards more accessible and cost-effective AI development for enterprises and researchers. This could democratize the use of advanced AI capabilities, enabling a wider range of organizations to build and deploy sophisticated AI solutions without prohibitive hardware or cloud expenses. The focus on agentic tasks and coding also indicates a move towards more practical, real-world applications of open-source AI.

Reflection AI's entry, backed by significant investment, signals increased competition in the open-source AI space, particularly as a Western counterpart to leading Chinese models. This development could influence the trajectory of AI regulation discussions by introducing more diverse voices and perspectives, advocating for the inclusion of open models in policy-making. The company's collaboration with US and UK AI safety institutes suggests a commitment to addressing security concerns proactively.