Reflection AI launches Beam, an open-weight model for coding and AI agents

Reflection AI is introducing Beam, its first open-weight AI model, and says it is designed for coding and agent-based workflows. J.D. Capelouto and Ashley Gold report for Semafor that the New York startup positions the model as an alternative for organizations that want more control over their AI systems.

The company says Beam performs better than other Western open models in programming and agentic tasks. Such tasks involve AI systems that can plan and carry out multi-step work, rather than simply produce a single response. Chief Executive Misha Laskin tells Semafor that Beam needs three to four times less computing power for reasoning than comparable open models. That could reduce operating costs and speed up responses, though Reflection has not published independent benchmark results in the report.

Reflection says Beam’s capabilities are comparable to Z.ai’s GLM-5.2 model and are approaching Alibaba’s Qwen 3.8-Max. Both are Chinese models. Qwen 3.8-Max still trails leading proprietary systems from OpenAI and Anthropic, according to Semafor.

Focus on sovereign AI

Reflection sees an opening among companies, governments, and other organizations that want to operate AI independently. These users may seek models whose weights can be downloaded, adapted, and run in their own infrastructure. Some may also avoid Chinese models because of security, policy, or procurement concerns.

The move comes as Chinese companies continue to hold a strong position in open AI models. Western organizations have increasingly adopted such models for lower-cost work, including customer support and routine software development. At the same time, officials have raised concerns that openly available models can be misused. Semafor notes that U.S. and U.K. government evaluators found recent Chinese models could assist hackers in exploiting software vulnerabilities.

Reflection says it is already training a more capable successor to Beam. The company also says it is working with the U.S. Center for Advancing Innovation and Standards for Super Intelligence and the U.K. AI Safety Institute to assess Beam’s capabilities.