Reflection AI, a New York-based startup founded in 2024 by former DeepMind researchers, has released its first open-weight model, called Beam. The company positions Beam as a practical, general-purpose model that excels particularly in coding and agentic tasks, according to cofounder and CEO Misha Laskin.
Performance and compute efficiency
Laskin said Beam requires three to four times less computing power to reason through problems than comparable open models, which makes it cheaper and faster to run. Reflection compares Beam’s capabilities to Z.ai’s GLM-5.2, released in June, and says Beam is “approaching” the performance of Alibaba’s Qwen 3.8-Max, though those models still trail the most advanced closed American systems from OpenAI and Anthropic.
Market context and security concerns
Chinese startups and technology giants have so far led globally in open-source AI. Several large Western companies have adopted Chinese models for cost-sensitive, repetitive tasks such as customer service and routine code generation. Beijing has also promoted open source as a diplomatic tool to support other countries’ technology development. At the same time, open models’ relative ease of download and fewer built-in safeguards have raised security concerns: US and UK government evaluators found that recent Chinese models could assist malicious actors in exploiting software vulnerabilities.
Reflection sees demand from businesses, governments, and other organizations that want to build sovereign AI systems but cannot or do not want to use Chinese models and still want to retain control over how AI is used. “They don’t really have very good options today,” Laskin said.
Funding, competitors, and next steps
Investors in Reflection AI include Nvidia, Sequoia, and Citigroup; the company confirmed in June that its latest funding round closed at a $25 billion pre-money valuation. With Beam’s release, Reflection joins other Western startups offering open-weight models, such as Thinking Machines Lab and Mistral. Reflection is already training its next model, which Laskin said will be “much more” powerful and performant than Beam.
Regulation and external testing
Beam’s debut comes amid heightened calls for AI regulation in the US following several alarming security incidents. Leading US AI labs and tech companies signed what US President Donald Trump described as a “morally binding” agreement on safety principles after a White House meeting. Trump has advocated for industry self-regulation rather than federal oversight, though political dynamics could shift if Democrats regain control of Congress in November.
Reflection says it wants to enable builders of open models to contribute to the regulatory discussion in Washington. Laskin argued that multiple voices — both from open and closed systems — should collaborate with governments on regulating AI. Reflection also said it is working with the US Center for Advancing Innovation and Standards for Super Intelligence and the UK’s AI Safety Institute to assess Beam’s capabilities.
