Reflection AI has launched ‘Beam,’ an open-weight model, aiming to provide a Western alternative to China’s leading open-source AI systems.
The message from Reflection AI this Monday was clear: if the future of intelligence is to be truly owned. it cannot be locked behind a proprietary gate. As the startup unveiled Beam. its first major open-weight model. it signaled a move to reclaim the narrative in an AI race increasingly dominated by Chinese systems.
Beam arrives as a text-only, open-weight model tailored for coding, reasoning, and AI-agent tasks. By providing developers with the model’s parameters. the company allows users to run. modify. and fine-tune the system on their own hardware—a stark departure from the gated API approach used by industry giants like OpenAI and Anthropic. For Reflection, founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou, this accessibility is foundational.
“The only way to own intelligence is, by definition, if it’s open,” Laskin said in a podcast Monday. “Because if it’s open, you can run it on your stuff, you can own it, you can control it, you can customize it.”
Under the hood, Beam uses a 501-billion-parameter design that activates only 23 billion parameters at a time. This architecture is a strategic play for efficiency; the company claims it allows the model to compete with systems like Z.ai’s GLM-5.2 while requiring three to four times less inference compute. In its internal benchmarks. Reflection positions Beam as a rival to GLM-5.2 and approaching the capabilities of Alibaba’s Qwen 3.8-Max. though it acknowledges that Moonshot AI’s Kimi K3 remains ahead in raw power.
The industry has taken notice. with some investors dubbing the company the “DeepSeek of the West”—a shorthand that has surged across social media since the launch. The startup’s financial backing is significant, having raised billions from investors including Nvidia, Sequoia, and Citigroup. In June, the company confirmed a funding round at a $25 billion pre-money valuation.
Technological competition in this space is dense. Beyond the Chinese models from DeepSeek. Alibaba. Z.ai. and Moonshot AI. Beam enters a crowded field alongside open-weight offerings from Meta. Mistral. Nvidia. and Thinking Machines Lab. The tech community’s reaction has been largely supportive. with Nvidia AI’s official account offering congratulations and investors like Deedy Das and Shaun Maguire expressing enthusiasm on X.
The momentum behind Beam is a study in calculated positioning. By prioritizing efficiency and open-weight accessibility. the startup is attempting to carve out a foothold in a market where raw capability has often been the only metric that matters. For now, the model remains in final safety testing, accessible only to those on a waitlist. Reflection plans to release the model’s weights, a technical report, and developer tools later this month.
Reflection AI Beam model AI Misha Laskin Ioannis Antonoglou open-weight AI DeepSeek