A New York AI startup founded by two former Google DeepMind researchers has released its first major model — and is pitching it as a homegrown answer to China’s increasingly powerful open AI systems.
Reflection AI on Monday unveiled Beam, an “open-weight” model designed for coding, reasoning, and AI-agent tasks. The company says it rivals some of China’s leading open models while using far less computing power.
Here’s what to know.
What is Beam?
Beam is Reflection AI’s first frontier open-weight model. Open-weight means developers can download the model’s parameters, run them on their own hardware, and modify or fine-tune them for specific use cases.
That is an alternative to using models from companies such as OpenAI and Anthropic, which are generally accessed via their own Application Programming Interfaces (APIs).
Reflection says Beam, a text-only model, has 501 billion total parameters, but activates only 23 billion at a time — a design intended to make it cheaper and faster to operate. The model can handle coding, multi-step reasoning, and other agentic work.
For now, Beam is in final safety testing and available through a waitlist. Reflection says it will release the model’s weights, a technical report, and developer tools later this month.
Who is behind Beam?
Reflection was founded in 2024 by Misha Laskin and Ioannis Antonoglou, both former DeepMind researchers.
“The only way to own intelligence is, by definition, if it’s open — because if it’s open, you can run it on your stuff, you can own it, you can control it, you can customize it,” Laskin, the CEO, said in a podcast released on Monday.
The startup has raised billions of dollars and is backed by investors including Nvidia, Sequoia, and Citigroup. In June, it confirmed a funding round at a $25 billion pre-money valuation.
How does Beam compare?
Reflection said Beam is competitive with Z.ai’s GLM-5.2 on coding and agentic tasks, and is approaching Alibaba’s Qwen 3.8-Max. It said that Moonshot AI’s Kimi K3 remains ahead on raw capability.
The startup is pitching itself on efficiency. It said Beam can achieve results comparable to GLM-5.2 on advanced reasoning benchmarks while using three to four times less inference compute — the processing required for a model to answer a prompt.
How are tech and business leaders reacting?
Beam’s launch is being welcomed as a potentially important US response to China’s open-model lead. It arrives in a crowded field that includes open models from Meta, Mistral, Nvidia, and Thinking Machines Lab.
Beam also faces obvious competition from powerful models released in China by companies including DeepSeek, Alibaba, Z.ai, and Moonshot AI.
The Wall Street Journal reported in March that some investors had described Reflection as the “DeepSeek of the West.” That comparison to DeepSeek has become the shorthand used in many X posts surrounding Beam’s buzzy launch.
On Monday, Nvidia AI’s official X account congratulated the company on the launch. Investors, including Deedy Das and
Shaun Maguire, also expressed their enthusiasm for the launch on X.
