"DeepSeek of the West" is about to break through the circle, pointing its sword at China, with Huida's support behind it

Reflection AI, a new artificial intelligence company supported by Huida, is participating more deeply in the global artificial intelligence competition with its first open-weight model (open-weight) Beam. This model is designed to compete with low-cost Chinese AI systems such as DeepSeek and Kimi in terms of coding and AI agent tasks, and has therefore been described by some investors as the "DeepSeek of the West."

Reflection AI released Beam on Monday, an open weight model designed for coding, reasoning and AI agent tasks. The company says Beam's performance is already comparable to some of China's leading open source models.

The New York-based artificial intelligence startup, founded by two former Google DeepMind researchers, is launching its first major model and is positioning Beam as the U.S.’s homegrown answer to China’s increasingly powerful open AI systems.

Beam is the first cutting-edge open weight model launched by Reflection AI. The so-called open weight model means that developers can download the model weights, run them on their own hardware, and modify or fine-tune them for specific application scenarios.

Such models also provide an alternative to the closed models offered by companies such as OpenAI and Anthropic. The latter usually provide services through their respective application programming interfaces (APIs), and users cannot directly obtain model weights.

Reflection AI pointed out that Beam is a pure text model with a total of 501 billion parameters, but only about 23 billion parameters are activated each time. This design aims to reduce operating costs and improve computing efficiency and speed. Beam can perform program coding, multi-step reasoning, and other AI agent tasks.

Currently, Beam is still undergoing final security testing, and users can apply for use through the waiting list. Reflection AI said it expects to release model weights, technical reports and developer tools later this month. The startup has raised billions of dollars in funding from investors including Huida, Sequoia Capital and Citigroup. In June this year, Reflection AI confirmed the completion of a round of financing, with a pre-money valuation of US$25 billion.

Reflection AI stated that Beam’s performance in coding and AI agent tasks is comparable to Z.ai’s GLM-5.2, and is gradually approaching Alibaba’s Qwen 3.8-Max. However, the company also pointed out that in terms of overall raw performance, Moonshot AI's Kimi K3 is still in the lead.

The release of Beam is seen as a potentially important response from the U.S. AI industry to China's rapid rise in the field of open AI architecture. Beam has also joined the ranks of open weight model competition launched by companies such as Meta, Mistral, Huida and Thinking Machines Lab, further intensifying market competition.

At the same time, Beam is also facing strong competition from Chinese AI companies, including models launched by companies such as DeepSeek, Alibaba, Z.ai and Moonshot AI, which have established a certain influence in the open AI field.