Reflection AI has taken the wraps off Beam, marking the company's first entry into frontier, open-weight AI models. According to the two-year-old startup, Beam delivers performance comparable to top Chinese open models on sophisticated reasoning benchmarks—but at a much lower cost. This claim could intensify the competition to create a Western counterpart to models like DeepSeek, Qwen, and Z.ai.

The announcement aligns with recent reporting that the company was preparing to launch imminently. In a detailed blog post published Monday, Reflection revealed that Beam is a text-only mixture-of-experts model. It was trained using high-compute reinforcement learning, specifically designed to excel at reasoning, coding, and agentic tasks while consuming "a fraction of the token cost and inference time compute" compared to competing models.

Beam features 501 billion parameters, with 23 billion active at any given time. It was pre-trained on 23.8 trillion tokens and supports a context window of 1 million tokens. For context, Z.ai's GLM 5.2 has approximately 744 billion total parameters with 40 billion active.

While Reflection's performance claims have not been independently confirmed, the company states that Beam achieves scores on par with Z.ai's GLM-5.2 on advanced reasoning benchmarks and surpasses current leading Western open models—all while using "3-4x less inference compute." Reflection describes Beam as a "workhorse model" tailored for enterprises, the public sector, and developers.

Reflection positions itself in contrast to closed labs such as Anthropic and OpenAI, as well as popular open models from Chinese developers, and Western competitors like Mistral, Meta, and Cohere. Its closest U.S. rival may be Inkling, the open model from Mira Murati's Thinking Machines Lab, which launched in July. Reflection's benchmarks indicate that Beam outperforms Inkling on four coding tests where both models report results. However, Inkling is multimodal, whereas Beam is text-only.

Founded in 2024 by two former Google DeepMind researchers, Reflection has raised approximately $4.7 billion from investors including Nvidia, Sequoia Capital, and Lightspeed Venture Partners, according to PitchBook. The company's most recent funding round valued it at a $25 billion pre-money valuation.

Reflection has also been securing compute—a critical resource for training frontier models that can attract customers away from Anthropic and OpenAI's closed models, as well as the more affordable open-weight models from Chinese labs. This summer, Reflection signed agreements collectively worth over $7 billion with SpaceX and Nebius to guarantee access to Nvidia's GB300 chips through 2029.

Beam and future models from Reflection are aimed at enterprises and sovereign nations. The company's pitch involves building "AI factories," a product that would enable institutions to create their own customized, local AI systems by training Reflection's models on their proprietary data. Nvidia CEO Jensen Huang, whose company is an investor in Reflection, has long advocated for the "AI factory" concept and has pushed to strengthen the open AI ecosystem—a vision that would also benefit Nvidia, whose GPUs would power those systems. According to Axios, hedge funds and trading firms are among those interested in building such systems. Reflection has already started testing the sovereign AI factory partnership concept with Shinsegae Group in South Korea.

Reflection plans to release Beam's weights and full technical details this month, with distribution through hyperscalers and neoclouds, and integrations across open source libraries at launch.