Reflection Launches Beam: Open-Weight Model Challenging Chinese AI at Reduced Costs

Reflection AI is officially unveiling Beam, its flagship frontier, open-weight AI model. The two-year-old startup asserts that Beam matches the performance of leading Chinese open models on advanced reasoning benchmarks at dramatically reduced costs, a claim that could intensify the race to develop a Western alternative to DeepSeek, Qwen, and Z.ai.

This announcement corroborates reporting from Axios over the weekend regarding the startup's imminent launch. In a detailed blog post shared on Monday, the company described Beam as a text-only mixture-of-experts model. It was trained using high-compute reinforcement learning to excel at reasoning, coding, and agentic tasks, operating at “a fraction of the token cost and inference time compute” compared to competitors.

Beam is a 501-billion-parameter model featuring 23 billion active parameters. It was pre-trained on 23.8 trillion tokens and boasts a 1-million-token context window. For context, Z.ai's GLM 5.2 features approximately 744 billion total parameters with 40 billion active.

While Reflection's performance claims have not been independently verified, the company reports that Beam scores on par with Z.ai's GLM-5.2 on advanced reasoning benchmarks. It also outperforms today's leading Western open models while consuming “3-4x less inference compute.” Reflection positions Beam as a “workhorse model” for enterprises, government agencies, and developers.

Reflection is positioning itself against closed labs such as Anthropic and OpenAI, as well as popular open models from Chinese developers and Western players like Mistral and Meta. Its most direct American rival is potentially Inkling, the multimodal model released in July by Mira Murati's Thinking Machines Lab. However, Reflection’s benchmarks indicate that Beam outperforms Inkling on four coding tests, though it is important to note that Inkling is multimodal while Beam is text-only.

The startup was founded in 2024 by two former Google DeepMind researchers and has raised approximately $4.7 billion from investors including Nvidia, Sequoia Capital, and Lightspeed Venture Partners, according to PitchBook. Its last funding round valued the company at a $25 billion pre-money valuation.

The company has also secured significant compute resources, a critical component for training frontier models that can lure customers away from closed solutions like those from Anthropic and OpenAI, or cheaper open-weight models from Chinese labs. During the summer, Reflection signed contracts collectively worth over $7 billion with SpaceX and Nebius to gain access to Nvidia's GB300 chips through 2029.

Reflection is targeting Beam and future models toward enterprises and sovereign nations with a pitch to build “AI factories.” This product would allow institutions to construct their own customized, local AI systems by training Reflection's models on proprietary data. Nvidia CEO Jensen Huang, whose company backs Reflection, has long championed the “AI factory” concept and the strengthening of the open AI ecosystem, a vision that would also benefit Nvidia by utilizing its GPUs to power these systems.

Axios reported that hedge funds and trading firms are among the entities eager to deploy such systems. Reflection has already begun testing the concept of a sovereign AI factory partnership with the Shinsegae Group in South Korea.

Reflection stated that it will release Beam's weights and full technical details this month. Distribution will occur through hyperscalers and neoclouds, accompanied by integrations across open source libraries at launch.

Reflection did not provide a response in time to TechCrunch's requests for additional information.