Reflection AI has officially launched Beam, a frontier open-weight AI model the Brooklyn-based startup says matches leading Chinese open-weight models on advanced reasoning benchmarks while costing a fraction of the compute to run. The release positions Beam as a Western alternative to popular open-weight systems from Chinese developers, and the company plans to publish the model weights and full technical details this month.<\/p>\n
Beam is a text-only mixture-of-experts model built around heavy reinforcement learning, aimed at reasoning, coding, and agentic workloads. Reflection describes it as a “workhorse model” for enterprises, the public sector, and developers.<\/p>\n
The model has 501 billion total parameters with 23 billion active per pass, was pretrained on 23.8 trillion tokens, and supports a 1 million token context window. For comparison, Z.ai’s GLM-5.2 has roughly 744 billion total parameters with 40 billion active.<\/p>\n
Reflection’s benchmarks, which have not been independently verified, show Beam scoring on par with Z.ai’s GLM-5.2 on advanced reasoning evaluations and outperforming today’s leading Western open-weight models while using what the company describes as 3-4x less inference compute than rivals. On four coding tests where both report results, Reflection’s benchmarks show Beam outscoring Inkling, the open model from Thinking Machines Lab released in July, though Inkling is multimodal and Beam is text-only.<\/p>\n
Reflection is explicitly positioning Beam against three fronts: closed labs like Anthropic and OpenAI, popular open-weight models from Chinese developers, and Western open-weight competitors including Mistral, Meta, and Cohere. The framing is meant to give enterprises and governments a frontier-grade model they can run and customize locally rather than rent through an API.<\/p>\n
Beam will be distributed through hyperscalers and neoclouds, with integrations across open source libraries at launch.<\/p>\n
Beyond the model itself, Reflection is selling the idea of “AI factories”: packaged systems that let institutions train Reflection’s models on their own proprietary data, on infrastructure they control. The pitch targets enterprises and sovereign nations that want a customized local AI stack without depending on a closed US provider or a Chinese open-weight model.<\/p>\n
Reflection has already begun testing the concept through a sovereign AI factory partnership with Shinsegae Group in South Korea. Hedge funds and trading firms are also among the buyers exploring such systems.<\/p>\n
Frontier training and inference depend on access to top-tier GPUs, and Reflection has moved aggressively to secure it. This summer the company signed deals collectively worth more than $7 billion with SpaceX and Nebius to lock in access to Nvidia’s GB300 chips through 2029.<\/p>\n
Reflection was founded in 2024 and has raised roughly $4.7 billion from backers including Nvidia, Sequoia Capital, and Lightspeed Venture Partners, according to PitchBook. The company’s last funding round set a $25 billion pre-money valuation.<\/p>\n
Beam is Reflection AI’s first frontier open-weight AI model, a text-only mixture-of-experts system with 501 billion total parameters and 23 billion active parameters, trained on 23.8 trillion tokens with a 1 million token context window.<\/p>\n
Reflection says Beam scores on par with Z.ai’s GLM-5.2 on advanced reasoning benchmarks while using 3-4x less inference compute. These performance claims have not been independently verified.<\/p>\n
Reflection is pitching packaged systems that let enterprises and sovereign nations train the company’s open-weight models on their own proprietary data on infrastructure they control, with a live pilot underway at Shinsegae Group in South Korea.<\/p>\n