Reflection AI has officially introduced Beam, its first frontier open-weight large language model. The Brooklyn-based startup, founded in 2024, says Beam delivers performance comparable to leading Chinese open models on advanced reasoning benchmarks while operating at significantly lower cost — a positioning that could intensify efforts to create a Western alternative to models such as DeepSeek, Qwen, and Z.ai.
The announcement corroborates an Axios report over the weekend that the company was nearing launch. Reflection published a detailed blog post on Monday outlining Beam as a text-only mixture-of-experts model trained with high-compute reinforcement learning to be effective at reasoning, coding, and agentic tasks while using what the company describes as “a fraction of the token cost and inference time compute” compared with rivals.
Technical specifications and claimed performance
- Beam has 501 billion total parameters, with 23 billion active parameters.
- The model was pre-trained on 23.8 trillion tokens.
- Beam supports a 1 million token context window.
For comparison, Z.ai’s GLM-5.2 has roughly 744 billion total parameters and about 40 billion active parameters.
Reflection’s performance claims have not been independently verified. The company reports that on advanced reasoning benchmarks Beam scores on par with Z.ai’s GLM-5.2 and outperforms current leading Western open models while using “3–4x less inference compute.” Reflection describes Beam as a “workhorse model” intended for enterprises, the public sector, and developers.
Market positioning and competitors
Reflection positions itself against closed labs such as Anthropic and OpenAI, popular open models from Chinese developers, and Western players including Mistral, Meta, and Cohere. Its most direct U.S. competitor may be Inkling, the open model released in July by Mira Murati’s Thinking Machines Lab. Reflection’s own benchmarks indicate Beam outperforms Inkling on four coding tests where both report results; however, Inkling is multimodal, while Beam is text-only.
Funding and compute commitments
Reflection was founded in 2024 by two former Google DeepMind researchers. According to PitchBook, the company has raised roughly $4.7 billion from backers including Nvidia, Sequoia Capital, and Lightspeed Venture Partners, and its last financing round placed a $25 billion pre-money valuation on the company.
The startup has been securing large-scale compute capacity, a key input for training frontier models capable of competing with Anthropic’s and OpenAI’s closed models as well as lower-cost open-weight Chinese models. This summer Reflection signed deals valued collectively at over $7 billion with SpaceX and Nebius to secure access to Nvidia’s GB300 chips through 2029.
AI factories and sovereign deployments
Reflection aims to sell Beam and future models to enterprises and sovereign nations. The company is pitching an “AI factory” product that would allow institutions to build customized, local AI systems by training Reflection’s models on their proprietary data. Nvidia CEO Jensen Huang, a Reflection backer, has long promoted the AI factory concept and strengthening the open AI ecosystem — a vision that would also benefit Nvidia by expanding demand for its GPUs.
Axios reported that hedge funds and trading firms are among the parties interested in building such systems. Reflection has begun testing a sovereign AI factory partnership with Shinsegae Group in South Korea.
Release plans and availability
Reflection says it will release Beam’s weights and full technical documentation this month, distributing through hyperscalers and neoclouds and offering integrations with open source libraries at launch.
Reflection did not respond in time to TechCrunch’s requests for further information.
(Source material: Reflection AI announcement, PitchBook, Axios, TechCrunch.)
