Reflection AI unveils Beam, an open-weight model it says rivals Chinese AI at lower cost

Reflection AI has unveiled Beam, its first frontier open-weight model, and says it matches the performance of leading Chinese open models on advanced reasoning benchmarks at a fraction of their inference cost. The announcement, made in a blog post Monday, confirms weekend reporting from Axios that the Brooklyn-based startup was close to a launch.

Beam is a mixture-of-experts model, meaning it activates only a subset of its parameters for any given task. It has 501 billion total parameters with 23 billion active at once, and was pretrained on 23.8 trillion tokens of text. It carries a context window, the amount of text a model can consider at once, of 1 million tokens. By comparison, Z.ai’s GLM-5.2 has roughly 744 billion total parameters with 40 billion active. Reflection describes Beam as text-only, built for reasoning, coding, and agentic tasks.

Reflection says Beam scores on par with GLM-5.2 on advanced reasoning benchmarks and beats today’s leading Western open models, while using three to four times less inference compute, the processing power needed to run the model once trained. The company has not had these claims independently verified. Reflection is pitching Beam as a “workhorse model” for enterprises, government agencies, and developers.

Measuring up against rivals

Reflection is positioning itself against closed labs like Anthropic and OpenAI, against open models from Chinese developers, and against Western players including Mistral, Meta, and Cohere. Its closest U.S. rival may be Inkling, the open model Mira Murati’s Thinking Machines Lab released in July. Reflection’s own benchmarks show Beam outscoring Inkling on four coding tests both companies report, though Inkling is multimodal, meaning it handles images as well as text, while Beam is text-only.

Reflection was founded in 2024 by two former Google DeepMind researchers and has raised about $4.7 billion from backers including Nvidia, Sequoia Capital, and Lightspeed Venture Partners, according to PitchBook. Its most recent funding round valued the company at $25 billion pre-money. This summer, Reflection signed deals worth more than $7 billion with SpaceX and Nebius to secure access to Nvidia’s GB300 chips through 2029, part of a push to lock up the computing power needed to train frontier models.

The company is also marketing what it calls “AI factories,” a product that would let enterprises and governments train Reflection’s models on their own proprietary data to build customized, local systems. Nvidia CEO Jensen Huang, whose company backs Reflection, has promoted this idea as a way to strengthen the open AI ecosystem, a shift that would also expand demand for Nvidia’s chips. Axios reported that hedge funds and trading firms have shown interest, and Reflection has begun testing a sovereign AI factory partnership with South Korea’s Shinsegae Group.

Reflection says it will release Beam’s full weights and technical details later this month, with distribution through major cloud providers and integration into open-source software libraries at launch. The company did not respond to TechCrunch’s requests for comment before publication. What remains to be seen is whether outside researchers can confirm Reflection’s benchmark claims once the weights are public.