Brooklyn startup Reflection AI is officially launching Beam as its first frontier, open-weight artificial intelligence model. The two-year-old company states that Beam matches the performance of leading Chinese open models on advanced reasoning benchmarks while cutting costs dramatically. This announcement confirms weekend reporting from Axios that the startup neared a launch.
Beam functions as a text-only mixture-of-experts model trained on high-compute reinforcement learning. Reflection designed the system to handle reasoning, coding, and agentic tasks using a fraction of the token cost and inference time compute required by rivals. The model contains 501 billion total parameters with 23 billion active parameters. It underwent pretraining on 23.8 trillion tokens and includes a 1 million token context window. For comparison, Z.ai features roughly 744 billion total parameters with 40 billion active in its GLM-5.2 model.
Beam matches performance using less compute
Independent verification of Reflection’s performance claims is currently unavailable. However, the startup reports that Beam scores on par with Z.ai’s GLM-5.2 on advanced reasoning benchmarks. The model also outperforms leading Western open models while consuming three to four times less inference compute. Reflection markets the product as a workhorse model tailored for enterprises, the public sector, and developers.
Reflection positions its technology against closed labs like OpenAI and Anthropic, popular open models from Chinese developers, and Western players such as Meta, Mistral, and Cohere. Its most direct United States rival may be Inkling, an open model released in July by Mira Murati’s Thinking Machines Lab. Reflection states that Beam outscores Inkling on four coding tests where both report results, though Inkling is multimodal while Beam handles text only.
The startup secured billions in funding and chips
Two former Google DeepMind researchers founded Reflection in 2024. PitchBook data shows the startup has raised approximately $4.7 billion from backers that include Nvidia, Sequoia Capital, and Lightspeed Venture Partners. Its most recent financing round valued the company at $25 billion pre-money.
The enterprise has actively secured computing power to train frontier models. This summer, Reflection established agreements worth more than $7 billion combined with SpaceX and Nebius to acquire Nvidia GB300 chips through 2029.
Read nextAleph Alpha Releases Kolibri Open-Weight Model for European SovereigntyReflection targets enterprises and sovereign nations
Reflection aims Beam and future releases at enterprises and sovereign nations. The company pitches AI factories, which are products allowing institutions to build customized local AI systems by training Reflection models on proprietary data. Nvidia CEO Jensen Huang has long supported the AI factory concept and the open AI ecosystem.
Hedge funds and trading firms are among the entities interested in building such systems, according to Axios reporting. Reflection already started testing a sovereign AI factory partnership with Shinsegae Group in South Korea.
Reflection will release Beam weights and technical details this month. Distribution will occur through hyperscalers and neoclouds alongside integrations across open source libraries at launch. Reflection did not respond in time to requests for additional information from TechCrunch.
