Reflection AI, a Brooklyn-based startup founded in 2024 by ex-Google DeepMind researchers, has launched Beam, an open-weight AI model designed to compete with leading Chinese AI systems like DeepSeek and Qwen. The 501-billion-parameter model, with 23 billion active parameters, has been pretrained on 23.8 trillion tokens and possesses a 1 million token context window. Reflection claims Beam matches or exceeds the performance of major Chinese open models on advanced reasoning benchmarks while using three to four times less inference compute, offering significant cost advantages.

Beam is a text-only mixture-of-experts model trained with high-compute reinforcement learning techniques, making it effective in reasoning, coding, and agentic tasks. Reflection markets Beam as a “workhorse” model aimed at enterprises, public sector entities, and developers seeking advanced AI capabilities at lower operational costs. The company positions itself against closed AI labs like OpenAI and Anthropic, as well as popular Western open models and Chinese AI systems, emphasizing Beam’s efficiency and performance balance as a key differentiator.

Reflection is also promoting the concept of “AI factories,” which allow organizations—particularly sovereign nations and enterprises—to customize AI by training Reflection’s models on proprietary data, enabling localized, secure AI deployments. This vision aligns with Nvidia CEO Jensen Huang’s push for an open AI ecosystem powered by Nvidia GPUs. Reflection has secured over $7 billion in compute contracts with partners including SpaceX and Nebius, ensuring it has the necessary resources to train and deploy its models through 2029, bolstering its competitive position in the AI industry.

The company expects to release Beam’s weights and full technical documentation this month, making the model widely accessible through hyperscaler platforms and integrations with open source libraries. This openness aims to accelerate adoption and foster an ecosystem around Beam, positioning Reflection as a significant player capable of challenging both Chinese AI dominance and closed Western AI providers. The startup has raised approximately $4.7 billion in funding so far, with a valuation of $25 billion, supported by investors like Sequoia Capital, Lightspeed Venture Partners, and Nvidia.