Reflection AI launches Beam, open-weight frontier model
Original: Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost
Why This Matters
A well-funded Western open-weight challenger to DeepSeek and Qwen raises competitive stakes across the market.
Reflection AI unveiled Beam, a 501B-parameter open-weight mixture-of-experts model, on Oct. 5, 2026. The Brooklyn startup claims it matches Chinese rival Z.ai's GLM-5.2 on reasoning benchmarks using 3-4x less inference compute.
Reflection AI has officially launched Beam, its first frontier open-weight model, targeting enterprises, developers, and sovereign nations. The company describes Beam as a text-only mixture-of-experts model with 501 billion total parameters and 23 billion active parameters, pretrained on 23.8 trillion tokens with a 1 million token context window. Reflection says Beam was trained using high-compute reinforcement learning to excel at reasoning, coding, and agentic tasks.
On advanced reasoning benchmarks, Reflection claims Beam matches Z.ai's GLM-5.2 (744B total / 40B active parameters) and outperforms leading Western open models — at 3-4x lower inference compute. Those figures have not been independently verified.
Reflection positions Beam against Anthropic, OpenAI, Meta, Mistral, and Cohere, as well as Chinese open-weight labs. Its most direct domestic rival is Thinking Machines Lab's Inkling, released in July; Reflection's own benchmarks show Beam leads on four coding tests, though Inkling supports multimodal inputs while Beam is text-only.
Founded in 2024 by two former Google DeepMind researchers, Reflection has raised roughly $4.7 billion from Nvidia, Sequoia Capital, and Lightspeed, with its last round pegging a $25 billion pre-money valuation. This summer the company locked in over $7 billion in compute deals with SpaceX and Nebius for Nvidia GB300 chips through 2029. Its enterprise pitch centers on "AI factories" — letting institutions train Beam on proprietary data to build local, customized AI systems.