Reflection AI Launches Beam, an Open-Weight Model Claiming to Match Chinese Rivals at Lower Compute Cost

Nvidia-backed Reflection AI announced its first open-weight frontier model, Beam, on October 5, 2026, claiming it matches leading Chinese open models on advanced reasoning benchmarks while using significantly less compute.

Beam is a text-only mixture-of-experts model with 501 billion total parameters and 23 billion active parameters. It was pre-trained on 23.8 trillion tokens and features a 1 million token context window. Reflection says Beam performs on par with Z.ai’s GLM-5.2 — which has roughly 744 billion total parameters and 40 billion active — while requiring “3–4x less inference compute.” Those performance claims have not been independently verified.

The Brooklyn-based startup, founded in 2024 by two former Google DeepMind researchers, has raised approximately $4.7 billion from investors including Nvidia, Sequoia Capital, and Lightspeed Venture Partners, with its last round valuing 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.

Reflection is positioning Beam against closed labs such as Anthropic and OpenAI, Chinese open-model developers including DeepSeek and Qwen, and Western open-model providers like Meta, Mistral, and Cohere. The company’s own benchmarks show Beam outscores Inkling — the open model from Mira Murati’s Thinking Machines Lab — on four coding tests where both report results, though Inkling is multimodal while Beam is text-only.

The company is targeting enterprises, developers, and sovereign nations with a concept it calls “AI factories” — customized, locally deployed AI systems built on Reflection’s models and trained on proprietary data. Reflection has already begun testing the approach with South Korea’s Shinsegae Group. Hedge funds and trading firms are also reported to be among interested parties.

Reflection says it will release Beam’s weights and full technical details later this month, with distribution through hyperscalers and neoclouds and integrations across open-source libraries.