Reflection AI prepares US rival to DeepSeek and Qwen

Nvidia-backed Reflection AI is nearing the release of its first open-weight model, positioning it against China’s DeepSeek and Alibaba’s Qwen. The company has not disclosed the model’s name, launch date, model card or benchmarks, but Axios reported that it is expected to trail top closed systems from OpenAI, Anthropic and Google while competing directly with leading Chinese open-weight models in coding, math and cost efficiency.

Reflection was founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou. Nvidia led an $800 million portion of the startup’s $2 billion Series B in October 2025, valuing Reflection at $8 billion, and Laskin later said the company had raised new financing at a $25 billion pre-money valuation.

The startup has committed to large-scale compute before shipping a flagship foundation model. Its deals include more than $1 billion in Nvidia compute from Nebius through 2029 and a SpaceX agreement tied to the Colossus 2 data center, with Reflection paying $150 million a month starting July 1, 2026, in a contract that could total $6.3 billion. The prospective release also carries a government and enterprise angle, giving Western buyers a potential alternative to Chinese open-weight systems for sensitive deployments.