Reflection AI, founded in March 2024 by two former Google DeepMind researchers, has raised $2 billion at an $8 billion valuation — roughly fifteen times its $545 million mark from seven months ago. Investors include Nvidia, Sequoia, Lightspeed, DST, B Capital, Citi, 1789 and Eric Schmidt, among others.

Misha Laskin, who led reward modeling on Gemini, and Ioannis Antonoglou, a co-creator of AlphaGo, are repositioning the company from autonomous coding agents to what they call an open frontier lab: an open-weights alternative to OpenAI and Anthropic and a Western counterpart to DeepSeek. Laskin says DeepSeek and Qwen were a “wake-up call,” warning that otherwise the global standard of intelligence will be built by someone else, not America.

Reflection says it has built a training stack capable of large-scale Mixture-of-Experts models at frontier scale and has secured a compute cluster. Laskin says the first model, largely text-based with multimodal work to follow, is targeted for early next year and will be trained on tens of trillions of tokens. The company plans to release model weights publicly while keeping datasets and full training pipelines proprietary — access rather than development. Its roughly 60 employees skew toward researchers and engineers.

Revenue is meant to come from large enterprises running the models on their own infrastructure, and from governments building sovereign AI.