Reflection AI raises $2B, positions as open US frontier lab

Reflection AI raised $2 billion at an $8 billion valuation, a roughly fifteen-fold increase from the $545 million valuation it carried seven months earlier. Investors included Nvidia, Sequoia, DST Global, Lightspeed, B Capital and GIC, alongside individual backers Eric Yuan and Eric Schmidt.

Founded in March 2024 by Misha Laskin, formerly of DeepMind’s Gemini reward-modelling team, and Ioannis Antonoglou, a co-creator of AlphaGo, Reflection had originally built autonomous coding agents before pivoting toward frontier large language models trained with a mixture-of-experts architecture on what the company said would be tens of trillions of tokens; its first text model was slated for release in early 2026. The company, now roughly 60 researchers and engineers, plans to publish model weights while keeping its datasets and training infrastructure proprietary, and to make money selling to enterprises and governments building sovereign AI systems on top of its models.

Reflection pitched itself explicitly as filling a gap it argued no other US lab occupied: an open-weight alternative to closed labs such as OpenAI and Anthropic, and a domestic counterweight to Chinese open-weight labs such as DeepSeek. Laskin framed China’s open-model progress as a “wake-up call,” arguing that American enterprises and governments were reluctant to build on Chinese models for legal and security reasons but had few well-funded domestic open alternatives to choose instead — a gap the raise was designed to fill before Reflection had shipped a model of its own.