Reflection AI, founded by ex-DeepMind leaders, has secured $2 billion to create open frontier-scale AI models, offering the U.S. an open source alternative to DeepSeek, OpenAI, and Anthropic.
Reflection AI has raised $2 billion at an $8 billion valuation, a fifteenfold leap from its previous $545 million valuation just seven months ago. The round saw participation from Nvidia, Disruptive, DST, 1789, B Capital, Lightspeed, GIC, Eric Yuan, Eric Schmidt, Citi, Sequoia, and CRV.
Founded in March 2024 by Misha Laskin, former DeepMind Gemini reward modelling lead, and Ioannis Antonoglou, co-creator of AlphaGo, the U.S.-based startup is positioning itself as America’s open frontier AI lab—an alternative to closed players like OpenAI and Anthropic, and a Western counterpart to China’s DeepSeek.
The company’s mission is to make frontier-scale AI openly accessible. Its large-scale LLM and reinforcement learning platform is designed to train Mixture-of-Experts (MoE) models at frontier scale, with its first frontier language model set for release in 2026, trained on tens of trillions of tokens.
Laskin said Reflection AI will release model weights publicly to encourage open development while keeping datasets and training pipelines proprietary.
“In reality, the most impactful thing is the model weights, because the model weights anyone can use and start tinkering with them,” he explained.
The firm’s commercial model targets enterprises and governments building sovereign AI systems.
“Once you get into that territory where you’re a large enterprise, by default you want an open model,” Laskin noted.
White House AI and Crypto Czar David Sacks welcomed the move, calling it “great to see more American open source AI models.” Hugging Face CEO Clem Delangue added that Reflection AI’s rise is “great news for American open-source AI.”
