Open frontier AI lab Reflection raises $2B to rival DeepSeek
Nvidia has led a $2bn round in Reflection AI, valuing the New York startup at $8bn. Founded by ex‑DeepMind researchers, the firm builds open frontier models and autonomous coding.
Reflection AI has secured $2 billion in fresh financing that has valued the New York–based startup at $8 billion, positioning the company as an American, openly oriented “frontier” AI lab amid intensifying competition with China’s DeepSeek.
As per reports, the round was led by Nvidia and included backers such as Eric Schmidt, Citi and 1789 Capital, alongside existing investors Lightspeed and Sequoia.
The company’s earlier financing this year totalled about $130 million and had valued Reflection AI at roughly $545 million.
Founders and focus
Reflection AI was founded in 2024 by former Google DeepMind researchers Misha Laskin (CEO) and Ioannis Antonoglou.
The team has focused on AI systems that automate software development, and the company has described itself as building open foundation models and advancing large-scale reinforcement learning to push frontier capability.
With the fresh funding, Reflection AI has said it intends to continue building and releasing frontier models “sustainably,” emphasising openness to enable broader participation in safety research and oversight rather than concentrating decisions within closed labs.
Challenging DeepSeek
DeepSeek has set the pace in 2025 by releasing reasoning models such as R1 under an MIT licence and by disclosing a headline training cost of $294,000, which sharpened the debate on efficiency and openness.
Reflection AI’s funding arrived directly in that context, with the startup positioning its research and tooling as openly available to developers and researchers in the US.
Microsoft also made DeepSeek’s R1 model available via Azure and GitHub, further reflecting the model’s reach and the competitive bar US labs have sought to meet.