Reflection AI, a New York-based startup, has just announced that it has raised $2 billion in venture capital at a valuation of $8 billion (about CZK 186.8 billion), including this investment. This represents a dramatic increase from the company's previous valuation of $545 million (around CZK 12.7 billion) in March of this year. The first reports of the talks came from The Information. Investors include Lightspeed Venture Partners, Sequoia Capital, and 1789 Capital, which is associated with Donald Trump Jr. Nvidia contributed the largest share, about $800 million (approximately CZK 18.7 billion).
This massive influx of funding will enable the company to develop new technologies. Founders Misha Laskin and Ioannis Antonoglou want to build Reflection AI into a U.S. provider of open-source artificial intelligence (AI) models, a capital-intensive endeavor. The money will primarily go toward development and expanding the team, which consists of around 60 researchers and engineers.
What Reflection AI Does
The company focuses on creating frontier open-source models, particularly for autonomous agents capable of programming. These models are intended to automatically generate, debug, and optimize code, going beyond simple code completion. Reflection AI plans to release a model next year trained on tens of trillions of tokens, the units of data used in training. Its approach includes reinforcement learning, where systems improve based on experience gained through long-term planning.
The founders bring experience from leading laboratories. Misha Laskin led reward modeling for the Gemini project and previously founded Claire AI, a company focused on forecasting product demand. Ioannis Antonoglou spent more than a decade at DeepMind, where he designed the architecture of AlphaGo, the first system to defeat a Go master. This background helps them build a team with a high concentration of reinforcement learning experts.
How It Differs from the Competition
Reflection AI aims to differentiate itself by releasing model weights for public use while keeping its training methods and datasets proprietary. This is similar to the approach taken by companies such as Meta with its Llama model, but the company focuses on large enterprises and government systems, where it offers ownership, infrastructure control, and customization. Unlike closed labs such as OpenAI or Anthropic, which keep their technologies under lock and key, Reflection AI supports open access for researchers and offers paid, customized versions for businesses and governments.
Compared with the Chinese laboratory DeepSeek, which it explicitly identifies as a rival, Reflection AI positions itself as a Western alternative emphasizing autonomous software development. Its agents are intended to handle the entire code development cycle, not just simple tasks, and aim to accelerate progress toward superintelligent systems. This approach attracts investors who see potential in openness combined with commercial applications for national security and enterprise needs.
With the new funding, Reflection AI plans to expand its offerings to include models that enable autonomous software engineering. The company is seeking to become a U.S. leader in open AI models, including support for both research and commercial deployment.
