Reflection raises $2B to be America’s open frontier AI lab, challenging DeepSeek

Reflection, a startup established by former Google DeepMind researchers, has successfully raised $2 billion, bringing its valuation to an impressive $8 billion. This marks a significant increase from its previous valuation of $545 million just seven months ago. Initially focused on autonomous coding agents, Reflection is now setting its sights on becoming a leading open-source alternative to established AI labs like OpenAI and Anthropic, while also positioning itself as a competitor to Chinese firms like DeepSeek. Founded in March 2024 by Misha Laskin, who previously led reward modeling for DeepMind's Gemini project, and Ioannis Antonoglou, co-creator of the renowned AlphaGo, the company’s strong foundation in advanced AI systems bolsters its vision. They believe that exceptional AI talent can drive the development of frontier models outside the confines of major technology corporations. Alongside their funding announcement, Reflection revealed the recruitment of top-tier talent from DeepMind and OpenAI, as well as the creation of an advanced AI training infrastructure that they promise will be accessible to all. A key aspect of Reflection's strategy involves identifying a scalable commercial model that aligns with its open intelligence vision. Currently, the team consists of approximately 60 AI researchers and engineers specializing in infrastructure, data training, and algorithm development, according to CEO Laskin. The company has also secured a compute cluster with ambitions to unveil a frontier language model next year, trained on “tens of trillions of tokens.” Laskin notes, “We have developed something once thought possible only in the world's top labs: a large-scale LLM and reinforcement learning platform capable of training massive Mixture-of-Experts (MoEs) models at frontier scale.” The Mixture-of-Experts architecture is crucial for powering frontier large language models (LLMs), capabilities that were previously the domain of large, closed AI laboratories. DeepSeek made a significant breakthrough when it pioneered the open training of these models, inspiring others such as Qwen and Kimi in China. Laskin emphasized the urgency of competing in this arena, stating, “DeepSeek and Qwen serve as our wake-up call; if we don’t act, the global standard of intelligence will be determined by others, not by America.” He highlighted the competitive disadvantage the U.S. faces, particularly as enterprises and governments may hesitate to utilize Chinese models for legal reasons. Response to Reflection's mission has been largely positive among American technologists. David Sacks, the White House's AI and Crypto Czar, expressed support on social media, stating, “It’s encouraging to see more American open-source AI models. A significant portion of the global market will prefer the cost, customization, and control that open-source solutions provide.” Clem Delangue, CEO of Hugging Face, echoed this sentiment, emphasizing the potential for American open-source AI to thrive. Reflection's approach to being “open” appears to focus on access rather than the full development of its models. Laskin indicated that while model weights will be made available for public use, the datasets and complete training processes will remain proprietary. “The most impactful element is the model weights, as anyone can utilize them for experimentation,” he noted. The company plans to generate revenue by catering to large enterprises and governments looking to develop their own “sovereign AI” systems. Laskin stated, “For large enterprises, an open model is essential. They seek ownership and the ability to run and customize the model according to their needs.” Reflection has yet to release its inaugural model, which will primarily focus on text but aims to incorporate multimodal capabilities eventually. The funds raised will be allocated towards obtaining the necessary computational resources for training these new models, with a target release date set for early next year. Notable investors in this funding round include Nvidia, Disruptive, DST, and several others.