Reflection AI, a startup founded by former Google DeepMind researchers, has completed a funding round of $2 billion, elevating its valuation to $8 billion. This financing marks a significant increase from its $545 million valuation seven months prior, positioning the company as a key player aiming to offer an open-source alternative to established frontier AI labs and a Western competitor to Chinese AI firms.
The company, launched in March 2024, was co-founded by Misha Laskin, who previously led reward modeling for DeepMind's Gemini project, and Ioannis Antonoglou, a co-creator of AlphaGo. Their collective experience in developing advanced AI systems underpins their strategy to build frontier models outside the established tech giants.
Alongside its latest funding, Reflection AI announced the recruitment of talent from DeepMind and OpenAI. The company has also developed an advanced AI training stack, which it states will be open for use. CEO Misha Laskin indicated that Reflection AI has "identified a scalable commercial model that aligns with our open intelligence strategy."
The team currently comprises approximately 60 individuals, primarily AI researchers and engineers. Reflection AI has secured a compute cluster and plans to release a frontier language model next year, trained on "tens of trillions of tokens," Laskin told TechCrunch. The company stated in a post on X that it has built a large-scale LLM and reinforcement learning platform capable of training massive Mixture-of-Experts (MoE) models at frontier scale, an architecture previously associated with large, closed AI labs.
Laskin emphasized the geopolitical context, noting the advancements of Chinese models like DeepSeek and Qwen. He stated this represents a "wake-up call" for Western AI development, suggesting that a lack of action could lead to non-American entities setting the global standard for intelligence, which could place the U.S. and its allies at a disadvantage due to potential legal impediments to using Chinese models.
The company's definition of "open" centers on providing public access to model weights, while keeping datasets and full training pipelines proprietary. This approach aligns with its business model, which seeks revenue from large enterprises building products on its models and from governments developing "sovereign AI" systems. Laskin stated that once enterprises require significant AI integration, they often prefer an open model for ownership, infrastructure control, cost optimization, and customization.
Reflection AI's first model, primarily text-based with future multimodal capabilities, has not yet been released. The newly acquired funds will be allocated to securing the necessary compute resources to train these new models, with the initial release anticipated early next year. Investors in the latest round include Nvidia, Disruptive, DST, 1789, B Capital, Lightspeed, GIC, Eric Yuan, Eric Schmidt, Citi, Sequoia, and CRV, among others.
