Ex-DeepMind founders raise $2B for Reflection to rival OpenAI and DeepSeek
Just a year after its founding, AI startup Reflection has secured $2 billion in new funding at an $8 billion valuation , marking one of the largest early-stage raises in the AI sector. The company, launched by former Google DeepMind researchers Misha LaskinDealroom has a profile for this one. Try Dealroom → and Ioannis AntonoglouDealroom has a profile for this one. Try Dealroom → , is positioning itself as the United States’ open alternative to leading closed frontier labs like OpenAI and Anthropic — and as a Western counterweight to China’s DeepSeek .
Reflection began with a focus on autonomous coding agents , but its ambitions have since expanded. The company now aims to develop large-scale language and reasoning models through an open approach , sharing model weights while retaining proprietary datasets and training pipelines. Its goal: to make frontier AI capabilities accessible to researchers and enterprises beyond Big Tech.
With its new capital, Reflection plans to train a frontier-scale language model on “tens of trillions of tokens” using a Mixture-of-Experts (MoE) architecture — the same type of system that underpins top-performing models like Gemini and DeepSeek. The company has also assembled a 60-person team drawn from DeepMind and OpenAI and built a custom AI training stack designed for openness and scalability.
Reflection’s business model revolves around dual access : researchers can freely use its open models, while large enterprises and governments can license or customize them for specific needs — including “sovereign AI” projects. Laskin argues that such open models give institutions greater control, transparency, and cost efficiency compared to closed systems.
The round, first rumored in September at around $1 billion, attracted major backers including Nvidia, Disruptive, DST, 1789, B Capital, Lightspeed, GIC, Eric Yuan, Eric Schmidt, Citi, Sequoia, and CRV . The announcement follows growing geopolitical urgency around AI independence, as U.S. technologists rally to ensure that “the global standard of intelligence” — as Laskin put it — is built in the West, not elsewhere .
Reflection’s first model, expected early next year, will be text-based with multimodal capabilities to follow , setting the stage for what could become America’s leading open frontier AI lab .
Source:
A.M.
