TheChek News Flow

The Takeaway

Reflection AI, a startup co-founded by ex-Google Deepmind researchers, is raising $1 billion to develop open-source AI models, as cheaper and more readily customized software from China gains in popularity.

Reflection AI, a one-year-old startup co-founded by former Google DeepMind researchers, is in talks to raise more than $1 billion to fund its efforts to develop open-source large language models to compete with the likes of China’s DeepSeek, France’s Mistral and U.S-based Meta Platforms, according to two people with direct knowledge of the matter. The company has raised most of its target, according to one of those people.

The New York-based startup will use some of the money for the cost-intensive development of new AI models. Co-founders Misha Laskin and Ioannis Antonoglou have told Reflection employees they believe there is an opportunity to establish Reflection AI as the preeminent U.S.-based provider of open-source AI models, said one of the people. Its ambitions show how the popularity of DeepSeek and other Chinese AI models has catalyzed U.S. AI companies to work on open-source.

In recent months, Meta, the most prominent American developer of open-source AI, has gone on a hiring spree to revamp its AI efforts after developers found the most recent model fell short of their expectations, particularly compared to DeepSeek. The social media giant has since discussed developing closed AI models. On the popular AI model leaderboard LMArena as of Monday, none of the open-source models in the top 30 come from American developers.

For the last year, Reflection AI has been developing a coding agent called Asimov that analyzes corporate data, such as emails and Slack messages, to generate relevant code for applications. It launched a preview last month and has already started to generate a small amount of revenue from corporations, said one of the people and a second familiar with the company’s financials.

The founders are expanding to developing open-source AI models after corporate demand for China’s models exploded. While many large companies are building applications on models from OpenAI and Anthropic, open source models cost less and give them more flexibility, such as being able to access the underlying training data and code. Companies can fine-tune AI models for specific business processes, such as sales, in a way they’re not able to do with proprietary AI models.

However, many U.S. companies can’t use models from DeepSeek or other Chinese AI firms due to data security concerns. OpenAI CEO Sam Altman has said that the company is aiming to release its own open-source model this summer.