Reflection AI, started by ex-DeepMind folks, just pulled in $2 billion, bumping its value to $8 billion. That’s a big jump from its $545 million tag seven months back. The cash comes from big names like Eric Schmidt, Citi, 1789 Capital, Lightspeed, and Sequoia. Founders Misha Laskin and Ioannis Antonoglou – who worked on Gemini and AlphaGo at DeepMind – want to automate coding and make AI open and accessible, while keeping U.S. tech in the lead. With about 60 staff from DeepMind and OpenAI, they’re gunning for rivals like China’s DeepSeek, Meta, and Mistral.
What They’re Building and Why
Their Coding Agent API handles boring coding jobs to speed up software work. They share model weights for free so researchers can tinker, but keep the full training setup under wraps for safety. The money goes to more research, beefier computers, hiring, and teaming up with schools, companies, and government. Laskin says it’s a wake-up call against letting others set global AI standards – they want America in charge. They’re training huge Mixture-of-Experts models on text, images, audio, video, and code.
Who Gets Involved and When?
Founded in March 2024, Reflection AI is open for partnerships now with businesses and governments. The focus is U.S.-based, but their open models could go global. No set dates for new releases yet, but expect more on coding agents soon.
Sparking the Open AI Race
This haul puts Reflection AI right up against DeepSeek’s cheap, high-powered models, Meta’s massive Llama downloads, and Mistral’s regulated tech. It’s all about mixing innovation with ethics, sharing smarts without losing control – one model at a time.
