Reflection AI, a startup founded by ex-DeepMind researchers, has raised $2 billion at an $8 billion valuation to build an open alternative to OpenAI, Anthropic, and China’s DeepSeek and Qwen.
Reflection AI Raises $2 Billion to Build Open American Alternative to OpenAI
Reflection AI, a startup founded in 2024 by former Google DeepMind researchers, has raised $2 billion at an $8 billion valuation — 15 times more than it was just seven months ago.
The company was founded by Misha Laskin, who led reward modeling on Project Gemini, and Ioannis Antoglu, one of the creators of AlphaGo, the legendary system that defeated the world Go champion.
Their mission is to build an open alternative to closed “frontier” labs like OpenAI and Anthropic, while also becoming a Western answer to Chinese giants DeepSeek and Qwen.
“We have created a world-class platform — a large-scale LLM and RL system for training Mixture-of-Experts models,” Reflection AI’s founders said.
Team, model, and scale
Reflection AI already has 60 experts, mostly engineers and researchers from DeepMind and OpenAI. The company plans to release its first language LLM in early 2026 — trained on tens of trillions of tokens.
Laskin explained that the team has built its own infrastructure for training MoE (Mixture-of-Experts) models — an architecture that provides the scalability and speed characteristic of advanced models like DeepSeek or Gemini.
Open strategy — with a business focus
Reflection AI positions itself as “open intelligence” — but in a format similar to Meta (Llama) or Mistral:
- model weights will be open;
- datasets and training pipelines will remain private.
“The most important thing is the model weights. They can be used, modified and developed,” says Laskin.
The company will allow free use for research, but will monetize work with large enterprises and governments that develop “sovereign AI systems” — national models running on their own infrastructure.
US vs. China: The technological race for intelligence
Laskin stressed that the emergence of DeepSeek and Qwen was a “wake-up call”:
“If we do nothing, then America will not create world standards for intelligence.”
Due to political and legal risks, many countries and companies cannot use Chinese models, so the demand for Western open alternatives is growing rapidly.
American technology leaders have already supported Reflection AI.
- David Sachs, White House AI & Crypto Czar, wrote: “Open models are the future, and the US should win this category.”
- Clem DeLang, CEO of Hugging Face, added: “The main thing now is to show the pace of open releases and data.”
Who invested $2 billion
The Reflection AI round was invested by leading industry players:
Nvidia, Disruptive, DST, 1789, B Capital, Lightspeed, GIC, Eric Yuan, Eric Schmidt, Citi, Sequoia, CRV and others.
The company will use the funds received to scale computing power and release the first open world-class LLM in early 2026.
A New Era of AI
Reflection AI wants to combine openness, scale, and national security, creating a platform that will change the global balance of AI development.
“If big companies build AI that only they control, that’s a risk. We want intelligence to belong to humanity,” concluded Laskin.
Related: OpenAI is preparing a new operating system with apps from ChatGPT
