Skip to main content

Kimi K3 Goes Open Source: A 3-Trillion-Parameter Beast Shakes Up AI

Moonshot AI has officially released the complete model weights of Kimi K3 on Hugging Face, along with a technical report on GitHub. This model, boasting 2.8 trillion parameters, 104 billion activated parameters, and a 1 million token context window, is the first open-weight model to cross the 3 trillion parameter threshold. The release comes just 11 days after its debut as a managed API on July 16.

But in those 11 days, the story has blown far beyond technical specs. K3 scored third globally (57 points) on the Artificial Analysis intelligence index, trailing only Claude Fable5 and GPT-5.6 Sol. It topped the front-end programming arena Arena with 1679 points. Hong Kong's Z.ai stock plunged 30% in a single day, while MiniMax dropped 16%. Moonshot AI's daily revenue surged at least sixfold, and its valuation anchor skyrocketed from $20 billion to $50 billion. The White House Office of Science and Technology Policy director publicly accused it of distilling Anthropic technology, and the Treasury Secretary warned of sanctions.

The License: More Than Meets the Eye

While the Kimi K3 License appears to be MIT, it hides some commercial strings. The agreement says any entity providing K3 inference or fine-tuning services to third parties with annual revenue over $20 million must sign a separate deal with Moonshot AI. Commercial products with over 100 million monthly active users or monthly revenue exceeding $20 million must prominently label "Kimi K3" on their interface. This is the first time a leading Chinese model company has explicitly priced the practice of "cloud vendors hitchhiking" into an open-source license. Those using it purely internally or through official certified inference partners are exempt.

Pricing Reality Check

Officially, K3's API costs $3 per million input tokens and $15 per million output tokens, compared to Claude Fable5's $10 and $50. But multiple testers report that K3 consumes more tokens for the same task, so the actual cost advantage per task isn't as big as the price sheet suggests.

Key Points

  • World's first open-weight model at 3 trillion parameters
  • Massive market impact: stock swings, revenue surge, valuation jump
  • License includes commercial clauses for large-scale users
  • Pricing is competitive but real-world costs may vary