Chinese startup Moonshot AI has unveiled Kimi K3, a massive 2.8 trillion parameter model that is the world's largest open-source AI system. Released on July 16, 2026, Kimi K3 is designed for autonomous software development and utilizes a vision-in-the-loop system. While Moonshot AI admits K3's overall performance trails top U.S. models like OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5, it has outperformed them on specific coding benchmarks at a fraction of the cost.
Kimi K3 model release
- ▪Kimi K3 features 2.8 trillion parameters, making it the largest open-source artificial intelligence model to date
- ▪Moonshot AI unveiled its Kimi K3 artificial intelligence model on July 16, 2026
- ▪Moonshot AI's Kimi K3 model has a 1 million-token context window and can process both text and images
Performance benchmarks versus US models
- ▪Moonshot AI acknowledged that Kimi K3's overall performance still trails the most powerful proprietary models, including GPT-5.6 Sol and Claude Fable 5
- ▪Kimi K3 outperformed Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol on specific evaluations including Program Bench and SWE Marathon
- ▪In blind testing by AI evaluator Arena, developers preferred Kimi K3 over leading U.S. models for front-end coding
Open-weights model strategy
- ▪The open-source nature of Chinese models allows developers to download them for free and customize them, though it requires technical expertise and rented cloud chips
- ▪Moonshot AI plans to release the weights of the Kimi K3 model to the public on July 27, 2026
API pricing structure
- ▪Kimi K3's API pricing is $0.30 per 1 million input tokens with a cache hit and $3.00 without a cache hit
- ▪Kimi K3 is priced at $15 per 1 million output tokens, which is cheaper than Anthropic's Claude Fable 5 at $50 per 1 million output tokens
Autonomous software development focus
- ▪Kimi K3 is designed for long-running autonomous software development tasks, analyzing large codebases, and coordinating programming tools
- ▪Kimi K3 utilizes a vision-in-the-loop system that examines screen captures, modifies code, and checks the visible output
US-China AI competition
- ▪U.S. export controls have restricted Chinese developers from accessing advanced AI processors, forcing them to focus on fundamental research and efficiency
- ▪Anthropic has accused Chinese AI firms, including Moonshot AI and DeepSeek, of violating rules by using model distillation to train their models
Story comments
Loading comments…