# Kimi K2

> Kimi K2 is a state-of-the-art AI model developed by MoonshotAI, featuring 1 trillion parameters and excelling in reasoning, coding, and autonomous tasks. It is designed for tasks like coding, reasoning, and tool use, making it ideal for automating complex workflows. Launched on July 11, 2025, Kimi K2 is optimized for agentic capabilities, enabling it to act and solve problems autonomously. The model is open-source and available for API access or local deployment.

Kimi K2 is a mixture-of-experts (MoE) language model with 32 billion activated parameters and 1 trillion total parameters. It was pre-trained on 15.5 trillion tokens using the MuonClip Optimizer, ensuring stability and efficiency. Its architecture includes 384 experts, a 128K token context length, and advanced features like MLA attention and SwiGLU activation. Kimi K2 sets new standards in various AI benchmarks, demonstrating superiority in coding, tool use, and reasoning tasks.

## Docs
- [Architecture](/architecture.md): Details about the mixture-of-experts architecture.
- [Training](/training.md): Information on the training process and the MuonClip Optimizer.
- [Benchmarks](/benchmarks.md): Performance metrics across various benchmarks.
- [Usage](/usage.md): Instructions on how to access and use Kimi K2 via API or local deployment.
- [Features](/features.md): Advanced features and capabilities of Kimi K2.
- [Testimonials](/testimonials.md): User feedback and experiences with Kimi K2.
- [FAQ](/faq.md): Answers to common questions about Kimi K2.

## Optional
- [Privacy Policy](/privacy-policy.md): Information on how user data is handled.
- [Terms of Service](/terms-of-service.md): Legal terms governing the use of Kimi K2 services.
- [Contact](/contact.md): Contact information for support and inquiries.