MoonshotAI: Kimi K2 0711
🧠 AI Modelmoonshotai
1T-parameter MoE model with 32B active per forward, efficient and powerful.
Kimi K2 Instruct is a state-of-the-art Mixture-of-Experts (MoE) language model from Moonshot AI. With 1 trillion total parameters, it activates only 32 billion per forward pass, balancing capability and computational efficiency. The model supports a context length of 131,072 tokens and processes text input and output. It features configurable sampling parameters such as frequency_penalty, max_tokens, temperature, and tool_choice, enabling fine-tuned generation. On benchmarks, Kimi K2 achieves an ELO of 1085 on code categories (rank #79), 1062 on data visualization (rank #81), and 1043 on game development (rank #86). Designed for advanced text generation tasks, it leverages the MoE architecture to deliver high performance while reducing active parameter count, making it suitable for large-scale AI applications.
💡Highlights
- ├─1T total parameters
- ├─32B active per forward
- └─131k context length
🎯For
- ├─AI researchers
- ├─developers building text applications
- └─ML engineers