MoonshotAI: Kimi K2 0711

MoonshotAI: Kimi K2 0711 is a cutting-edge large-scale Mixture-of-Experts (MoE) language model developed by Moonshot AI. With 1 trillion total parameters and 32 billion active per forward pass, Kimi K2 is specifically optimized for agentic capabilities, demonstrating superior performance in advanced tool use, complex reasoning, and efficient code synthesis. This model sets new benchmarks in coding (LiveCodeBench, SWE-bench), reasoning (ZebraLogic, GPQA), and tool-use (Tau2, AceBench) tasks. Designed for high-performance applications, Kimi K2 supports long-context inference up to 128K tokens, making it ideal for intricate and extensive tasks. It features a maximum output of 4K tokens and is available at a competitive pricing of $0.50 per 1M input tokens and $2.40 per 1M output tokens. Access to this powerful model is available through our PRO tier. Kimi K2's innovative training stack includes the MuonClip optimizer, ensuring stable large-scale MoE training. Its robust capabilities make it an excellent choice for developers and businesses seeking advanced AI solutions for chat, code generation, and complex problem-solving.