Kimi K3

Kimi K3 is Moonshot AI's flagship model for long-context coding, complex reasoning, and knowledge work. It combines native vision with a 1M-token context window for repository analysis, long-document research, and multi-step agent workflows.

Kimi K3 is designed for long-horizon coding, complex reasoning, and agent-style work. Use the Kimi K3 benchmarks on this page to shortlist models, then validate it with your own repository, prompts, tools, latency target, and evaluation set.

Review Kimi K3 input, output, and cache-read pricing separately. Coding and research tasks with long responses are usually more sensitive to output cost, while repeated long-context prompts benefit more from cache pricing.

Create a TokenHub API key, set the model ID to `kimi-k3`, and use an available endpoint shown on this page. Existing OpenAI-compatible applications can usually keep their request flow and update the API key, Base URL, and model ID.

Choose by workload, not a generic ranking. Start with Kimi K3 when long-context coding, knowledge work, or native vision are central; then compare Kimi K3 vs DeepSeek V4 Pro on API price, output volume, latency, tool use, and your own evaluation results.

Kimi K3 supports a 1M-token context window. The Kimi K3 API can therefore be used for long documents, large codebases, research material, and multi-turn agent workflows; actual cost depends on tokens sent and generated per request.

Kimi K3 includes native vision capability. Before building an image or multimodal workflow, check the currently published modalities and endpoint support on this Kimi K3 model page, because support can vary by endpoint.