Kimi K3 is Moonshot AI's flagship open-weight large language model, positioned by Together AI as a natural fit for long-running agent and coding workloads. Together AI's documentation frames the model around its roughly one-the cataloged API limit and its integration with Moonshot's own Kimi Code CLI and VS Code extension, describing Kimi K3 as the backbone that Kimi Code was built around. The same guide highlights that the same provider plumbing can be reused across Together AI's broader serverless catalog, so developers can route Kimi Code through the hosted Kimi K3 endpoint and switch to other open-source models without changing their agent setup.
On the Together AI endpoint, the model is exposed with streaming, tool calling, vision input, and JSON output, along with prompt caching, giving it the surface area needed for tool-using agents and structured retrieval. Together AI lists pricing in line with other flagship-tier open-weight hosts, charging a few dollars per million input tokens and a higher rate for output, while keeping a cache-read discount that makes repeated context reuse economical. The combination of a very large context window, open weights, and an agent-friendly capability stack makes Kimi K3 through Together AI a practical choice for coding assistants, document-heavy RAG pipelines, and multi-step workflows that need both long memory and flexible tool use.
