There are four distinct routes to Kimi K3, and they suit very different people. Picking wrong is the most common reason someone concludes the model is awkward or expensive when it is neither.
1. The Kimi apps (easiest)
The consumer apps — web, mobile, and Kimi Work — run K3 Max as the default. This is the route for research, writing, document analysis, and general assistant use, with no API key and no per-token billing to think about.
Which capabilities and limits apply on which plan changes over time, so check the app rather than trusting a blog post — including this one.
2. The Kimi API (for building)
| Base URL | https://api.moonshot.ai/v1 |
| Model ID | kimi-k3 |
| Compatibility | OpenAI-compatible and Anthropic-compatible |
The practical consequence of that last row: if you already have code written against the OpenAI SDK, you change a base URL and a model string. Same for the Anthropic SDK. This is the lowest-friction migration path of any frontier model.
3. Third-party routers
K3 is available through aggregators such as OpenRouter, which is useful if you want to compare models behind one API key or need fallback routing across providers.
Two caveats. Router pricing includes a margin, so it is generally above Moonshot's direct rate. And provider-level limits can differ from the official specification — output ceilings in particular are sometimes capped lower than the model supports. Check the specific route before you rely on a number.
4. Self-hosting (for a small number of organisations)
The weights are open under the Kimi K3 License and supported by vLLM, SGLang, and TokenSpeed. This is a real option and a demanding one: ~1.4 TB of weights means a multi-node GPU deployment before you account for KV cache on a 1M-token window.
See hardware requirements for the honest version, and the licence explainer before building a business on it.
Which route should you pick?
- Just want to use it? The Kimi apps.
- Building a product? The Kimi API, direct.
- Comparing several models behind one key? A router, accepting the margin.
- Hard data-residency rules and a GPU fleet? Self-host.
More on Kimi K3
Start with the complete Kimi K3 guide for the overview, or go deeper:
Ready to go deeper?
Read the full Kimi K3 guideFrequently Asked Questions
How do I get access to Kimi K3?
Four ways: the Kimi consumer apps (no API key needed), the Kimi API directly at https://api.moonshot.ai/v1 with model ID `kimi-k3`, third-party routers such as OpenRouter, or self-hosting the open weights. Most developers want the direct API; most individuals want the apps.
Do I need to rewrite my code to use Kimi K3?
Usually not. The API is both OpenAI-compatible and Anthropic-compatible, so existing code against either SDK typically needs only a base URL and model ID change. Expect to re-tune prompts rather than rewrite plumbing — reasoning defaults differ, and K3's cannot be disabled.
Is Kimi K3 available on AWS, Azure, or Google Cloud?
Not as a first-party managed model the way Claude is on Bedrock and Vertex AI. You reach K3 through Moonshot's API, a third-party router, or by deploying the open weights on your own infrastructure — which can of course be inside any of those clouds.
