How to Use Kimi K3 in Claude Code: Cheap Coding Setup (2026)

TL;DR: Kimi K3 is Anthropic-compatible. Set ANTHROPIC_BASE_URL to Moonshot's Anthropic base and ANTHROPIC_AUTH_TOKEN to your Moonshot key, then pick a Kimi K3 model, no translating proxy, unlike OpenAI's GPT or xAI's Grok. It's a strong value pick with large context. It's still a global endpoint swap, so to move back to Claude you edit env vars and restart, unless you use an agent that treats the model as a setting.

Kimi K3 + Claude Code: no proxy needed

This is the easy camp. Moonshot publishes an Anthropic-compatible endpoint for Kimi, so Claude Code can point directly at Kimi K3, no translating proxy, unlike GPT (OpenAI-compatible) or Grok (OpenAI-compatible), which both need a translator in front. You bring your own Moonshot key and usage is billed to you.

The config

What to set

ANTHROPIC_BASE_URLPoint it at Moonshot's Anthropic-compatible base URL (the /anthropic endpoint).
ANTHROPIC_AUTH_TOKENYour Moonshot API key, usage billed to your Moonshot account.
ModelSelect a Kimi K3 model using the exact name the endpoint exposes.

Model names, versions, exact pricing and install commands change frequently, always check the provider's official documentation for the current details.

Why Kimi K3 for coding

Cheap, capable

The Kimi line is a popular capability-per-dollar pick for everyday agent coding.

Large context

Kimi is known for big context windows, handy on larger codebases in one session.

No proxy tax

Anthropic-compatible, so you skip the extra translating service that GPT and Grok need.

The cleaner option: Kimi K3 as a one-click setting

The endpoint swap works, but Claude Code is single-provider by design, so mixing Kimi and Claude means reconfiguring each time. If you want to run Kimi K3 most of the time and jump to Claude on demand, use an agent where the provider, base URL, key and model are settings you pick from a list, and where Kimi, being Anthropic-compatible, just works with no proxy at all.

What that looks like with Wuwei

Kimi K3 in the model list

Pick it in the UI; Claude, DeepSeek, Qwen, GLM and OpenAI-compatible endpoints sit alongside.

No proxy, and one-click switch

Kimi works natively (no translator, unlike Grok); move between Kimi and Claude mid-task without editing config or restarting.

Free, open source, local

MIT-licensed, runs on your machine, reads your real files; you only pay Moonshot's usage.

FAQ

Can Claude Code use Kimi K3?

Yes. Moonshot exposes an Anthropic-compatible endpoint for Kimi. Set ANTHROPIC_BASE_URL to Moonshot's /anthropic base and ANTHROPIC_AUTH_TOKEN to your Moonshot key, then select a Kimi K3 model, no proxy needed.

Do I need a proxy for Kimi K3 in Claude Code?

No. Kimi is Anthropic-compatible, so Claude Code points straight at it, unlike OpenAI's GPT or xAI's Grok, which are OpenAI-compatible and need a translating proxy.

Is Kimi K3 good for coding agents?

The Kimi line is a popular value pick for agent coding, with strong capability per dollar and large context windows that help on bigger repos.

How do I switch between Kimi K3 and Claude easily?

Use an agent where the model is a setting. Wuwei is free and open source, Kimi K3 is already supported and works natively (no proxy), and switching to Claude is one click instead of an env-var edit and restart.

How to use Kimi K3 in Claude Code