ChatModelsPricingDocsFor agents
Log inGet API Key
×

Use Kimi with Claude Code

$3/Mtok in

Kimi K3's tool-calling is tuned for agentic work, which makes it the strongest open-weights brain for Claude Code sessions, at a fraction of frontier pricing.

Get a key, free starter creditsKimi K3 API details

Facts

At a glance

Model

Kimi K3

moonshotai/kimi-k3

Price

$3 / M input · $15 / M output

Context window

256k tokens

Tool

Claude Code

Setup below, under two minutes

Quickstart

Set up Kimi in Claude Code

  1. 1

    Point Claude Code at CompanyFabric. Claude Code reads the ANTHROPIC_* variables, the value is your CompanyFabric key (sk-cf-…), no Anthropic account needed:

    export COMPANYFABRIC_API_KEY="sk-cf-…"
export ANTHROPIC_BASE_URL="https://api.companyfabric.com"
export ANTHROPIC_API_KEY="$COMPANYFABRIC_API_KEY"
  2. 2

    Start a session on this model:

    claude --model moonshotai/kimi-k3
  3. 3

    Every turn's exact cost lands in your dashboard usage feed, set a per-key budget cap if the agent runs unattended.

Full tool guide: Use any model with Claude Code →

FAQ

Frequently asked questions

Do I need a Moonshot AI account to use Kimi in Claude Code?

No, one CompanyFabric key covers Kimi K3 and every other model in the library. Billing is one prepaid balance with exact per-call metering; BYOK at 0% is available if you already have provider keys.

What does Kimi K3 cost in Claude Code?

$3 / M input · $15 / M output, the same metered price as any other route to the model. Set a per-key budget cap to bound what the tool can spend.

Can I switch models later without touching Claude Code again?

Yes, the endpoint stays the same; the model is a string. Any catalog id works in the same configuration, including cheaper tiers for bulk work.

Explore

Related setups

Use Gemini with Claude Code

Claude Code's agent loop with Gemini's million-token context: point the harness at a model that can hold your whole repo in one window. The messages translation layer makes non-Anthropic models a --model flag, not a different tool.

Use DeepSeek with Claude Code

The cheapest way to run long agent sessions: DeepSeek's commodity token prices under Claude Code's agent loop. Ideal for bulk refactors and test generation where volume dwarfs per-token quality differences.

Use GPT-5.5 with Claude Code

Run OpenAI's flagship inside Claude Code for a second opinion on hard problems, same session, same tools, different reasoning style. Useful as a review pass over changes drafted by another model.

Use Kimi with Cline

The open-weights pick for Cline: K3's agentic tool-calling handles the act loop, and the price difference compounds over hundreds of daily tool calls.

One key. Every model. Exact prices.

Run Kimi in Claude Code with free starter credits. Free starter credits included, no subscription required.

Get your API keyBrowse all models