MoonshotAI (Kimi)

MoonshotAI's Kimi K3 is a 2.8 trillion parameter open-weight multimodal reasoning model. It ranks #1 in multiple coding and agentic arenas, with a 1M token context window and strong performance on long-horizon tasks.

Setup

# 1. Get your API key from https://platform.moonshot.cn/
export KIMI_API_KEY=sk-...

# 2. Verify detection
arcana doctor

# 3. Run
arcana run "hello"

Available Models

Highlights

• #1 in coding arenas — top-ranked on SWE-bench, Terminal-Bench, and web app development
• 1M token context — analyze entire codebases, long documents, or multi-file refactors
• Multimodal — accepts text, images, and video input
• Always-on reasoning — deep thinking is enabled by default for complex tasks
• Open-weight — can be self-hosted for privacy-sensitive workloads

Config File

{
  "provider": "moonshotai",
  "model": "kimi-k3"
}

Environment Variables

Via OpenRouter

You can also access Kimi K3 through OpenRouter with a single key:

export OPENROUTER_API_KEY=sk-or-...
# Then use kimi-k3 directly — Arcana routes through OpenRouter

Pricing Notes

Approximate costs per 1M tokens via MoonshotAI API:

Tips

• kimi-k3 excels at long-horizon coding tasks — give it entire repositories to work with
• coding-kimi-k3 is much cheaper and great for everyday code generation
• Kimi K3's 1M context makes it ideal for analyzing large codebases or document sets
• Use kimi-k3 as a strong alternative to Claude Opus for complex agentic work