Author(s): Chew Loong Nian – AI ENGINEER

Originally published on Towards AI.

Why this matters right now

A Chinese lab just shipped a terminal coding agent that does almost everything Claude Code does, released the entire thing under the MIT license, and pointed it at a model that costs $0.60 per million output tokens. Claude Code’s default model, Opus 4.8, costs $25 for the same million tokens. That is roughly 42 times more expensive on the part of the bill that actually hurts. I spent the morning reading Moonshot’s repo line by line, pulling the install script apart, and trying to find the catch. The catch is smaller than you would expect.

The article explains that Moonshot’s terminal agent is called Kimi Code CLI and is truly open to build and distribute under MIT, with an emphasis on how it differs from other “terminal coding agents” that are moving toward closed-source distribution. It details what the CLI does (single-binary install, a TUI, OAuth/API key login, and a workflow that reads/edits code and runs shell/web tasks), then highlights standout capabilities such as video input, conversational MCP configuration, built-in subagents, lifecycle hooks, and editor integration via an Agent Client Protocol. It argues the core value is not just the agent wrapper but the pricing and model loop: Kimi Code CLI runs on Kimi K2.6 open-weight models with dramatically cheaper output-token costs, making agent usage far more affordable than Claude Code’s and others’ more expensive proprietary setups. The piece compares how each incumbent approaches openness and licensing, concludes that Kimi Code CLI is an open alternative aligned with developers’ need to avoid lock-in and pricing shocks, and ends with a “verdict” that while the repo is still young and not for everyone (e.g., enterprise constraints), the direction toward open, MIT-licensed tooling is the most durable edge for 2026.

Read the full blog for free on Medium.

Published via Towards AI