Kimi Code CLI
An open MIT shell; Moonshot access isn't restricted from Russia.
Verified: Sep 13last change: Sep 23next price check: Dec 12
Facts with sources
- The coder / explore / plan subagents run in isolated contexts. Source (in Russian)
- Access via OAuth or key; the bill depends on how much the agent read and wrote. Source (in Russian)
Strengths
- Single binary for macOS, Linux and Windows, no Node.js, MIT licence; sign in with a Kimi membership or a Moonshot API key.
- In AI Coding Daily's test the same K3 model cost $0.49 in Kimi Code versus $0.85 in OpenCode with the same result: all tests passed.
- kimi web starts a local browser UI with full session history; in swarm mode four subagents added an avatar upload field in parallel.
- Picks up the same AGENTS.md and .agents skills other agents use with no setup; accepts screen recordings as input and MCP servers.
Weaknesses
- Em Dash cancelled after 6 months: since the K3 launch responses are slow and quota drains fast, and moving from $39 to $99 changed nothing.
- Plan mode asked two questions or none; in auto-approve mode the plan approved itself and the coder agent started without asking.
- No cost or elapsed time shown for a session; spend on the Moonshot platform dashboard lags by about 10 minutes.
- Plan terms changed without explanation, "10x agent credits" became "5x"; new paid memberships have been waitlisted since 2026-07-19.
Example from practice
Hardening a CSV contact importer in Laravel
The reviewer gave the same prompt to Kimi Code and to OpenCode, both on K3: make a Laravel contact CSV importer robust to UTF-8, odd line endings and other edge cases. Both runs passed the hidden evaluation tests with no errors (one memory-dependent test skipped). Kimi Code cost $0.49 against $0.85 through OpenRouter, OpenCode took 8 minutes 19 seconds, and the Kimi Code transcript was about half as long. The video is sponsored by Moonshot; the reviewer states the company had no say over the results.
source: AI Coding Daily, I Tried Kimi Code with Kimi K3: Deep-Dive Review
What reviewers say
Reviewer opinions on their own tasks and hardware, not measurements. Links go to the videos; transcripts live in docs/raw-data.
- I Tried Kimi Code with Kimi K3: Deep-Dive Review · AI Coding Daily, Aug 13The reviewer (sponsored by Moonshot) finds K3 a stable model and Kimi Code rough around the edges, but was impressed by kimi web and swarm mode.
- Kimi Code Plan after 4 months of use (Honest Review) · Em Dash, Sep 4In the reviewer's view Moonshot builds excellent models but cannot serve them; after six months he cancelled the subscription and moved to another provider.
Who it is for
- Engineering team in Russia without a security sign-off
Models in this agent
Score is from our manager-task benchmark; models without a score are not measured yet.
What changed
- Fireworks released Ember-1, a Kimi K3 derivative that matches its quality with 40% fewer tokens →
- Kimi K3 arrived on Amazon Bedrock – the first open-weights model there with explicit prompt caching →
- Moonshot open-sourced the Kimi K3 weights, 2.8 trillion parameters: the largest open-weights release to date →
How to brief an agent like this
The four-line brief is the Foundation module: prompt structure, LLM agents and MCP, an AI policy for the team. It starts with the free open module.