Kimi Code membership guide
Kimi Code is a developer-focused benefit within the Kimi membership plan, providing high-performance AI coding capabilities. You can use this benefit through Kimi Code CLI, Claude Code, Roo Code, and other supported tools.
Key advantages
| Advantage | Description |
|---|---|
| Broad Compatibility | Works with Kimi Code CLI, Claude Code, Roo Code, and other mainstream coding agents |
| Standard / HighSpeed Tiers | The same model at two speeds — HighSpeed delivers roughly 5–6× the output speed of Standard and switches on demand |
| Ultra-Fast Responses | Generation speeds up to 100 tokens/s, significantly boosting coding efficiency |
| High-Frequency Concurrency | Approximately 300–1,200 requests per 5-hour window (depending on your plan), with up to 30 concurrent streams |
Quick start
Choose the path that fits your situation:
- New Users: Go to kimi.com/code, sign in, and subscribe to a Coding Plan.
- Existing Subscribers: Access the console to manage your API Keys and start using Kimi Code.
Obtaining an API key
- Sign in to the Kimi Console.
- Navigate to the API Keys page.
- Click Create New API Key.
- Copy and securely store your API Key (it is only displayed once upon creation).
Do not share your API Key with others or commit it to public code repositories.
One-click login
In Kimi Code CLI, you can use the /login command for quick authorization without manually copying an API Key:
The system will automatically complete device authorization and account binding — the entire process takes just a few seconds.
Device management
- Each account can be used across multiple devices.
- Device authorizations that have been inactive for 30 days will automatically expire; you will need to run
/loginagain to re-authorize. - You can view and manage authorized devices in the console.
How to switch models
The HighSpeed model is now available. Kimi Code offers two tiers — Standard and HighSpeed — built on the same model with identical coding ability, sharing the same Base URL, API Key, and membership benefits. HighSpeed delivers roughly 5–6× the output speed of Standard, so when you want instant responses and fast iteration, a one-click switch gives you a smoother coding experience. The main differences:
| Item | Standard | HighSpeed |
|---|---|---|
| Model ID | kimi-for-coding | kimi-for-coding-highspeed |
| Output speed | Baseline | ~5–6× faster than Standard |
| Credit usage | Baseline | ~3× that of Standard |
| Coding ability | Full | Same as Standard |
| Best for | Everyday coding tasks | Instant responses, fast iteration |
| Membership | Available to all Kimi Code members | Requires an Allegretto plan or above |
Ways to switch to the target model:
- Official Kimi Code CLI: type
/modelin a session to switch directly between Standard and HighSpeed — no config changes needed. - Kimi Code for VS Code: pick the target model from the dropdown menu in the input bar; if HighSpeed isn't listed yet, restart VS Code or reinstall the extension.
- Third-party tools: set the tool's Model ID to the target model; all other settings stay the same. For where to find it in each tool, see Using in Third-Party Coding Agents.
- Stable model IDs: Both IDs are stable identifiers; the backend updates the model they map to as models improve, with no client config changes needed.
- Enter it exactly: The high-speed ID must be
kimi-for-coding-highspeed. If mistyped or set to another value, the request silently falls back to the standardkimi-for-coding— no error, but no acceleration either. - 401 without access: If your plan doesn't include HighSpeed access, calling it returns a
401; upgrade to Allegretto or above.
Why doesn't the whole task feel 5–6× faster? "5–6×" refers to model output speed (how fast text/code is generated). A coding task's total time is made up of "model output + tool calls (reading/writing files, running commands, web lookups, etc.) + script execution" — how long tool calls and script execution take depends on your project and commands, and HighSpeed doesn't change that part. So if the overall task doesn't feel 5–6× faster, it's usually because tool calls / script execution took up most of that turn, not because model generation slowed down.