Kimi K3
Take the hardest problems here: large refactors, unfamiliar code, and long debugging sessions.
Coding Plan
Kimi K3, GLM-5.3, and DeepSeek V4.1 Flash in the coding tool you already use. No 5-hour window, no weekly quota, no monthly token allowance. Give them the long, parallel work: tests, refactors, migrations, the runs that go on all day.
The current Coding Plan lineup includes Kimi K3, GLM-5.3, and DeepSeek V4.1 Flash. Work with them in your terminal or in an app, and switch to suit each task. Model availability and selection may change under the Coding Plan terms.
Take the hardest problems here: large refactors, unfamiliar code, and long debugging sessions.
Run most of the day on it: features, fixes, tests, and reviews.
Hand it the quick, simple jobs at the lowest cost: small edits, renames, and boilerplate.
Codex CLI connects through a local Responses proxy described in the docs.
For Claude Code, ask us for setup information.
OpenCode and similar tools connect with the base URL in the docs.
Standard Code is planned.
No monthly token allowance. Run agents for as long as you need; at peak hours, requests may be slower. Made for developers who keep agents running all day, not for occasional use.
Queueing and concurrency conditions, model changes, billing, and training terms are described in the Coding Plan terms.
Your coding tool controls project access and local history.
Standard Thinking’s data terms cover the inference requests sent to us.
Check which files, instructions, and conversation context the tool can use.
Check what persists between tasks and the controls for removing stored context.
Coding Plan Inputs and Outputs may be used for our own model training and evaluation as described in the Coding Plan terms. Standard API and Product Plan inference covered by ZDR is not used for training. Read the Coding Plan terms.
Coding Plan is designed for a developer’s own work and applications, subject to the access configuration in the applicable order. For a product workload that needs a separately agreed base allocation, auto scaling, or an uptime SLA, request information about Product Plan; terms are set in the order.
Coding Plan customers may serve an application’s end users under the applicable order, and under the same general conditions as Standard API, subject to the access configuration confirmed for their subscription. For a separately agreed base allocation, auto scaling, or an uptime SLA, Product Plan is the next step; its terms are set in your order.
You can switch from Weekly to Monthly any time, with unused days credited at your Weekly fee divided by 7 per day against your Monthly fee, and from Monthly to Weekly at renewal. Both renew until you cancel, and access runs to the end of the period you paid for. How coding plans count usage compares the units and reset windows other plans publish.
No 5-hour window, no weekly quota, and no monthly token allowance. Run agents for as long as you need; at peak hours requests may be slower, but a baseline is kept. Read the Coding Plan terms.
Coding Plan inputs and outputs may be used for our model training and evaluation under the Coding Plan terms. Standard API and Product Plan inference under Zero Data Retention is not used for training. Read the terms.
OpenCode and any tool that speaks OpenAI chat completions connect with the base URL. Codex CLI connects through a local Responses proxy described in the docs. For Claude Code, ask us for setup information. Standard Code is planned.