Services aimed at AI coding work, with subscription and monthly plans adapted to clients such as Claude Code, Codex CLI, Cursor, Cline and Roo Code.
CCSub relays AI model APIs from inside mainland China, so requests do not have to leave the regi…
This category is organised not by model but by how you use it: these services are adapted specifically for AI coding clients — Claude Code, Codex CLI, Cursor, Cline, Roo Code, Continue and others.
"Adapted" means three specific things: supporting the protocol the client speaks (Anthropic native or OpenAI Responses); handling streaming properly, since coding is unusually sensitive to time-to-first-token; and supporting tool calls with long context, because a codebase context easily runs to hundreds of thousands of tokens. Miss any one of them and the client will misbehave on a real project — often not at the start, but once the context grows.
The most useful reference when choosing is what other people have actually run through it. Our verification status and API probe results can rule out services that plainly do not support the protocol, but in the end you should test with your own project.
Pay attention to the pricing model as well. Coding tools consume tokens continuously and in volume, so the difference between a subscription (a fixed monthly allowance) and pay-as-you-go becomes very visible under heavy use. Estimate your monthly usage before choosing.
The picks below are computed live from the current measurements, not fixed conclusions — change the data and the picks change with it. Each states its basis, and you can open the detail page to check it yourself.
The top 1 services by overall score in this category. 1 of them are confirmed to expose a /v1/models endpoint — that figure comes from probing, not from the services' own claims.
| Service | Region | Uptime | Latency | Models | Payment | API |
|---|---|---|---|---|---|---|
|
CCSub
ccsub.net
|
China direct | 80.33% | 4687 ms | 32 | WeChat Pay · Alipay · USDT | Has /v1/models |
The questions we see most often about this category. The guidance applies to the category as a whole, not to any single service.
Check the protocol, not the marketing copy. Claude Code needs /v1/messages; Codex CLI needs /v1/responses. Send one minimal request to each endpoint — a normal response means the protocol is implemented. Our detail-page probe results are a reasonable first filter.
Because perceived latency in a coding tool is mostly time-to-first-token. If a service buffers the whole response before returning it, the editor feels sluggish even when the total time is identical. Test by sending a request that produces a long answer and watching how quickly the first character arrives.
It is one of the easiest traps to fall into. Context grows fast as a project grows, and some services behave perfectly on short requests until the context gets long, then start erroring or dropping content. Test with your own project in its later stages, not with a hello world.
It depends on how steady your usage is. Large, steady volume makes a subscription cheaper at the margin; volatile usage, or a fresh start, makes pay-as-you-go safer because you are not paying for an allowance you never use. Convert last month's actual consumption into both models before deciding.
This page shows you what exists; the pages below show you who is more stable, faster and cheaper. Read together, they tell you more than either alone.