Using Codex from China: config, models and Alipay top-ups

The two OpenAI texts sit on different pages: the list of API-supported countries and territories on developers.openai.com has no mainland China entry, and the help.openai.com article archived on 2026-10-07 states that "If a location is not included in the list below, our API is not supported there" and that no separate list of unsupported countries and territories is published. This page quotes those sentences as written and draws no further conclusion. Configuration is a separate matter: the official docs show how to point model_provider at a provider table you define, connecting Codex to any endpoint that speaks Responses.

Two places: the top-level model_provider points at your custom provider id, and you then define a table for that id under model_providers with base_url, wire_api and env_key. The official reference says responses is the only supported wire_api value and the default when omitted; env_key is the environment variable supplying the provider API key; requires_openai_auth means the provider uses OpenAI authentication, defaults to false, and makes Codex ignore env_key when it is true. You may name the table anything except openai, ollama or lmstudio, which the official docs say are reserved built-in IDs that cannot be overridden.

For QCode it is https://api.qcode.cc/openai, with no /v1 and no trailing slash. QCode's guide states the Codex base_url must be exactly that value and that Codex uses the Responses protocol. The endpoints page adds that /api and /openai/v1 are path prefixes rather than standalone endpoints, and that SDKs append /v1/messages, /chat/completions or /responses themselves. In the official docs the base_url field means the API base URL for the model provider.

No. QCode's docs say the protocol is decided by the request path and that not every model family can use every path: the Responses path Codex uses serves the GPT family and explicitly does not serve the Chinese model families, Claude models only work over the Anthropic protocol, and GPT models cannot use the Anthropic endpoint. So in Codex you use gpt-6.1-sol, gpt-6-sol, gpt-5.6-sol or gpt-5.6-terra, and for Claude you switch to a client that speaks Anthropic. QCode's docs state that the same key also works for Claude Code and that the key itself does not distinguish protocols.

Top up in CNY with Alipay or WeChat Pay, and activation after the top-up is self-service and immediate. The payment methods page routes mainland China and CNY payments to Alipay, with the fee not passed through, and WeChat Pay as the alternative; both credit the CNY balance. Its printed limits are CNY top-ups from ¥100 and USD top-ups from $1 to $2,000, while the site's top-up page lists a ¥2,000 per-payment cap for Alipay, about ¥900 per payment for WeChat Pay, whose checkout page also accepts cards. Other options: Visa / Mastercard / AMEX through the international checkout with the balance credited immediately, Apple Pay or Google Pay using the card already bound in your wallet, or cryptocurrency, which covers many coins and networks and is verified by hand after the on-chain transfer, so a small first transfer is advised.

Billed per token; per-model rates are listed on /models. QCode's guide states that costs are calculated from actual token consumption and not per tool, and that the same key also works for Claude Code; which models are offered and at what rate follows the /models page. Use /cost for the tokens used in a session and the console for overall quota.