RelayVano API relay for Codex: one address, one key, GPT models

Coding-agent routes EN2026-11-01

What one address and one key changes for Codex

Pointing Codex at RelayVano comes down to two pieces of information: a base address and an API key. Everything else stays in the tool you already use, with the same chat window, the same workflow and the same habits. You are not learning a new client, and you are not moving your prompts somewhere unfamiliar. You are simply telling Codex where to send requests and how to authenticate them.

That simplicity is deliberate. RelayVano is an independent relay made for developers worldwide, and the whole site, dashboard and docs come in six languages: 中文, English, Español, हिन्दी, Português, Français. Prices show in yuan on the Chinese pages and in approximate dollars on the others, so a reader in Beijing and a reader in São Paulo can each work in the language and currency display that suits them without guessing what a token costs.

One API key reaches every model through https://api.relayvano.com, using three protocol paths: OpenAI-compatible /v1, Anthropic /v1/messages and native Gemini /v1beta. For Codex, you work with the OpenAI-compatible /v1 path. The practical benefit is that your key stays the same while the model behind it changes, so a test run on one GPT model and a later switch to another does not mean a second account, a second bill or a second round of configuration.

There is also a clear reason to try the setup before committing much budget: new accounts start with $0.40 of credit. That is enough to send a handful of real requests through Codex, see the responses arrive, and open the usage page to see exactly what each call was charged.

How Codex fits the RelayVano model rules

RelayVano does not pretend that every model speaks every protocol. The pairing rules are simple, and they are the same everywhere on the platform, so you can plan around them instead of discovering them through failed requests.

Codex, Cursor, Cherry Studio, Cline, OpenCode, OpenAI SDKs and cURL pair only with GPT, DeepSeek and the other non-Claude, non-Gemini models.
Claude models work only in the official Claude Code and Claude Desktop clients, version 2.1.285 or newer.
Gemini models use the native Gemini protocol only, in Gemini CLI, and they are not OpenAI-compatible.

For your purposes, that means Codex talks to the GPT family and to DeepSeek, and to the other non-Claude, non-Gemini models in the catalogue. There are no exceptions to that mapping, so if a model name does not appear in the Codex-compatible group, spending time on it in Codex is time you do not need to lose.

The rule cuts both ways, and it is worth being precise. Claude models are not usable through Cursor, Cherry Studio, Cline, OpenCode, any SDK or cURL, so do not plan a Claude workflow inside Codex. Gemini models are not reachable through the OpenAI-compatible path either, so a Gemini request belongs in Gemini CLI with the native protocol. RelayVano keeps those boundaries because the clients and the protocols genuinely differ, and a clean mapping is easier to reason about than a flexible one that fails in confusing ways.

Inside Codex, your shortlist is therefore straightforward: GPT-5.6 Sol, GPT-6 Astra, GPT-6.1 Sol, DeepSeek V4 Flash, DeepSeek V4 Pro, Qwen3.7 Plus, Qwen3.8 Max, Kimi K3, MiniMax M3, MiMo V2.6 Pro, Hunyuan 4, Grok 4.6 and Grok 4.7. Which one you pick depends less on fashion than on the task and the per-token fare you are comfortable with.

Setting up Codex on RelayVano in plain steps

First, create your RelayVano account and sign in to the dashboard. The account is yours to use from anywhere, and the interface, documentation and support all follow the six-language setup described above, so nothing in the process depends on a particular working language.

Next, generate an API key from your dashboard and copy it somewhere safe. This single key is the credential you will use for Codex, and it is the same key that works across the models Codex supports. You do not need a separate key per model, per project or per prompt size.

Then open your Codex configuration and set the base address for the provider to the RelayVano OpenAI-compatible endpoint at https://api.relayvano.com/v1, and place your key in the environment variable or settings field Codex uses for the API key. If your setup already points at another OpenAI-compatible endpoint, this is usually a two-field change: the address and the key. Nothing else about your Codex setup has to change.

After that, choose a model from the compatible group, send a short test prompt, and check the usage page in the dashboard. The usage page lists every request with its charge, which means you can confirm the connection works and read the actual cost of the request rather than estimating it from memory. If something looks off, logged-in members can write to us under Messages in the dashboard, so a question about a specific request has a natural place to go.

What pay-as-you-go costs on RelayVano

RelayVano sells one thing: pay-as-you-go balance. There is no monthly fee, no subscription commitment and no expiry on the balance you add, so you can start small, top up when a project picks up pace, and stop without leaving unused credit stranded behind a recurring charge.

Prices below are approximate US dollars per million tokens, with the actual charge made in yuan. Prompt tokens, reply tokens, cache stores and cache hits are priced separately, so a workload that reuses a long context can be much less expensive than one that sends it fresh every time. A few fares relevant to Codex work:

GPT-6.1 Sol: $0.1958 prompt / $0.9792 reply / $0.0389 cache hit, against an official list price of $2 / $10 / $0.1 cache hit.
GPT-5.6 Sol: $0.3736 prompt / $2.09 reply / $0.0306 cache hit, against an official list price of $4 / $20 / $0.4 cache hit, with that OpenAI promotion stated to run at least until 2026-11-21.
GPT-6 Astra: $0.7403 prompt / $3.70 reply / $0.0736 cache hit, against an official list price of $10 / $50 / $1 cache hit.
DeepSeek V4 Flash: $0.3097 prompt / $1.24 reply / $0.008333 cache hit, and DeepSeek V4 Pro: $1.37 prompt / $4.10 reply / $0.0458 cache hit.

Here is an example, not a forecast. Suppose you send 200 requests a day, each with 1,000 prompt tokens and 500 reply tokens on GPT-6.1 Sol, and you do not use caching. The prompt side is 200,000 tokens, which works out to about $0.0392, and the reply side is 100,000 tokens, which works out to about $0.0979. The example total is roughly $0.137 per day. The same 200 requests on a pricier model would cost more, and the same requests with heavy cache hits would cost less, which is exactly why the usage page is worth checking as your patterns settle.

One detail that matters for long-running agent sessions: if a caller disconnects mid-stream, the request is still finished and billed on the tokens really used. A dropped connection therefore does not mean the work vanishes, and it does not mean you pay for tokens that were never generated. You are charged for what the model actually produced.

Top-ups, referrals, refunds and everyday support

Adding balance happens through Stripe checkout. Cards work everywhere, and WeChat Pay is available where your account supports it. Because RelayVano is prepaid rather than subscription-based, the size of each top-up is entirely your decision, and the balance you have never expires. If you overestimated a month of usage, unspent top-ups can be refunded within 9 days.

There is also a referral program built around the same pay-as-you-go model. People who sign up through your link generate 16% of every top-up for you, and 4% one level further. Earnings can be withdrawn from $5.56 or moved straight into your balance, which makes it easy to put a referral reward to work on real requests instead of letting it sit as a separate pot.

Support stays inside the product you are already using. Logged-in members write to us under Messages in the dashboard, which keeps questions, key issues and billing checks attached to the account that needs them. For a setup that involves one address, one key and a short test request, that is a short path from problem to answer.

RelayVano gives Codex users a straightforward way to work with GPT and DeepSeek models on a pay-as-you-go balance, with fares shown per million tokens, charges itemised per request and a six-language experience for developers worldwide. Start with the $0.40 of credit in a new account, point Codex at the RelayVano endpoint, send a test prompt and read the first line of usage data. If it looks the way you expected, sign up and top up when you are ready.