Codex on the Vana API relay: GPT-6.1 Sol at $0.1389 per million

Daily workEN2026-11-02

§ One address, three protocols, and Codex

Freelancers who use Codex usually have one practical question: where does the request go, and what does it cost at the end of the month? Vana API is built around a short answer. There is one API address, https://api.vanaapi.com, and three protocol shapes sit behind it. That address answers API requests only. The dashboard, where you create keys, check your balance and read the usage log, lives on the main site, https://vanaapi.com.

The setup is deliberately small. Create a key in the dashboard, top up a prepaid balance, and point your client at https://api.vanaapi.com. There is no monthly fee, and calls are charged per real token as they happen. Every call is listed in the usage log in the dashboard, so a freelancer working across several projects can see what each request actually cost.

For Codex specifically, the route is the OpenAI format. GPT models answer on /v1/chat/completions, and they also answer on /v1/responses, which is the endpoint Codex uses. That means the model list for Codex runs through the OpenAI-compatible side of Vana API, where the catalogue includes GPT, Grok, DeepSeek, GLM, Kimi, Qwen, MiniMax and MiMo.

As a concrete detail, GPT-6.1 Sol (model id gpt-6.1-sol) is listed at $0.1389 per million input tokens and $0.8333 per million output tokens on Vana API, against an official list price of $2 / $10. The example that follows shows what that means for a working day rather than a benchmark.

§ A worked example: a day of Codex work

Say you run this every day. It is an example, and the numbers come straight from the published price list. You send 100,000 input tokens and 50,000 output tokens to GPT-6.1 Sol through Codex on one afternoon.

Input: 100,000 tokens is 0.1 of a million. 0.1 × $0.1389 = $0.01389.
Output: 50,000 tokens is 0.05 of a million. 0.05 × $0.8333 = $0.041665.
Example total: $0.01389 + $0.041665 = $0.055555.

That is the whole arithmetic for the afternoon. No seat fee, no subscription line, no minimum volume. If the day is quieter, say 20,000 input tokens and 10,000 output tokens, the example becomes 0.02 × $0.1389 = $0.002778 for input and 0.01 × $0.8333 = $0.008333 for output, for an example total of $0.011111.

Now compare the shape of the same example on GPT-5.6 Sol (model id gpt-5.6-sol), priced at $0.3472 input and $1.81 output per million. For 200,000 input tokens and 100,000 output tokens, the example is 0.2 × $0.3472 = $0.06944 plus 0.1 × $1.81 = $0.181, giving an example total of $0.25044. Both examples are simple because Vana API bills on real tokens and records every call in the usage log.

For a freelancer, that visibility matters as much as the rate. You can open the dashboard, look at the usage log, and see which client, which model and which request produced the charge. When a client invoice needs a breakdown, the log is already there.

§ Choosing the right route for each model

One address does not mean one blind endpoint. Vana API separates three protocols, and each one has a matching client.

Claude models: only from official Claude Code or Claude Desktop, version 2.1.285 or later, which speak Anthropic Messages (/v1/messages) themselves. Other clients, any SDK, cURL and scripts are not supported for these models.
Gemini models: only from Gemini CLI, over the Gemini native protocol (/v1beta/models/...). The Gemini native protocol is not OpenAI-compatible, and Gemini CLI is the only supported client for these models.
GPT, Grok, DeepSeek, GLM, Kimi, Qwen, MiniMax, MiMo: OpenAI format (/v1/chat/completions), from Cursor, Cline, Roo Code, OpenCode, Cherry Studio, any OpenAI SDK or cURL. GPT models also answer on /v1/responses, which is what Codex uses.

For a freelancer who lives in Codex, the third line is the one that matters most. You keep your editor, you point it at the Vana API address, and you pick a model id from the list. GPT-6.1 Sol, GPT-5.6 Sol and GPT-6 Astra (model id gpt-6-astra, $0.6944 input and $3.54 output per million) all sit behind that same OpenAI format.

The wider catalogue is there too. Grok 4.7 is $0.4861 / $1.53 per million, DeepSeek V4 Flash is $0.2778 / $1.04, GLM-5.3 is $0.5556 / $1.67, Kimi K3 is $1.04 / $5.21, Qwen3.8 Max is $0.6944 / $2.08, MiniMax M3 is $0.1389 / $0.5556, and MiMo V2.6 Pro is $0.2083 / $0.3472. All are available through the OpenAI format from the clients listed above.

§ Plans, credit and the money side

Beyond pay-as-you-go usage, Vana API offers monthly plans, each scoped to a single model family. For GPT work, the plan is GPT up to $4,000 / 30 days: $103.61 for 30 days, at most $1,200 a week, GPT models only. For Claude work, there are two options: Claude up to $1,000 / 30 days at $64.72 for 30 days, at most $320 a week, Claude models only; and Claude up to $2,400 / 30 days at $153.61 for 30 days, at most $750 a week, Claude models only.

Signing up adds $0.10 of credit, which is enough to make a first request and see how the usage log fills in. If the prepaid balance is topped up and then left unused, the unused balance can be refunded within 3 days of the top-up, minus what was used and the payment processor's fee. That is a short window, so it is worth keeping an eye on the balance rather than topping up far ahead.

There is also a referral programme. You receive 14% of what people you invite pay, and 5% one level further. Earnings can be withdrawn from $13.89. For a freelancer with a network of other independents, that is a small addition to the same account that already handles API spend.

The practical picture is straightforward. Keys come from the dashboard, balance comes from a top-up, and every call is charged per real token with no monthly fee. When the usage log looks unfamiliar, the model id and the rate are both published, so the numbers can be checked line by line.

§ Why the setup stays small

Minimal setup is not a slogan here, it is a consequence of the shape. One address, https://api.vanaapi.com, serves three protocols. The dashboard sits separately at https://vanaapi.com, so the request path and the account path never get tangled. A key created in the dashboard is the only credential you need to carry across clients.

For a Codex user, the routine is short. Add the key, set the base address to https://api.vanaapi.com, choose the OpenAI format, and select a model id such as gpt-6.1-sol or gpt-5.6-sol. If you also work with Gemini, the same address handles that traffic over its native protocol from Gemini CLI. If a project needs Claude, the route is Anthropic Messages from official Claude Code or Claude Desktop, version 2.1.285 or later.

What you gain from that arrangement is choice without extra plumbing. A single balance funds calls across the OpenAI format, the Gemini native protocol and Anthropic Messages. A single usage log records them all. And the cost of any single request can be worked out from published per-million rates, as the examples above show, without waiting for an invoice to arrive.

That is the whole offer: one address, three protocols, minimal setup, and honest token pricing you can check in the dashboard. For a freelancer who wants Codex to run smoothly and the bill to stay legible, Vana API keeps the moving parts to a key, an address and a model id.