Codex pricing on Relay365: pay-as-you-go with a $0.25 signup credit

Give every project its own tagged pair of Relay365 keys

Welcome to the Relay365 duty desk. Relay365 is an independent third-party relay, run like a round-the-clock duty desk, and every lane sits on one table, so you can compare routes without juggling separate accounts. If your plan is to run Claude Code and Codex side by side, the first hour is well spent on one habit: give each project a tag, and keep the two tools on separate Relay365 keys.

Start in the dashboard and create two keys for the same project. Bind one key to a Claude lane for Claude Code, and bind the other to a GPT lane for Codex. Relay365 API keys support tags, and a key can be bound to a lane, so the tag answers which project a key belongs to while the lane answers which model family it may reach. Tag both keys with the same project name, for example billing-service, and you still keep clean separation between the two tools.

Copy the quick-wiring snippets from the dashboard into your editor rather than typing a base address by hand, and paste the Claude key into Claude Code and the Codex key into your Codex setup. The desk suggests a small naming convention as well, such as a project tag plus a tool suffix, so that a glance at the key list tells you which pairing is live. A new Relay365 account also arrives with a $0.25 signup credit, which is enough for a first smoke test on a short prompt.

Confirm the client and lane pairing before you send real work

Relay365 client rules are firm, and checking them early saves a confusing failed call later. Claude models work only from the official Claude Code and Claude Desktop clients, and Claude Code 2.1.285 or newer is required. If your current setup is older, update the client first, then point it at the Relay365 key you tagged for that project.

Cursor, Cherry Studio, Cline, OpenCode, SDKs and cURL pair only with GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax and Hunyuan lanes. That is exactly why the Codex key deserves its own slot: the GPT lane is the one to bind there, while the Claude key stays reserved for the official clients. Two keys, two lanes, one tag, and no crossed wires.

Gemini runs on its native protocol only, through Gemini CLI, and it is not OpenAI-compatible. A single reply uses :generateContent, and a stream uses :streamGenerateContent?alt=sse. Gemini cannot go through /v1/chat/completions, so if you add a Gemini lane later, give it a separate key and a separate tag rather than reusing the Codex key.

One more desk note for the first hour: if a call returns an error, check the pairing before you touch the model. A lane mismatch looks like a broken key on the surface, and it is far easier to fix than a mis-set rate or a missing tag. Once the pairing is right, move on to the numbers.

Read the four token prices on the same table

Relay365 prices input, output, cache write and cache read separately, so you can see exactly where a request spends. Rates are per 1 million tokens in approximate USD, while the account itself is kept in CNY. For Claude Fable 5.1, wire name claude-fable-5-1, the table shows input $4.00, output $20.01, cache write $4.93 and cache read $0.2375. Against the official list price of input $10, output $50 and cache read $0.25, the Relay365 output rate is about 40% of official.

For GPT-6 Astra, wire name gpt-6-astra, Relay365 lists input $0.7444, output $3.72 and cache read $0.075 per 1 million tokens, against an official list price of input $10, output $50 and cache read $1, so the Relay365 output rate is about 7% of official. Gemini 2.5 Flash, wire name gemini-2.5-flash, shows input $0.2389, output $1.91, cache write $0.2389 and cache read $0.0236, and MiniMax M3, wire name minimax-M3, shows input $0.1528, output $0.6111 and cache read $0.0306. For those two families no official price is quoted, so the desk makes no comparison and simply shows the Relay365 figures. The pricing page carries the full table for every lane.

Worked example on the Claude side: for example, if you send 100,000 input tokens and 20,000 output tokens a day to claude-fable-5-1, the input cost is $0.40 and the output cost is $0.4002, so about $0.8002 a day and about $24.01 across 30 days, before cache write and cache read. On the GPT side, for example, if you send 200,000 input tokens and 50,000 output tokens a day to gpt-6-astra, that is $0.14888 plus $0.186, or about $0.33488 a day and about $10.05 across 30 days. Both examples ignore cache tokens, which are billed on their own rates.

Because cache write and cache read carry their own lines, a project that leans on context caching should be checked against both columns, not just the headline input and output numbers. The desk recommends writing the four figures next to your project tag while you are setting things up, so the next invoice reads like a familiar chart instead of a surprise.

Trace every request in the usage log and the ledger

Relay365 keeps a usage log of the latest 200 requests, and each entry shows the time, model, input, output, cache recall, charge and result of the call. That is enough to answer the questions a duty desk gets asked most often: which model answered, how much was charged, and did the request succeed or fail. When something looks off, this log is the place to start.

One honest caveat: the log does not show which key made the call. That is precisely why the tagged-key habit matters. If your Claude Code key and your Codex key both carry the project tag billing-service, you can at least narrow a charge to that project, and the model column tells you which lane served it. The dashboard also keeps a ledger of wallet, bounty and pass accounts, so prepaid balance, referral earnings and any duty pass stay visible in one place.

The dashboard supports six interface languages, and in-site messages reach the desk directly once you log in. If a model name looks unfamiliar in the log, or a charge does not match your expectation, send a message with the time and model from the log entry. The desk can work from the same record you are looking at.

Check billing, limits and duty pass behaviour in the first hour

Relay365 billing is prepaid through the wallet, with no monthly fee, and every call is settled on the token counts the upstream reports. A call that fails upstream with a 502 or 503 is not billed, while a disconnected stream is billed for what was actually used. Know these two rules before you stress-test a key, because they decide what a failed run actually costs.

There are also two request limits worth writing down: more than 120 calls in 60 seconds on one key, or more than 8 requests in flight for one account, will return HTTP 429. If you script parallel work, keep the in-flight count under that ceiling and space out bursts on a single key. A 429 is a queue signal from the desk, not a broken configuration.

Optional duty passes are available for Claude and GPT only, and each pass comes with a dedicated key that never touches the wallet. The Claude pass runs 14 days with $390 of quota and $210 per week for $50.97, and the GPT pass runs 30 days with $2350 of quota and $700 per week for $81.53. Quota is drawn at the duty-pass rate x2.5, and a call already let through is drawn in full even if it runs past the quota or the 7-day ceiling, with the overrun counting against the earliest-expiring card. When a pass is nearly used up, each running call holds $3 x the draw rate and an extra parallel call gets HTTP 429 until the running ones finish. Because a pass key is dedicated, project tagging is easier to manage on ordinary wallet keys, and a pass is better treated as a deliberate, separate lane for heavy Claude or GPT runs.

Finally, check the money rules while the account is fresh. Refunds are available within 14 days of a top-up, and unused balance can be refunded minus what was consumed and the payment channel fees, while bounty and promotional credit are not refundable. The referral bounty pays 18% at level 1 and 6% at level 2 of what referred members pay, and it can be withdrawn from $11.11 or moved into the wallet. Know which bucket a balance sits in before you move it.

Put the desk on watch for the next shift

An hour of setup buys you something specific: two tagged keys, one for Claude Code and one for Codex, a confirmed client pairing, a clear view of the four token prices, and a usage log that can explain what each call did. Relay365 keeps every lane on one table, so the next step is simply more traffic, watched by the same log and the same desk.

When you are ready, create a Relay365 account, set your project tags, and let the first request through. The desk will be on the other side of the log, ready to help you read the numbers.