GPT-5.6 Sol pay-as-you-go rates on the Relay365 desk

A note from the night desk about the bill you keep losing track of

Dear developer on the late shift, if your AI spend has become something you try not to look at directly, you are not the only kind of person who ends up here. The request goes out, the answer comes back, and somewhere a counter ticks quietly upward. Relay365 was built for exactly that feeling. We are an independent third-party relay run like a round-the-clock duty desk, and our job is to put every lane, every rate and every call in front of you at once.

So here is the short version, written the way the desk would say it out loud. One dashboard holds the API keys, the usage log, the wallet ledger and the duty passes. Six interface languages let you work in the one you actually want to use. And four token types, input, output, cache write and cache read, are priced separately, so you can see where the money is going instead of guessing at a single blended number.

If you have been juggling separate subscriptions, scattered invoices and a mental spreadsheet that never quite balances, this is the desk handing you one table and a log you can actually read. The rest of this letter is a walk through that table, region by region, model by model, with the figures left in plain sight.

Six languages and one table of lanes

Relay365 opens with a choice of six interface languages, which sounds like a small detail until you are the person who never once wants to translate a dashboard at midnight. Pick the language you are comfortable in and the lanes all sit on the same table underneath it. A lane is simply a route to a model family, and Relay365 keeps those routes together so that switching between them is a change of setting rather than a change of platform.

The dashboard gives you API keys with tags, and each key can be bound to a lane. That binding is what keeps a quiet, cheap route from quietly absorbing traffic meant for a heavier one. It also gives the usage log something to describe. Quick-wiring snippets sit close by, so the first call does not require a hunt through documentation while you are still deciding how the rest of the night should go.

• Every lane sits on one table, with input, output, cache write and cache read priced separately.
• Each key can be tagged and bound to a lane, so traffic stays where you put it.
• Six interface languages mean the desk speaks your language, not just the billing system's.

One more practical point before you keep reading. Claude models run only from the official Claude Code and Claude Desktop clients, and Claude Code 2.1.285 or newer is required. Cursor, Cherry Studio, Cline, OpenCode, any SDK and cURL pair only with GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax and Hunyuan lanes. Gemini runs on its native protocol only, through Gemini CLI, using :generateContent for a single reply and :streamGenerateContent?alt=sse for a stream, and it does not go through /v1/chat/completions.

Reading the usage log line by line

Here is the part of Relay365 that tends to quieten the late-night nerves. The usage log keeps the latest 200 requests, and each entry shows the time, the model, the input, the output, the cache recall, the charge and the result of the call. Seven fields, one row, no interpretation required. You can open the log and see the shape of your spend directly: which model answered, how much was read from cache, what it cost and whether it succeeded.

One honest caveat, because a duty desk should not oversell: the log does not show which key made a given call. It shows the traffic itself. That is why the tags and lane bindings matter so much. Give each key a name that means something, bind it to the route you expect, and the log becomes far easier to reason about even without a per-key attribution column.

Behind the log sits the ledger, which keeps wallet, bounty and pass accounts in view together. Messages to the desk live in the same place, in-site, so a question about a charge does not require a new email thread. Sign in first, write to the desk, and the conversation stays attached to the account rather than scattered across inboxes.

Duty passes for the two lanes that need them

Some nights you want a steady route with a ceiling you can reason about instead of a wallet you meter by hand. That is what a duty pass is for. Duty passes are optional and currently cover Claude and GPT; every other model always uses the wallet. A pass comes with a dedicated key that never touches the wallet, which means pass traffic and wallet traffic do not quietly merge into one number at the end of the month.

The Claude pass for 14 days costs $50.97 and draws on up to $700 of quota at the duty-pass rate x2.5, with a weekly limit of $380. At list prices that is at most about $280 of usage, and less on models whose pass rate sits above list. The GPT pass for 30 days costs $81.53 and draws on up to $3000 of quota at the duty-pass rate x2.5, capped at $900 per week, which is at most about $1200 of usage at list prices, and again less where the pass rate sits above list.

There is also a 30-day Claude pass at $94.03 with up to $1400 of quota and a $450 per week limit, but for most people the two passes above are enough to show the shape of the offer. A pass is not a subscription and it is not a promise of extra volume. It is a prepaid block of quota on a dedicated key, with a stated weekly ceiling, so the ceiling is visible before you start spending rather than afterwards.

The rates, and what they add up to in a month

All rates below are per 1 million tokens in approximate USD, and the account itself is kept in CNY. Take Claude Fable 5, whose wire name is claude-fable-5: input is $3.79, output is $18.94, cache write is $4.66 and cache read is $0.2917. Against the official list price of $10 input, $50 output and $1 cache read, the Relay365 output rate is about 38% of official. That is a concrete line on a concrete table, not a vague promise.

Now let us do the arithmetic that actually keeps developers awake. For example, if you send 10,000 output tokens a day to Claude Fable 5, that is 0.01 of a million tokens, so about $0.1894 a day and roughly $5.68 across 30 days. If the same 10,000 output tokens a day go to GPT-5.6 Sol, whose wire name is gpt-5.6-sol, the output rate is $2.09 per 1 million tokens, so about $0.0209 a day and roughly $0.63 across 30 days. Same volume, very different line on the statement.

Other lanes sit in the same table with their own figures. DeepSeek V4 Flash, wire name deepseek-v4-flash, lists input at $0.2889, output at $1.15 and cache read at $0.006944. Gemini 2.5 Flash, wire name gemini-2.5-flash, lists input at $0.2389, output at $1.91, cache write at $0.2389 and cache read at $0.0236. No official price is quoted for these families, so Relay365 publishes no comparison for them. For GPT-5.6 Sol, the official list price is $4 input, $20 output and $0.4 cache read, and the Relay365 output rate is about 10% of official. The pricing page carries the full table, region by region.

How the desk settles accounts

Relay365 bills from a prepaid wallet with no monthly fee, and every call is settled on the token counts the upstream reports. If a call fails upstream with a 502 or a 503, it is not billed. If a stream disconnects, you are billed for what was actually used. That distinction matters on a long night, because it means a broken run does not quietly become a full one.

Top-ups can be refunded within 14 days, with unused balance refunded minus what was consumed and the payment channel's fees. Bounty and promotional credit are not refundable. A new account starts with $0.25 in signup credit, and the referral bounty pays 18% at level 1 and 6% at level 2 of what referred members pay, withdrawable from $11.11 or movable into the wallet.

There are limits worth knowing before you script against the desk. An HTTP 429 is returned when one key makes more than 120 calls in 60 seconds, or when one account has more than 8 requests in flight. Plan for those ceilings and the log will stay clean.

Relay365 is an independent third-party relay, and everything above is on the same desk: one dashboard, six languages, one table of lanes, four separately priced token types and a usage log that shows every request in the latest 200. If you would like to see it with your own traffic on it, create a Relay365 account and start from the pricing page.