Cursor + Relay365: an API relay setup for GPT-6 Astra

Wiring GPT-6 Astra into Cursor as an OpenAI-compatible lane

Cursor talks to GPT-6 Astra through Relay365 by way of an OpenAI-compatible provider entry. Point the base URL at the Relay365 API endpoint, hand the desk a key bound to the GPT lane, and send the model name gpt-6-astra. That is the short version. The more useful version is what the desk does around the request: Relay365 is an independent third-party relay run like a round-the-clock duty desk, every lane sits on one table, every request can be traced in the usage log, and the four token types — input, output, cache write and cache read — are priced separately instead of being folded into a single number.

The setup is deliberately narrow, and the desk would rather say so up front. Cursor, Cherry Studio, Cline, OpenCode, any SDK and cURL pair only with GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax and Hunyuan lanes. Gemini 2.5 Flash runs on its native protocol only, through Gemini CLI, using :generateContent for single replies and :streamGenerateContent?alt=sse for streams; it cannot go through /v1/chat/completions, so it is not an OpenAI-compatible option for Cursor.

One more boundary is worth stating plainly. Claude Fable 5.1 runs only from the official Claude Code and Claude Desktop clients, with Claude Code 2.1.285 or newer. Relay365 keeps that rule firm, which is exactly why the Cursor conversation here stays on the GPT side of the table.

What the desk charges for GPT-6 Astra, token by token

Rates on Relay365 are per 1 million tokens in approximate USD, while the account itself is kept in CNY. For GPT-6 Astra (wire name gpt-6-astra), input is $0.7444, output $3.72 and cache read $0.075. The official list price for the same model is input $10, output $50 and cache read $1, so the Relay365 output rate is about 7% of official.

For example, if you send 100,000 input tokens and 200,000 output tokens a day on GPT-6 Astra, the desk works out roughly $0.07444 of input and $0.744 of output, or about $0.81844 a day and about $24.55 over 30 days, using those rates and no other assumption. The same shape of day on GPT-5.6 Sol (gpt-5.6-sol) comes to about $0.45536, or about $13.66 over 30 days, at input $0.3736 and output $2.09. These are examples, not quotes.

Billing runs on a prepaid 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. A disconnected stream is billed for what was actually used. That last detail matters more than a headline rate when a Cursor session runs long and the connection wobbles halfway through.

The trade-offs the duty desk will not hide

Here is the honest list. First, Relay365 is an independent third-party relay, not a first-party channel from any model maker, so the desk is one more hop between Cursor and the upstream. Second, the pricing table is a table, not a promise: the account is kept in CNY while the rates are quoted in approximate USD, so your bill carries a conversion step. Third, the usage log shows the latest 200 requests with the time, model, input, output, cache recall, charge and result of each call — but it does not show which key made the call, so if you run several keys, attribute traffic by process rather than by reading the log.

Fourth, rate limits are real: HTTP 429 appears after more than 120 calls in 60 seconds on one key, or more than 8 requests in flight for one account. A busy Cursor agent can reach the second one sooner than the first. Fifth, refunds cover unused balance within 14 days of a top-up, minus what was consumed and the payment channel's fees; bounty and promotional credit are not refundable. Sixth, Gemini stays on its own protocol and Claude has its own client rule, so the one-table promise at Relay365 is about lanes, pricing and records — not about making every model speak the same protocol.

Against those, the desk offers a few things that are hard to fake. The dashboard holds API keys with tags, a key bound to a lane, quick-wiring snippets, a ledger of wallet, bounty and pass accounts, in-site messages to the desk (log in first) and six interface languages. A new account starts with $0.25 in signup credit. 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.

Duty passes for steady GPT traffic

For predictable GPT traffic, Relay365 also offers duty passes — 值守卡 in the desk's own words — for Claude and GPT only. Other models always use the wallet. A pass comes with a dedicated key that never touches the wallet, and its quota is drawn at the duty-pass rate x2.5.

The GPT 30 days · $3000 · $900 per week pass costs $81.53 for 30 days and gives up to $3000 of quota, which is at most about $1200 of usage at list prices, and less on models whose pass rate sits above list. The larger GPT 30 days · $7200 · $2100 per week pass costs $194.03 for 30 days, with up to $7200 of quota and at most about $2880 of usage at list prices, and less on models whose pass rate sits above list.

The trade-off is the x2.5 rate and the weekly ceiling. A pass is a prepaid commitment to a lane's traffic, not a discount on everything. If your Cursor usage is spiky and light, the wallet at $0.7444 input and $3.72 output per 1 million tokens may serve you better. If the volume is regular, the pass turns a variable bill into a known one and keeps its key away from your wallet balance.

Why the desk is still a good deal

Put the trade-offs beside the numbers and Relay365 still reads as a good deal. GPT-6 Astra output on Relay365 is about 7% of official list price, with input $0.7444 against $10 and cache read $0.075 against $1. That gap is wide enough to absorb the inconvenience of one more hop, a conversion step and a log that records the request but not the key.

The desk also keeps the operational picture in one place. Every lane on one table; every request traceable in the usage log; input, output, cache write and cache read priced separately so you can see where the tokens actually went. Failed upstream calls are not billed, disconnected streams are billed only for what was used, and there is no monthly fee to keep paying while a project sleeps.

If you want to try it, create a Relay365 account, take the $0.25 signup credit, bind a key to the GPT lane, and point Cursor at the relay's OpenAI-compatible base URL. Then watch the usage log on your first day and decide for yourself whether the desk earns its keep.