Grok 4.7 on the Relay365 API relay: one table, honest trade-offs

Putting Grok 4.7 on the same table as everything else

Most model try-outs come with paperwork attached: a new subscription, a new billing page, a new console to learn. Relay365 is run differently. It is an independent third-party relay, an AI API relay run like a round-the-clock duty desk, and every lane sits on one table. One prepaid wallet, one set of published token rates, one usage log. When you want to test Grok 4.7, you are not opening a second relationship with a second vendor — you are switching a request to a different lane on the same desk.

Grok 4.7 carries the wire name grok-4.7 and sits beside Grok 4.6, whose wire name is grok-4.6. That pairing is the honest starting point for a trial: both are available, both are priced per token, and you can send a small batch to one, look at the charge recorded for it, and decide from there. Nothing about the setup asks you to commit to a monthly plan before you know what the traffic will look like.

One detail worth reading carefully on the table is that Relay365 prices four token types separately: input, output, cache write and cache read. For Grok 4.7, the input rate is $0.6097 per 1 million tokens, the output rate is $1.83, and the cache read rate is $0.1528, all in approximate USD while the account itself is kept in CNY. The table shows no separate cache write rate for this lane, which is worth knowing before you plan a prompt-heavy workflow around it.

What Grok 4.7 costs when you do the arithmetic

Rates are easier to judge once they are turned into a day of traffic. For example, if you send 1,000,000 input tokens and 100,000 output tokens a day to Grok 4.7, that is about $0.6097 for the input plus $0.183 for the output, or roughly $0.7927 a day. Add a cache-heavy pattern and it stays legible: 5,000,000 cache read tokens on the same lane is about $0.764. The point of publishing input, output, cache write and cache read as separate lines is precisely so that a number like this can be assembled without guesswork.

The comparison lane is honest, too. Grok 4.6 is priced a little lower on every line that applies: input $0.5653, output $1.69 and cache read $0.1417 per 1 million tokens. So choosing 4.7 over 4.6 is a decision about which model you actually want, not a decision hidden in the pricing. If the two behave the same for your workload, the cheaper line is the sensible one; if 4.7 earns its place, the difference is small enough to see plainly in the usage log.

Other lanes on the same table make the trade-off sharper. GPT-6.1 Sol, wire name gpt-6.1-sol, is listed at input $0.1986, output $0.9944 and cache read $0.0389 per 1 million tokens, with its output rate about 10% of the official list output price of $10. For example, if you send 1,000,000 input tokens and 100,000 output tokens a day to that lane, you are looking at about $0.1986 plus $0.09944, or $0.29804 a day. Claude Opus 5.5, wire name claude-opus-5-5, sits at input $1.61, output $7.59, cache write $1.90 and cache read $0.0736, with its output rate about 38% of the official list output price of $20; the same 1,000,000 input plus 100,000 output example comes to about $1.61 plus $0.759, or $2.369.

The trade-offs: wallet access, pass rules and client rules

Here is the first honest trade-off. Duty passes are optional and apply to the Claude and GPT lanes only; every other model, including Grok 4.7 and Grok 4.6, always uses the wallet. A pass comes with a dedicated key that never touches the wallet. So if your plan is to try Grok 4.7, there is no pass to lean on — you top up the prepaid wallet and pay per token as you go, with no monthly fee attached.

Two passes show what that option looks like when it is available. The Claude pass covers 30 days with $6000 of quota, no weekly cap, for $506.53, drawing at the duty-pass rate x2.5, which is at most about $2400 of usage at list prices. The GPT pass covers 30 days with $7200 of quota and $2100 per week for $194.03, also at the duty-pass rate x2.5, at most about $2880 of usage at list prices. The trade-off is clear: a pass commits you to a quota block and draws at x2.5 the pass rate, while wallet usage is settled at the plain token rates and stays as flexible as your traffic.

Client rules are firm and worth reading before you wire anything up. The Grok lane works with the client families developers already use, including Cursor, Cherry Studio, Cline, OpenCode, SDKs and cURL, alongside the GPT, DeepSeek, Kimi, GLM, Qwen, MiniMax and Hunyuan lanes. Claude models work only from the official Claude Code and Claude Desktop clients, Claude Code 2.1.285 or newer. Money is straightforward too: a new account receives $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 moved into the wallet. Within 14 days of a top-up, unused balance can be refunded minus what was consumed and the payment channel's fees, while bounty and promotional credit are not refundable.

Reading the request log after the call

The duty desk keeps a usage log of the latest 200 requests, showing the time, model, input, output, cache recall, charge and result of each call. That is the instrument that makes a trial of Grok 4.7 meaningful: you are not guessing from a monthly total, you can see which request cost what and whether it succeeded. One honest note — the log does not show which key made the call, so if you run several keys at once, keep your own notes on naming.

Around the log, 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 once you log in, and six interface languages. Billing follows the same plain logic: every call is settled on the token counts the upstream reports, a call that fails upstream with a 502 or 503 is not billed, and a disconnected stream is billed for what was actually used. Limits are stated up front as well — HTTP 429 applies when there are more than 120 calls in 60 seconds on one key, or more than 8 requests in flight for one account.

Why the desk is still a good deal

Pull the threads together and the case for Relay365 is not that one model is cheaper than another, because that is not what the table claims. It is that every lane sits on one table, every request can be traced in the usage log, and the four token types are priced separately. You can move between Grok 4.7, Grok 4.6, GPT-6.1 Sol and Claude Opus 5.5 without opening a new subscription, and you can check the charge for each call the same afternoon you make it.

The trade-offs stay visible rather than being buried. Grok has no duty pass, so it runs on the wallet. Pass rates draw at x2.5, so a pass is a commitment rather than a discount. The refund window is 14 days from a top-up, minus consumption and channel fees. The log covers the latest 200 requests and does not name the key behind them. Claude models stay inside Claude Code and Claude Desktop. None of that is hidden, and none of it stops the desk from being a sensible place to try something new.

What you get for it is a quiet, round-the-clock arrangement: one wallet, published rates, a readable record of every request, and enough flexibility to change lanes the moment a different model makes more sense. If you would like to see what Grok 4.7 looks like on your own traffic, create a Relay365 account and put it on the table.