Skip to content

How to choose a Claude Code relay: checks you can run yourself

· 6 min read · OpusVIP

Short answer

Convert every price into what you pay per $1 of usage at list price. Then check that the model multiplier is 1x, cache is billed at Anthropic's cache rates, failed requests are free, and every log line can be recomputed. Finish with a small real task. On OpusVIP, $0.15 buys $1 of usage at list price.

When you pick a relay, look past the price to how billing works and whether you can verify it. Each of the seven checks below is something you can do yourself, and OpusVIP has a clear answer to every one. New to relays? Start with What is a Claude Code relay?

1. Convert every price to "what does $1 of usage at list price cost me?"

Sites describe top-ups and multipliers in different ways: some advertise a discounted top-up and then add a 2x multiplier, others quote a flat per-token price. To compare them, convert everything to "what do I pay for $1 of usage at list price?" OpusVIP bills in credits: 1 credit = $0.01 of usage at list price, and 100 credits cost $0.15, so $0.15 covers $1 of usage: 15% of list price. The model multiplier is 1x, so credits are deducted at list price: Opus 5.5 input is 400 credits per 1M tokens ($0.60 to you), Fable 5.1 input 1,000, Sonnet 5.5 input 200, and Haiku 5.5 input 10.

2. Is the multiplier really 1x?

Check the model multiplier on the pricing page and confirm that usage is billed at list price. Every OpusVIP model is 1x, the same for every user group.

3. Is cache billed at Anthropic's cache rates?

Claude Code resends a large amount of context on every turn, and cache reads cost only a fraction of regular input (Anthropic's cache-read pricing varies by model, from about 1/10 to 1/40 of the input price). OpusVIP prices both cache reads and writes at Anthropic's cache rates and lists them separately in the log.

4. Can you audit every request in the log?

A good log shows at least the time, model, input / output / cache read and write token counts, and the credits deducted. Take any entry and recompute it with the formula; see Billing and caching explained.

5. Are failed requests charged?

OpusVIP does not deduct credits for upstream errors or requests that don't complete. You can confirm this against your log and balance.

6. Which clients are supported?

OpusVIP is built for the official Claude Code CLI, so the request shape is well defined and behavior is predictable. Requests from other clients aren't accepted. If you also need other clients, check that a relay supports them before you choose it.

7. Run a full test with a small balance

  1. Top up a small amount first.
  2. Set it up following the docs, run claude -p "hello", then start a session and type /model to confirm the model.
  3. Do a small real task, then check the model, token counts and credits deducted in the log.
  4. Run the same task twice and see whether the second run shows cache reads.

Once everything checks out, top up more. More ways to verify are in How to verify the model and usage of every request.


Questions? Email hello@opusvip.dev.