The nine checks
| # | Check | What to look for |
|---|---|---|
| 1 | Endpoint compatibility | OpenAI-compatible and Anthropic-compatible endpoints, so existing clients need no code change |
| 2 | Model coverage | Several families, not one vendor — Claude, GPT, DeepSeek, Qwen, Kimi, GLM |
| 3 | Live model list | A queryable per-key list (e.g. GET /v1/models) rather than a static page that drifts |
| 4 | Pricing basis | A published share of official rates per family, not an opaque blended number |
| 5 | Top-up rails | Payment networks you can actually use; card, or crypto such as TRC-20 / BSC / Solana |
| 6 | Credit expiry | Credits that never expire — expiry is a hidden price increase |
| 7 | Limits and quotas | Stated gateway limits, and clarity that upstream provider limits still apply |
| 8 | Observability | Per-key usage and spend in a console, plus an error format you can parse |
| 9 | Support and scope | A reachable support channel and an honest statement of what the service is not |
1–3: will your code keep working?
Compatibility is the check that costs the most to get wrong. If the gateway speaks the OpenAI and Anthropic schemas, your existing SDK, streaming handler and tool-calling code keep working — you change a base URL, not a codebase. Coverage matters because the reason to use a gateway is to stop opening one account per vendor. And a live model list matters because static catalogues silently rot: the authoritative answer should come from an API call, not a marketing page.
4–6: what will it actually cost?
Prefer gateways that publish a share of each provider’s official rate, per family. A single blended discount figure hides the fact that the discount differs by model. Then check the top-up rail: a gateway that only accepts a payment method you do not have is not cheaper at any price. Finally, check expiry. Credits that expire turn an unused month into a donation; QuantoPort credit packs never expire and the minimum top-up is $10.
7–9: what happens when something breaks?
Ask what the gateway limits are, and confirm that upstream provider limits still apply on top. Make sure usage and spend are visible per key, and that errors come back in a parseable, documented format. Then check support: a channel you can reach, in a language you can use — QuantoPort support is on Telegram at @thunderpeak_support in English and Chinese.
Red flags
- A pricing page that never states the basis of its discount.
- No way to query the live model list with your own key.
- No public statement of what the service is not, or implied affiliation with the model vendors.
- Credits with an expiry date that is not stated up front.
- No visible per-key usage reporting.
Frequently asked
How many models should a gateway cover? Enough that you can switch models without a new vendor relationship — in practice several families, not one.
Is a cheaper share always better? No. Compare the share per family, the payment rail, credit expiry and the support channel together.
How do I verify coverage before paying? Create a key, call GET /v1/models, and send a small request to each family you plan to use.
Run the checklist on QuantoPort
One key, one balance, published per-family rates, credits that never expire.