How to Choose an OpenAI API Gateway: 2026 Checklist
A 7-point checklist covering billing transparency, protocol compatibility, channel types and provider risk — verify any API gateway in 10 minutes.
Search "OpenAI API gateway" and you'll find hundreds of resellers claiming discounts from 30% to 90% off official prices. The cheap ones usually cost you in three currencies: opaque billing, channels that die without notice, and prepaid balances that evaporate. Here is a 7-point checklist that filters out 90% of bad gateways in ten minutes.
Who actually needs a gateway
- Developers without overseas cards — no way to open official accounts, so a gateway is the pragmatic path
- Multi-model users — you run GPT, Claude, Gemini and DeepSeek in parallel and don't want four keys and four invoices
- Tool users — Cursor, Cline, Cherry Studio and friends all take a custom base_url
Enterprise buyers with invoicing needs can use gateways too, but pay extra attention to points 6 and 7.
The checklist
1. Is the price list public — and does it match the bill?
A trustworthy gateway publishes every model's unit price and keeps console charges consistent with it. Open the pricing page, pick any model, multiply by your usage, and compare with the billing records. If they don't reconcile, walk away.
For example, zaimod exposes its pricing as a public endpoint (https://api.zaimod.com/api/pricing) with 175 priced models and per-call token records.
2. Is the billing unit stated clearly?
The classic trap: "gpt-4o-mini for just $0.1" — per 1K tokens or per 1M? Same price for input and output? A serious listing reads "USD per 1M tokens, input X / output Y". If that line is missing, ask support; if they can't answer, move on.
3. Protocol compatibility: are native endpoints exposed?
A gateway that only speaks OpenAI format will degrade Claude thinking blocks and tool calling. Check for:
- OpenAI
/v1/chat/completions - Claude-native
/v1/messages - Gemini-native
/v1beta/models/{model}:generateContent
All three means official SDKs connect without downgrade.
4. Channel type: official relay or reverse-engineered?
"Official relay" means reselling official API access — stable but pricey. "Reverse" means scraping the web app — dirt cheap, dies randomly. Group names usually tell you: look for "official", be suspicious of "unlimited". Official relays typically run 40–60% below list; anything far cheaper is a red flag.
5. Balance and expiry rules
Three questions: any top-up fees? Does the balance expire? Refund policy? The mainstream answer is "credits never expire" — that's zaimod's policy too. Avoid anything that zeroes your balance after 30 days unless you can forecast usage precisely.
6. Invoicing and bank transfer
For companies: invoices, bank transfer, volume top-up channels. Usually on the contact page or one question to support away.
7. Live support and status
The classic pre-runaway signal is silence. Before committing: is there live chat, how fast does it respond, where do you report incidents? Public status pages and community groups are a plus.
A real price snapshot
November 2026 snapshot (USD per 1M tokens, input/output, subject to live pricing):
| Model | Input | Output |
|---|---|---|
| gpt-4o-mini | $0.16 | $0.62 |
| gpt-4.1 | $2.00 | $4.00 |
| claude-sonnet-5 | $4.00 | $20.00 |
| gemini-3.1-flash-lite-preview | $0.25 | $1.50 |
| deepseek-v4-flash | $3.00 | $9.00 |
The final test before you commit
Top up the minimum (usually $1–5) and run three checks:
- Send a streaming request with the cheapest model
- Run a real coding task with your primary model (e.g. claude-sonnet-5)
- Reconcile token usage in the console the next day against your local counters
Pass all three and you have a keeper.
Bottom line
A gateway's value is not "cheap" — it's collapsing N platforms into one endpoint and one bill. Judge in this order: billing transparency > channel type > protocol compatibility > price. Absurdly cheap eventually becomes dead balance.
The zaimod price table is a public API — use it as your reconciliation sample →