Chinese models in our catalog
68
Aggregator comparison
ChinaAPI and OpenRouter, compared on facts you can check.
Both services put many models behind one OpenAI-compatible key. They differ in what the catalog is for, how the rate is set, what it costs to put money in, and which endpoints the base URL serves. Those are the rows below. Which model is better is not a row here, because we have not measured it.
Positioning
One is a catalog of everything; one is a catalog of Chinese models.
OpenRouter's value is breadth: one account reaching many providers, including the Western frontier labs. ChinaAPI's catalog is narrower on purpose. Every model in it comes from a Chinese vendor, and for 6 of those vendors we publish a hub for one specific reason: their own developer console expects a mainland-China entity or phone number, so a developer outside that boundary often cannot finish sign-up at all.
So the honest framing is not which service wins. It is whether your shortlist is mostly Chinese models. If it is not, breadth matters more than we do. If it is, the rows below are the ones that will actually decide it.
Chinese vendors covered
11
OpenAI-compatible endpoint families
5
Base URL
https://api.chinaapi.ai/v1
Side by side
7 rows, every one of them sourced.
ChinaAPI cells state our own published position. OpenRouter cells paraphrase OpenRouter's own documentation on the retrieval date given under sources; their page is authoritative.
| Dimension | ChinaAPI | OpenRouter |
|---|---|---|
| Chinese models covered | 68 models from 11 Chinese vendors, generated from our own catalog on August 10, 2026. | Not stated here. We could not source a dated, re-checkable count, so we publish none — see the known gaps below. |
| How the rate is set | Token-model rates are synchronized from the gateway and may follow providers' official China list prices where configured. Media models show a transparent USD service rate that may include a service margin covering provider input/output billing, payment processing, chargeback exposure and operations; any margin is included in the displayed rate and is never added separately. | OpenRouter's documentation says the price you pay for inference is the underlying provider's own rate, passed through unchanged. |
| Fee on adding credit | Credit is prepaid and drawn down by metered usage at the published ChinaAPI list rates. The top-up page publishes no separate purchase fee on top of that rate. | OpenRouter's documentation lists a 5.5% fee with a $0.80 minimum on card credit purchases, and a 5% fee on cryptocurrency purchases. |
| Payment methods | Online card payments are currently unavailable. The current route is a prepaid redemption code: you tell us the amount and your dashboard account email, then redeem the code in the dashboard wallet. | OpenRouter's documentation says it accepts major credit cards, AliPay and cryptocurrency payments in USDC. |
| Key management | One key and one base URL cover every vendor in the catalog, so adding a model is a configuration change rather than a new account, a new console and a new invoice. | OpenRouter authenticates with an API key sent as a bearer token, and documents a separate class of management keys for creating and administering keys programmatically. |
| OpenAI-compatible surface | One base URL serves /v1/audio/speech, /v1/audio/transcriptions, /v1/chat/completions, /v1/images/generations, /v1/video/generations, because the catalog spans text, image, video and audio models. | OpenRouter's documentation describes itself as a drop-in replacement for OpenAI, and states that it implements the OpenAI API specification for the completions and chat completions endpoints. |
| Where you can sign up | Sign-up needs no mainland-China company entity and no mainland-China phone number. A new dashboard account starts with $2 of trial credit. | Not stated here. We found no published eligibility list to cite — see the known gaps below. |
Compatibility
What one base URL actually serves.
This list is generated from the catalog, so it grows when the catalog does. Text, image, video and audio models all answer on the same key.
POST https://api.chinaapi.ai/v1/audio/speechPOST https://api.chinaapi.ai/v1/audio/transcriptionsPOST https://api.chinaapi.ai/v1/chat/completionsPOST https://api.chinaapi.ai/v1/images/generationsPOST https://api.chinaapi.ai/v1/video/generations
If your code already targets an OpenAI-compatible client, switching is a base URL and a key. Nothing else in the request changes.
The two base URLs, for reference:
https://api.chinaapi.ai/v1https://openrouter.ai/api/v1
curl -X POST https://api.chinaapi.ai/v1/chat/completions \
-H "Authorization: Bearer $CHINAAPI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-v4-pro",
"messages": [{"role": "user", "content": "Hello from ChinaAPI"}]
}'Scope
Why there is no quality column here.
Every row on this page is a number you can check against a published source: a price, a context window, or a statement about who may open an account. There is no row for reasoning, coding or instruction-following, because we have not measured them. Our bench harness is not published yet, and until it is, a ranking from us would be a preference presented as a measurement.
That is a real limitation, not a hedge, so here is the position it leaves us with: use this page to rule options out on cost, context and account eligibility, and use your own evaluation set to choose among whatever survives. If a comparison page tells you which model is smarter and shows you no task, no sample count and no failure definition, it is selling something.
Known gaps
What we could not verify, and therefore did not write.
A missing row is a deliberate choice. Each gap below is something a comparison page is expected to state and that we could not source from a page we can re-check, so we left it out rather than fill it in from memory.
OpenRouter
How many Chinese models OpenRouter lists. Their catalog is a live listing that changes without a dated, citable page, so any number here would be a snapshot we could not honestly date or re-check. The only model count on this page that we stand behind is our own, and it is generated from our catalog.
OpenRouter
Which countries and territories can open an OpenRouter account. We did not find a published eligibility list equivalent to the ones OpenAI and Anthropic maintain, so we make no claim either way.
Related comparisons
Same rule everywhere: checkable facts only, no quality ranking.
Sources and updates
ChinaAPI model names, context values and USD prices on this page are generated from the ChinaAPI model catalog on August 10, 2026. Live pricing in the dashboard is authoritative.
Third-party figures were read from each vendor's own documentation on 2026-08-10. That is a dated snapshot, not a live feed: the vendor's own page stays authoritative, and rates change without notifying us. Check the source before you commit a budget to a number here.