Global routing
Requests are routed across leading providers with health-aware node selection, tuned for responsive calls under production load.
MULTI-PROVIDER
NEW
Kimi K3, GLM-5.2 and DeepSeek V4 Pro · now routable
One API for every
One API for every
model you build with.
Kimi, GLM, Qwen, Wan, DeepSeek and BytePlus, alongside GPT, Claude and Gemini — through a single OpenAI-compatible endpoint. Broader model choice, reliable routing, and usage-based billing with no minimum commitment.
OpenAI-compatible · change one line · pay only for what you use
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["ONEAI_API_KEY"],
base_url="https://api.1ai.cloud/v1", # the only line you change
)
response = client.chat.completions.create(
model="kimi-k3",
messages=[{"role": "user", "content": "Say hello in one sentence."}],
)
print(response.choices[0].message.content)import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.ONEAI_API_KEY,
baseURL: "https://api.1ai.cloud/v1", // the only line you change
});
const response = await client.chat.completions.create({
model: "claude-sonnet-5",
messages: [{ role: "user", content: "Say hello in one sentence." }],
});
console.log(response.choices[0].message.content);curl https://api.1ai.cloud/v1/chat/completions \
-H "Authorization: Bearer $ONEAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "glm-5.2",
"messages": [{"role": "user", "content": "Say hello in one sentence."}],
"stream": false
}'25+
Supported models
99.9%
Availability
<100ms
Avg routing latency
7×24
Support coverage
Why 1AI Cloud
Broader model choice,
Broader model choice,
one production-ready API.
Most gateways give you one region's models. 1AI Cloud gives you the Chinese frontier labs and the Western ones behind the same key, the same schema and the same invoice.
Usage-based billing
Pay for the tokens and seconds you actually consume. No seats, no minimum commitment, no contract before you can run your first benchmark.
PAY AS YOU GO
Unified model surface
Kimi, GLM, HappyHorse, Qwen, Wan, DeepSeek and BytePlus sit next to GPT, Claude and Gemini in one catalogue with one request schema.
ONE SCHEMA
Multi-region reliability
Multi-node scheduling is tuned continuously so API access stays responsive across the regions your users are actually in.
MULTI-NODE
OpenAI-compatible
Standard /v1/chat/completions semantics. Existing SDKs, agents and frameworks keep working — you change base_url.
DROP-IN
Named human support
Technical and commercial support from named contacts on WeChat and Telegram — not a ticket queue — for teams scaling past prototype.
7×24
Routing & reliability
A gateway earns its keep
A gateway earns its keep
on the bad days.
Any proxy works when every upstream is healthy. What matters is what happens when one provider rate-limits you mid-launch.
Declare a fallback order, not a single model
Send an ordered list of models. If the first upstream returns a rate limit, a timeout or a 5xx, the request moves down the list inside the same call — your application sees one response, not a retry loop you had to write yourself.
{
"model": "claude-sonnet-5",
"fallbacks": ["gpt-5.4", "glm-5.2", "kimi-k3"],
"messages": [ ... ]
}Streaming, tools and structured output
SSE streaming, function calling and JSON-schema structured output are normalised across providers, so an agent written against one model runs against the others without a rewrite.
SSE · TOOLS · JSON SCHEMA
One usage ledger
Every call — Chinese or Western model, text or video — lands in the same usage record, so cost attribution per key, per model and per project is one query rather than six invoices.
PER KEY · PER MODEL · PER PROJECT
Swap models without a deploy
Because the schema is shared, moving a workload from a frontier model to a cheaper one is a config change. Benchmark on the expensive model, serve on the one that pays for itself.
CONFIG, NOT CODE
Scoped keys per environment
Issue separate keys for staging, production and each teammate, and revoke any one of them without rotating the rest or touching the others' traffic.
REVOCABLE
Model catalogue
One key. Every frontier lab.
Filter by provider or modality. Prices shown are per 1M tokens unless the row states otherwise — media models bill per generated second or per image.
Highlights
What the China models bring.
Kimi K3
Up to 1M context with native vision and configurable reasoning — built for coding, knowledge work and long-running agent tasks.
GLM-5.2
1M context and up to 128K output, plus function calling and structured output for long-running development workflows.
Qwen3.8 Max Preview
1M context with reasoning, function calling and built-in tools. This model remains in preview — treat availability as subject to change.
DeepSeek V4 Pro
1M context, up to 384K output, tool calls and thinking modes for agentic and heavy reasoning workloads.
Wan & HappyHorse
Text-to-video generation billed per generated second, at 720P and 1080P, for creative media pipelines alongside your text models.
Seedream 5.0 Pro
Image generation with the first image free, then per-image billing that steps up only above 2.36M pixels.
Quick start
Four steps to your first call.
If you already call an OpenAI-compatible endpoint, step four is a one-line diff.
STEP 01
Create account
Register and verify your account in the console.
STEP 02
Top up
Alipay or WeChat Pay. Balance is usage-based — no plan to pick.
STEP 03
Create a key
Issue a scoped API key in one click. Revoke it just as fast.
STEP 04
Start calling
Point base_url at 1AI Cloud and pick a model.
# 1 — install any OpenAI-compatible SDK
pip install openai
# 2 — set your key
export ONEAI_API_KEY="your-api-key"
# 3 — verify the key and list what you can route to
curl https://api.1ai.cloud/v1/models \
-H "Authorization: Bearer $ONEAI_API_KEY"
# 4 — send a request. swap `model` for any id in the catalogue.
curl https://api.1ai.cloud/v1/chat/completions \
-H "Authorization: Bearer $ONEAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"kimi-k3","messages":[{"role":"user","content":"ping"}]}'
Pricing
Provider reference prices,
Provider reference prices,
published in full.
Text models are priced per 1M tokens in the currency shown. Media models bill per generated second or per image. Every row states its own billing unit.
These are provider reference prices, last checked 22 Jul 2026, published so you can
verify them against the source before you commit. Promotions, regions and model availability
change — the linked official pages are always authoritative.
Trust
Reasons to route production
Reasons to route production
traffic through us.
A gateway sits between your application and every model you depend on. That position has to be earned, so here is exactly what it involves.
Encrypted in transit
All API traffic runs over TLS. Keys are presented as bearer tokens and are never embedded in URLs or query strings.
Your prompts are yours
Prompts and completions are passed through to the upstream provider to serve your request. They are not used to train models.
Verifiable pricing
Every price on this page links to the provider's own pricing page, with the date we last checked it. Nothing is quoted without a source.
No lock-in
The API is OpenAI-compatible by design. If you ever want to leave, you change base_url back — there is nothing proprietary to unwind.
Revocable, scoped keys
Issue one key per environment or per teammate. Revoking a compromised key does not interrupt any other traffic.
Named humans
Three named sales and technical contacts, reachable on WeChat and Telegram, listed publicly on this page. No anonymous support queue.
All routing regions operational
·
99.9% availability target
·
7×24 support coverage
Built for
Where teams put it to work.
Coding agents
Long-context models with tool calling, and a cheaper fallback for the routine steps in a multi-step run.
RAG & knowledge work
1M-context models let you skip aggressive chunking, and swap to a cheaper model for re-ranking passes.
Creative media
Image and video generation billed per image or per generated second, on the same key as your text models.
Model evaluation
Benchmark ten models against one prompt set without opening ten accounts or reconciling ten invoices.
FAQ
Questions worth asking
Questions worth asking
before you route traffic.
Do I have to rewrite my application?+
No. The endpoints follow the OpenAI schema, so any OpenAI SDK, agent framework or
LangChain-style integration works once you point base_url at 1AI Cloud and
supply a 1AI Cloud key. In most codebases that is a one-line change.
How is billing calculated?+
Usage-based. Text models bill per 1M input and output tokens in the currency shown
on the pricing table; image models bill per image; video models bill per generated second. There is
no seat licence and no minimum commitment — you top up a balance and draw against it.
Are the prices on this page what I actually pay?+
The tables show provider reference prices, last checked 22 July 2026, with a link to
each provider's official pricing page so you can verify them. Providers change prices and run
promotions; confirm your effective rate in the console before you plan a budget around it.
What happens if a provider goes down mid-request?+
You can declare an ordered fallback list on the request. If the first upstream
returns a rate limit, a timeout or a 5xx, the call moves to the next model in the list and your
application receives a single response rather than an error you have to retry around.
Which payment methods are supported?+
Alipay and WeChat Pay. If you need invoicing or another settlement method for a
team account, talk to sales — the contacts are listed at the bottom of this page.
Are my prompts used for training?+
No. Prompts and completions are passed to the upstream provider to serve your
request. Note that each upstream provider also applies its own data policy — if you have strict
data-handling requirements, raise them with sales so the right models are recommended.
Can I use it from outside mainland China?+
Yes. Routing is multi-region, and the Chinese frontier models are the main reason
teams outside the region use the gateway — they are otherwise awkward to reach and to pay for.
Ready to start?
Create a key, top up whatever you feel like risking, and send your first request in the next five minutes.