AIWave API Documentation

The AIWave API gives you one OpenAI-compatible endpoint for DeepSeek, GLM, Kimi, Qwen and ERNIE. If your code already talks to OpenAI, you change two lines — the base_url and the API key — and every Chinese AI model below becomes available. Zero Data Retention, no separate accounts per provider, one balance in USD.

Base URL and authentication

Every endpoint lives under a single base URL. Authentication uses a bearer token, exactly as the OpenAI API does, so existing SDKs and HTTP clients work without modification.

Base URL:  https://aiwave.live/v1
Header:    Authorization: Bearer sk-YOUR_KEY

Create a key in the console. Keys are scoped to your account balance; there is no per-provider signup.

Quick start: migrate from OpenAI in two lines

The official openai-python SDK works unchanged. Point it at AIWave and pick any model ID from the table below.

from openai import OpenAI

client = OpenAI(
    api_key="sk-YOUR_KEY",
    base_url="https://aiwave.live/v1",   # ← only change
)

response = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[{"role": "user", "content": "Explain MoE routing in two sentences."}],
)
print(response.choices[0].message.content)

Node.js is the same idea with the openai-node SDK:

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.AIWAVE_API_KEY,
  baseURL: "https://aiwave.live/v1",
});

const res = await client.chat.completions.create({
  model: "glm-5",
  messages: [{ role: "user", content: "Hello!" }],
});

API reference

Endpoint	Method	Reference
/v1/chat/completions	POST	Chat Completions API →
/v1/embeddings	POST	Embeddings API →
/v1/models	GET	List Models API →

Guides

• Authentication — API keys, bearer tokens, SDK setup
• Streaming responses — server-sent events in Python and TypeScript
• Error codes — 400, 401, 429, 503 and which are safe to retry
• Rate limits — backoff, jitter and concurrency control

Available models

These are the model IDs currently served by the gateway, grouped by provider. Pass the ID verbatim as the model field. For per-model context windows, capabilities and price, browse the model directory; live rates are on the pricing page.

Provider	Models	Model IDs
DeepSeek	10	deepseek-chat deepseek-r1 deepseek-r1-distill-qwen-14b deepseek-r1-distill-qwen-32b deepseek-reasoner deepseek-v3 deepseek-v3.2 deepseek-v3.2-think deepseek-v4-flash deepseek-v4-pro
GLM (Zhipu AI)	9	glm-4.5 glm-4.5-air glm-4.6 glm-4.7 glm-4.7-flash glm-4.7-flashx glm-5 glm-5-turbo glm-5.1
Kimi (Moonshot AI)	6	kimi-k2.5 kimi-k3 moonshot-v1-32k moonshot-v1-32k-vision-preview moonshot-v1-8k moonshot-v1-8k-vision-preview
Qwen (Alibaba)	8	qwen3-235b-a22b-instruct-2507 qwen3-235b-a22b-thinking-2507 qwen3-32b qwen3-8b qwen3-coder-480b-a35b-instruct qwen3.5-122b-a10b qwen3.5-27b qwen3.5-397b-a17b
ERNIE (Baidu)	27	ERNIE 4.5 Turbo ERNIE 5.0 ernie-3.5-128k ernie-3.5-8k ernie-3.5-8k-preview ernie-4.0-8k ernie-4.0-8k-latest ernie-4.0-8k-preview ernie-4.0-turbo-128k ernie-4.0-turbo-8k ernie-4.0-turbo-8k-latest ernie-4.0-turbo-8k-preview ernie-4.5-8k ernie-4.5-turbo ernie-5.0 ernie-5.0-thinking-preview ernie-5.1 ernie-char-8k ernie-char-fiction-8k ernie-lite-8k ernie-lite-pro-128k ernie-novel-8k ernie-speed-128k ernie-speed-8k ernie-speed-pro-128k ernie-tiny-8k ernie-x1.1
MiniMax	1	minimax-m2.5

Model availability changes as providers ship new versions. Call GET /v1/models at runtime rather than hard-coding a list; see the changelog for additions and deprecations.

Streaming responses

Set stream: true to receive server-sent events. The chunk format matches the OpenAI specification, so streaming parsers written for OpenAI work unchanged.

stream = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Write a haiku about latency."}],
    stream=True,
)
for chunk in stream:
    delta = chunk.choices[0].delta.content
    if delta:
        print(delta, end="", flush=True)

The stream terminates with a data: [DONE] sentinel. Full parameter reference is in the Chat Completions documentation.

Error codes

Errors use standard HTTP status codes with a JSON body describing the failure.

Status	Meaning	What to do
400	Bad Request	Malformed JSON, missing messages, or an unknown parameter.
401	Unauthorized	Missing or invalid Authorization: Bearer header.
403	Forbidden	The key is valid but not permitted to use the requested model.
404	Not Found	Unknown model ID. Call GET /v1/models for the current list.
429	Too Many Requests	Rate limit or quota exhausted. Back off exponentially and retry.
500	Internal Server Error	Unexpected gateway error. Safe to retry with backoff.
503	Service Unavailable	Upstream model provider is overloaded. Retry, or fall back to another model.

{
  "error": {
    "message": "Model not found",
    "type": "invalid_request_error",
    "code": "model_not_found"
  }
}

For 429 and 503, retry with exponential backoff and consider falling back to a second model — the same request body works across every provider, which makes fallback a one-line change.

Compatibility notes

• Request and response shapes follow the OpenAI Chat Completions specification.
• Tool/function calling and JSON mode are supported by models that implement them upstream — capability varies by provider, so check the model page before relying on it.
• Token accounting is returned in the usage object on non-streaming responses.
• Framework integrations that accept a custom base_url — LangChain, LlamaIndex, the Vercel AI SDK, Cursor, Cline and similar tools — work without a dedicated provider plugin.

Where to go next

• How to access Chinese AI models for production
• Migrating an OpenAI app to Chinese models
• AI API cost comparison 2026
• Full model directory · Pricing · Provider comparisons