Build agents on the OpenAI-compatible API

Recommended models available now

Chosen for agents: reliable tool calls, structured output, context that holds many turns. Multi-step loops cost grows linearly with steps, so get the flow working on a cheap model, then upgrade to see the difference.

Platform prices per 1M tokens, billed on actual usage. The list follows the live pricing API; retired models never show up here. Pick a model and the config and cost below update to match.

Exact configuration

The API base and the selected model qwen3.8-max are already filled in. Replace the key with your own.

from openai import OpenAI

client = OpenAI(base_url="https://token.easyapi.com/v1", api_key="sk-your-key")

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get current weather for a city",
        "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]},
    },
}]

resp = client.chat.completions.create(
    model="qwen3.8-max",
    messages=[{"role": "user", "content": "What is the weather in Shanghai today?"}],
    tools=tools,
    tool_choice="auto",
)
call = resp.choices[0].message.tool_calls[0]
print(call.function.name, call.function.arguments)

Full OpenAI SDK · Python integration guide →

Get one request working first

Swap in your key and run it as is. A 200 means the base URL, key and model are all right; otherwise match the error against the list below.

curl https://token.easyapi.com/v1/chat/completions \
  -H "Authorization: Bearer sk-your-key" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.8-max","messages":[{"role":"user","content":"Hello"}]}'

What to know

• tools, tool_choice, stream and response_format are passed to the upstream as is. Support varies by model; the parameter table on each model's detail page says what takes effect.
• Multi-turn agents accumulate messages. The context window is a hard limit: a request over it fails instead of being truncated.
• Watch for 429 under high concurrency on one key. Both platform rate limiting and an exhausted balance return 429; the error message says which.
• The models fallback list and provider.sort let a request declare backup models and routing preference, switching automatically on upstream failure. See the API docs.

Common errors

tool_calls is empty and the model answered in prose

Usually the model does not support tool calls or tool_choice is unset. Pick a model from the recommended list and set tool_choice to auto or a specific function.

No deltas arrive when streaming

Confirm stream is true and the client is not buffering. curl with -N shows the raw SSE and pinpoints the issue fast.

Returned JSON fails to parse

Use response_format with json_object or json_schema and say 'output JSON only' in the system prompt. Models without response_format rely on prompting alone and are not guaranteed.

429 Too Many Requests

Read the message: insufficient quota means top up; rate limit means too much concurrency, so slow down or spread across keys.

Cost example

At the current price of qwen3.8-max, one typical request with 6,000 input + 800 output tokens, 2,000 times a day:

You pay only for tokens actually used. No platform fee, routing fee or monthly minimum. Real usage is listed request by request in the console logs.