Chat

POST /v1/chat/completions follows the OpenAI Chat Completions API: messages, tools, tool_choice, response_format, temperature, max_tokens, stream and the other OpenAI parameters. Use any model whose type is text.

completion = client.chat.completions.create(
    model="anthropic/claude-fable-5.1",
    messages=[
        {"role": "system", "content": "You are a concise assistant."},
        {"role": "user", "content": "Explain webhooks in one sentence."},
    ],
)
print(completion.choices[0].message.content)

const completion = await client.chat.completions.create({
  model: 'anthropic/claude-fable-5.1',
  messages: [
    { role: 'system', content: 'You are a concise assistant.' },
    { role: 'user', content: 'Explain webhooks in one sentence.' },
  ],
})

The response is an OpenAI chat.completion with an ID starting with chatcmpl-. Its usage adds the cost:

"usage": {
  "prompt_tokens": 38,
  "completion_tokens": 19,
  "total_tokens": 57,
  "cost": 0.0000133,
  "currency": "USD",
  "cost_details": { "input": 0.0000038, "cached_input": 0, "output": 0.0000095 }
}

Streaming

Set stream: true to receive server-sent events. Each line is data: {chunk}, the whole stream shares one chatcmpl- ID, and it ends with data: [DONE]. The last chunk has an empty choices array and carries usage with the cost. You don’t need stream_options.include_usage.

stream = client.chat.completions.create(model="anthropic/claude-fable-5.1", messages=messages, stream=True)
for chunk in stream:
    if chunk.choices:
        print(chunk.choices[0].delta.content or "", end="")
    elif chunk.usage:
        print("\ncost:", chunk.usage.cost)

Validation, model selection and the balance check happen before the first byte, so those failures return a normal JSON error. An error after output has started arrives as data: {"error": {...}}.

If your client disconnects mid-stream, generation continues to the end and you are charged for the tokens actually produced, as with OpenAI.

Reliability

If a model’s upstream capacity fails before any output is produced, GEEKIDEA retries on other capacity automatically. You only see 503 model_overloaded when every option has failed; it is safe to retry later.

Get started

Generate

Workflows

Async and webhooks

Account

Streaming

Reliability

​Streaming

​Reliability

Streaming

Reliability