Streaming

With "stream": true the answer arrives as server-sent events while the model writes it.

Request a stream

cURL

curl -N https://api.getworg.com/v1/chat/completions \
  -H "Authorization: Bearer $WORG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4.1-flash:cached",
    "messages": [{"role": "user", "content": "Count from 1 to 3."}],
    "max_tokens": 512,
    "stream": true,
    "stream_options": {"include_usage": true}
  }'

Python

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.getworg.com/v1",
    api_key=os.environ["WORG_API_KEY"],
)

stream = client.chat.completions.create(
    model="deepseek-v4.1-flash:cached",
    messages=[{"role": "user", "content": "Count from 1 to 3."}],
    max_tokens=512,
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)
print()

Format

Stream (ids shortened)

: keepalive

data: {"id":"chatcmpl-req_Kzzo...","object":"chat.completion.chunk","created":1793729649,"model":"deepseek-v4.1-flash:cached","choices":[{"index":0,"delta":{"role":"assistant","content":""},"logprobs":null,"finish_reason":null}]}

data: {"id":"chatcmpl-req_Kzzo...","object":"chat.completion.chunk","created":1793729649,"model":"deepseek-v4.1-flash:cached","choices":[{"index":0,"delta":{"content":"1"},"logprobs":null,"finish_reason":null}]}

data: {"id":"chatcmpl-req_Kzzo...","object":"chat.completion.chunk","created":1793729649,"model":"deepseek-v4.1-flash:cached","choices":[{"index":0,"delta":{"content":","},"logprobs":null,"finish_reason":null}]}

data: {"id":"chatcmpl-req_Kzzo...","object":"chat.completion.chunk","created":1793729649,"model":"deepseek-v4.1-flash:cached","choices":[{"index":0,"delta":{"content":" 2, 3."},"logprobs":null,"finish_reason":null}]}

data: {"id":"chatcmpl-req_Kzzo...","object":"chat.completion.chunk","created":1793729649,"model":"deepseek-v4.1-flash:cached","choices":[{"index":0,"delta":{},"logprobs":null,"finish_reason":"stop"}]}

data: {"id":"chatcmpl-req_Kzzo...","object":"chat.completion.chunk","created":1793729649,"model":"deepseek-v4.1-flash:cached","choices":[],"usage":{"prompt_tokens":12,"completion_tokens":9,"total_tokens":21,"prompt_tokens_details":{"cached_tokens":0}}}

data: [DONE]

Each event is one data: line with a JSON chunk, then a blank line. The stream ends with data: [DONE].
Text arrives in delta.content, reasoning in delta.reasoning_content. The chunk with finish_reason has an empty delta.
Lines that start with a colon, such as : keepalive, are comments. SSE clients and the OpenAI SDKs skip them.
Some chunks carry "usage": null. Read usage only from the usage chunk below.

The usage chunk

With "stream_options": {"include_usage": true}, the last chunk before [DONE] has an empty choices array and the usage of the whole request. Without it there is no usage chunk. Worg bills the same either way.

Tool calls in a stream

A function call arrives in pieces in delta.tool_calls. The first piece for each call has its id and function name; later pieces add to function.arguments. Join the pieces by index, and parse the arguments after the chunk with finish_reason: "tool_calls". Some text in delta.content can come before the call.

Errors

Until the first chunk, nothing is sent: an error before then is a normal HTTP error with a JSON body, as on Errors and retries. A combo can still move to its next lane.
After the first chunk the status is already 200. If the model stops with an error, a chunk carries an error object with code upstream_error, and the request stays on that lane.
A stream that ends without data: [DONE] was cut. Retry the whole request.

When your client disconnects

The model keeps writing after your client leaves. Worg keeps reading the stream for up to 2 minutes after you disconnect and bills the tokens the model wrote. In one test a client that left after 2 seconds was billed for all 458 output tokens.

Set max_tokens to the most you want to pay for. Closing the connection does not stop the cost.

Next: Tool calling Reasoning Errors and retries