Streaming

Server-sent events in both formats — the event shapes differ, and mixing them up reads as silence.

Set stream to true and the response becomes text/event-stream. Both endpoints stream, but they emit different event shapes.

Anthropic format

Copycurl -N https://aiprimetech.io/v1/messages \
  -H "x-api-key: $CLAUDEAPIKEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-sonnet-4-6","max_tokens":512,"stream":true,
       "messages":[{"role":"user","content":"Count to five"}]}'

Copyevent: message_start
data: {"type":"message_start","message":{"id":"msg_...","usage":{"input_tokens":9}}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"One"}}

event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"end_turn"},"usage":{"output_tokens":14}}

event: message_stop
data: {"type":"message_stop"}

Text arrives in content_block_delta events at delta.text. Final output token counts arrive in message_delta, not message_start.

OpenAI format

Copydata: {"choices":[{"delta":{"content":"One"},"index":0}]}

data: {"choices":[{"delta":{},"finish_reason":"stop","index":0}]}

data: [DONE]

Text arrives at choices[0].delta.content, and the stream ends with a literal data: [DONE] sentinel that is not JSON.

Python, Anthropic SDK

Copyfrom anthropic import Anthropic

client = Anthropic(api_key=class="s">"sk-your-key", base_url=class="s">"https://aiprimetech.io")

with client.messages.stream(
    model=class="s">"claude-sonnet-4-6",
    max_tokens=512,
    messages=[{class="s">"role": class="s">"user", class="s">"content": class="s">"Count to five"}],
) as stream:
    for text in stream.text_stream:
        print(text, end=class="s">"", flush=True)

Operational notes

• Use curl -N when testing by hand, or curl buffers the whole stream and it looks like nothing is happening.
• Disable proxy buffering in front of your own app (proxy_buffering off in nginx) or clients see the reply arrive in one lump at the end.
• A dropped connection mid-stream still bills the tokens already generated.
• Streaming does not change the price — only when the bytes arrive.