Streaming Text Completions

🚧

Legacy API

The Text Completions API is a legacy API. Future models and features will require use of the Messages API, and we recommend migrating as soon as possible.

When creating a Text Completion, you can set "stream": true to incrementally stream the response using server-sent events (SSE). If you are using our client libraries, parsing these events will be handled for you automatically. However, if you are building a direct API integration, you will need to handle these events yourself.

Example

curl https://api.anthropic.com/v1/complete \
     --header "anthropic-version: 2023-06-01" \
     --header "content-type: application/json" \
     --header "x-api-key: $ANTHROPIC_API_KEY" \
     --data '
{
  "model": "claude-2",
  "prompt": "\n\nHuman: Hello, world!\n\nAssistant:",
  "max_tokens_to_sample": 256,
  "stream": true
}
'

event: completion
data: {"type": "completion", "completion": " Hello", "stop_reason": null, "model": "claude-2.0"}

event: completion
data: {"type": "completion", "completion": "!", "stop_reason": null, "model": "claude-2.0"}

event: ping
data: {"type": "ping"}

event: completion
data: {"type": "completion", "completion": " My", "stop_reason": null, "model": "claude-2.0"}

event: completion
data: {"type": "completion", "completion": " name", "stop_reason": null, "model": "claude-2.0"}

event: completion
data: {"type": "completion", "completion": " is", "stop_reason": null, "model": "claude-2.0"}

event: completion
data: {"type": "completion", "completion": " Claude", "stop_reason": null, "model": "claude-2.0"}

event: completion
data: {"type": "completion", "completion": ".", "stop_reason": null, "model": "claude-2.0"}

event: completion
data: {"type": "completion", "completion": "", "stop_reason": "stop_sequence", "model": "claude-2.0"}

Events

Each event includes a named event type and associated JSON data.

Event types: completion, ping, error.

Error event types

We may occasionally send errors in the event stream. For example, during periods of high usage, you may receive an overloaded_error, which would normally correspond to an HTTP 529 in a non-streaming context:

event: completion
data: {"completion": " Hello", "stop_reason": null, "model": "claude-2.0"}

event: error
data: {"error": {"type": "overloaded_error", "message": "Overloaded"}}

Older API versions

If you are using an API version prior to 2023-06-01, the response shape will be different. See versioning for details.