Streaming responses from Anthropic Claude models

This example demonstrates how to stream the response of an Anthropic model with Python. Instead of waiting for the complete answer, the script prints each fragment as the model produces it and reports the time to first token.

Requirements

• Python 3.8+
• An Anthropic API key

Steps for running this example in the shell

1. Install dependencies:

python -m venv .venv

# macOS/Linux:
source .venv/bin/activate

# Windows Command Prompt:
.venv\Scripts\activate.bat

# Windows PowerShell:
.venv\Scripts\Activate.ps1

pip install -r requirements.txt

2. Export your API key as an environment variable:

export ANTHROPIC_API_KEY="..." # Windows cmd: set ANTHROPIC_API_KEY="..." # Windows PowerShell: $env:ANTHROPIC_API_KEY="..."

3. Run the script:

python anthropic-claude-streaming.py

Output

When you run the script, it sends a user prompt to a Claude model (claude-haiku-4-5) and prints the answer as it is generated. Once the stream ends, it prints the model identifier, the time to first token, the total latency, and the token counts. The example uses the messages.stream() helper, which accumulates the streamed events so that get_final_message() returns the complete message together with its usage.