Using Claude 3 Opus with Anthropic API in Python
With the recent release of the Claude 3 family, this was the perfect opportunity to use the Claude API access that I was recently granted.
The Claude 3 family boasts a huge improvement from the prior family, Claude 2. The introductory post goes into greater detail.
In this post, we’ll explore how to invoke Claude 3 Opus using the Anthropic SDK.
Getting started
For the purposes of this post, we’ll leverage the Python Anthropic SDK.
1pip3 install anthropicTo authorize requests, please export the ANTHROPIC_API_KEY:
1export ANTHROPIC_API_KEY="sk..."How to invoke Claude 3 Opus
The API for invoking Anthropic models is simple.
The model id we’ll use is the Opus release from Feb. 29th, 2024: claude-3-opus-20240229.
If you wish to target other models, please refer to Anthropic’s model page for more information.
1from anthropic import Anthropic
2
3model = "claude-3-opus-20240229"
4
5client = Anthropic()
6
7message = client.messages.create(
8 max_tokens=1024,
9 messages=[{"role": "user", "content": "Hello! How are you?"}],
10 model=model,
11)
12print(message.content)
13print(f"Input tokens: {message.usage.input_tokens}")
14print(f"Output tokens: {message.usage.output_tokens}")Output:
1[ContentBlock(text="Hello! I'm doing well, thank you for asking. As an AI language model, I don't have feelings, but I'm functioning properly and ready to assist you with any questions or tasks you may have. How can I help you today?", type='text')]
2Input tokens 13
3Output tokens 53In the example above, the Anthropic() client was instantiated and automatically referenced the environment variable from earlier.
Then, using the Messages API, we created a new message.
How to invoke Claude 3 Opus with Streaming
Similarly to before, we also have the option to stream output using Anthropic’s API.
1from anthropic import Anthropic
2
3model = "claude-3-opus-20240229"
4
5client = Anthropic()
6
7stream = client.messages.create(
8 max_tokens=1024,
9 messages=[
10 {
11 "role": "user",
12 "content": "Hello! How are you?",
13 }
14 ],
15 model=model,
16 stream=True,
17)
18
19message = ""
20input_tokens = None
21output_tokens = None
22for event in stream:
23 match event.type:
24 case "message_start":
25 input_tokens = event.message.usage.input_tokens
26 case "content_block_start":
27 message += event.content_block.text
28 case "content_block_delta":
29 message += event.delta.text
30 case "message_delta":
31 output_tokens = event.usage.output_tokens
32 case "content_block_stop" | "message_stop":
33 ...
34
35print(message)
36print(f"Input tokens: {input_tokens}")
37print(f"Output tokens: {output_tokens}")Output:
1Hello! As an AI language model, I don't have feelings, but I'm functioning well and ready to assist you. How can I help you today?
2Input tokens: 13
3Output tokens: 35For the sake of the example, I concatenated the “text”-based events together to showcase the full output.
For asynchronous eventing, please refer to the Anthropic SDK documentation.