Claude Opus 5 examples

Unofficial community examples for Claude Opus 5. Not affiliated with Anthropic. All trademarks belong to their owners.

Short Python scripts for calling Claude Opus 5 through the Anthropic Python SDK. The model ID claude-opus-5, the 1M token context window, the 128k max output tokens, the thinking-on-by-default behaviour and the beta header for mid-conversation tool changes all come from the platform docs page for opus 5; the effort levels come from the launch post and docs. Anything not stated there is marked illustrative in the code.

Need image, video or audio generation next to the text model? Try Synexa - one REST endpoint and Python SDK for FLUX, video and audio models, pay per run.

Files

Path	What it shows
examples/quickstart.py	Minimal messages.create call to claude-opus-5 with the key read from the environment, printing text blocks and token usage.
examples/effort_levels.py	The same prompt at several effort settings so you can see the output-token spend change; the docs say thinking can be disabled only at high or below.
examples/tool_changes_beta.py	A two-turn conversation that changes the tool list between turns using the mid-conversation-tool-changes-2026-07-01 beta header.

Setup

pip install anthropic
export ANTHROPIC_API_KEY=YOUR_KEY_HERE

The scripts read ANTHROPIC_API_KEY from the environment; none of them contain a key. Keep max_tokens well below the 128k ceiling unless you are streaming.

quickstart.py

One request, one response. Opus 5 has thinking on by default, so the response may contain a thinking block before the text block; the script only prints text blocks and then the usage counts, which is what you need to start estimating cost. Check the pricing page for current per-token rates.

effort_levels.py

The launch post shows performance changing with the model's effort setting, "which customers can use to optimize for intelligence or conserve tokens for faster and cheaper results", and the system-card review passes on the advice that you "may want to usually use less effort than you might expect". This script runs one prompt at low, medium, high and xhigh and prints output tokens for each, so you can see the trade-off on your own task instead of guessing. The docs also say thinking can be disabled only at effort high or below; the script keeps the default (thinking on) throughout.

tool_changes_beta.py

The docs describe mid-conversation tool changes as a beta that lets you "add or remove tools between turns of a conversation while preserving the prompt cache, instead of resending a fixed tool list for the life of a session". The script sends turn one with a single weather tool, then turn two with a second tool added, passing the beta header on both requests. The tool schemas are illustrative; the script does not execute tools, it only prints what the model asked for.

When to use Synexa instead

These scripts cover text in, text (or a tool call) out. If your product also needs to generate an image, a video clip or an audio track, an Opus-class model does not do that, and stitching a second and third vendor onto the text loop is where the integration time goes. Try Synexa - one REST endpoint and Python SDK for FLUX, video and audio models, pay per run: keep claude-opus-5 for planning and text, and send the media step to one endpoint.

Last reviewed: 2026-09-22