Claude API Course is Anthropic’s official code first curriculum at github.com/anthropics/courses, with twenty two thousand stars as of mid 2026. Five Jupyter notebook sub courses covering the Claude API fundamentals, prompt engineering interactive tutorial, real world prompting, prompt evaluations, and tool use. Free, CC BY NC 4.0 license, no certificate. Python only. Default model is Claude 3 Haiku to keep API costs down.
Claude API Course: What You’ll Learn
This is the code first companion to Anthropic Academy on this site. Where the Academy is video first with certificates, this GitHub repo is Jupyter notebooks you clone and run. Five sub courses, twelve to eighteen hours total. No videos, no certificate. The repo is the closest thing to a Claude SDK reference maintained by Anthropic.
Claude API Course: The Five Sub Courses
- Anthropic API Fundamentals. Six notebooks. Getting an API key, Messages format, comparing Claude models, model parameters, streaming responses, vision and multimodal.
- Prompt Engineering Interactive Tutorial. Nine chapters plus appendix. Beginner chapters on basic prompt structure, being clear and direct, assigning roles. Intermediate chapters on separating data from instructions, formatting output, precognition, using examples. Advanced chapters on avoiding hallucinations, building complex prompts. Available also as AWS Workshop and Google Sheets variant.
- Real World Prompting. Five lessons. Prompting recap, medical prompt, prompt engineering process, call summarizer, customer support AI. Prerequisite is the interactive tutorial.
- Prompt Evaluations. Nine lessons. Intro to evals, workbench evals, code graded evals, classification evals, Promptfoo introduction, custom graders, model graded evals, custom model graded.
- Tool Use. Six lessons. Tool use overview, first simple tool, structured outputs, complete workflow, tool choice, chatbot with multiple tools. The capstone is a multi tool agentic chatbot.
Prerequisites
Python fluency. An Anthropic account with billing enabled. An API key from the Anthropic Console. Real API spend required, no mock responses or pre recorded outputs are shipped. Default model is Claude 3 Haiku to keep costs down. As of mid 2026 Claude 3 Haiku is a legacy model predating Claude 3.5, 3.7, 4, and 4.5. You can swap in any current Claude model by changing the model ID.
Distinguishing From Anthropic Academy
This is the most common confusion. Anthropic has two educational offerings and they are complementary. We covered Anthropic Academy on this site as Post 04.
Anthropic Academy at anthropic.skilljar.com is video first. Roughly twenty courses. Skilljar certificates. Broad coverage from non technical AI literacy through MCP and cloud deployment.
Anthropic Courses on GitHub at github.com/anthropics/courses is code first. Five notebook sub courses. No certificates. Narrow API focused coverage.
If you learn better from video, do the Academy. If you learn better from runnable code, do the GitHub repo. If you want both, do the GitHub repo first to get your hands on the SDK, then do the Academy to fill in conceptual gaps.
Maintenance Status
As of mid 2026 the repository has uneven maintenance. Real World Prompting and API Fundamentals were updated in late 2025. The other three sub courses, Prompt Engineering Interactive Tutorial, Prompt Evaluations, and Tool Use, have not been updated since September 2024. The SDK patterns they teach are still valid but predate newer features like extended thinking and MCP support.
A Worked Example, the Tool Use Capstone in Depth
The Tool Use sub course ends with a multi tool agentic chatbot in lesson six. You declare multiple tools as Python functions, hand them to the Claude SDK, and the chatbot uses them as needed. The pattern is the same function calling pattern every other agent course on this site teaches, but expressed in the Anthropic SDK directly without smolagents, LangGraph, or AutoGen abstractions. By the end of the lesson you have the foundation to build any tool using agent on the Claude API.
Claude API Course: Common Mistakes to Avoid
- Using Claude 3 Haiku verbatim in 2026. The README defaults to it for cost. Swap to a current model like Claude Sonnet 4.5 for better behavior on most tasks.
- Confusing with Anthropic Academy. They are complementary, not duplicative. See the dedicated section.
- Expecting TypeScript samples. The repo is Python only. TypeScript developers must translate.
- Trusting the older sub courses for current SDK patterns. Three of five sub courses were last updated September 2024. The SDK has evolved.
Claude API Course: Best Practices
- Do the sub courses in the order the README suggests. API Fundamentals, then Prompt Engineering, then Real World Prompting, then Prompt Evaluations, then Tool Use.
- Swap Claude 3 Haiku for a current model early. The legacy model is fine for cost sensitive testing but lacks current capabilities.
- Pair with the Anthropic Academy on this site for the video complement.
- Run the Tool Use capstone before doing any agent framework course on this site. It is the foundation.
Claude API Course: Frequently Asked Questions
Is the Anthropic GitHub courses repo free?
Yes. CC BY NC 4.0 license. Free to read and run. No redistribution under the NonCommercial clause.
Does it issue a certificate?
No certificate, no badge. Portfolio piece is the multi tool agentic chatbot from Tool Use lesson six.
What language is it?
Python only via the anthropic Python SDK.
What model does it use by default?
Claude 3 Haiku, explicitly to keep API costs down. Swap to a current model by changing the model ID.
How is this different from Anthropic Academy?
GitHub is code first with notebooks. Academy is video first with certificates. Complementary.
Is the repo maintained?
Unevenly. Two of five sub courses updated in late 2025. Three last touched September 2024.
Claude API Course is the code first companion to Anthropic Academy. Pair with the Academy for video and certificates, with the HF and Microsoft MCP courses for MCP depth.