You have used Claude. But which Claude?

The Claude app (claude.ai, the desktop and mobile apps) is the chat product you talk to. The Claude API is the developer platform other products are built on. Same intelligence underneath, completely different audiences and ambitions. Here is what each one is for.

Claude “The App”

This is the chat product at claude.ai, the desktop app and the mobile app – probably the most commonly used and understood one – below screenshot shows the windows desktop.

Used by everyone and anyone such as Knowledge workers, writers, analysts, students, founders. If you can type a question, you can use it.

The aim: Open it, ask, get work done. Everything is bundled in — chat, web search, file creation, connectors to Google Drive, Gmail, Slack, artifacts for code and documents, skills for domain expertise, projects for separation of areas and memory across chats. You pay a monthly subscription and Anthropic runs the infrastructure – such as free, pro, max 5x and 20x.

Claude API (the developer platform)

Raw access to the models, plus the same building blocks the app uses exposed for you to compose into your own product. This is for the developers, engineers, product teams, people building AI into their own software with agents and workflows. You write the code, manage the tokens, hooks, MCP and design the UX.

Key takeaway: if you want to build something with Claude, you want the API.

What about Claude Console?

A central location where you can do administration such as generate API keys, monitor usage / costs, set spending limits and test prompts in the Workbench before wiring them into your code. The API is what your application calls. The Console is how you manage it. You will probably need both.

Claude Code?

Claude Code is the CLI for developers. It is built on the API but you talk to it like the app — it edits files, runs tests, executes shell commands, all from your terminal. If you write software this is super useful for you too – yes it works great with the terminal but also with Visual Studio.

Summary

The app is for people who want answers. The API, console, Claude code is for people who want to build. Pick the one that matches what you are trying to do. Hopefully this helps you understand what is out there (and subject to fast change!)

Every major model out there can summarise documents, write code and answer multi step questions – then if you decide to go with a specific vendor based on costs then great, but I don’t.

Doom and gloom?

I have read many books such as life 3.0 and If anyone builds it everyone dies and yes its doom and gloom for most of it where much is discussed around the warnings of alignment failure and lack of governance and guard rails. Some will read these books and dismiss it as sci-fi but once I read through these books it was too hard to dismiss. So I got thinking, which company out there is treating AI safety at its core – that lead me to Anthropic and its Constitution.

The Constitution that changed my thinking

The Constitution covers honesty, avoiding harm, being helpful and being transparent about uncertainty.  Claude will tell you when its unsure, it will refuse to do things that could cause harm not because someone developed a filter but because it’s a core principle of the model itself. I haven’t done the Constitution justice here you should read it for yourself https://www.anthropic.com/constitution but you will probably see why I am so curious about exploring Claude further.

However, if there is one important concept to take from today it’s the priority order of the Constitution below

Safety first, then ethics, then Anthropic’s rules, then user helpfulness.

Claude is exactly the model I want to use.

For those entering the AI space whether professionally or personally I wanted to give a quick overview on the different models on offer within the Claude family – when you would use them and why.

Read more: Overview of Claude AI Models: Which One to Choose?

Claude Opus — Deep reasoning & complex decisions

What it is

• The highest‑intelligence Claude model
• Designed for complex, ambiguous, multi‑step reasoning

Cost & performance

• Cost: High
• Latency: Moderate (slower, but intentional)

Reasoning

•  Full advanced reasoning support

Best used for

• Advanced software development and system design
• Large‑scale or enterprise architecture decisions
• Long‑running tasks that require sustained context
• Strategic planning and complex multi‑step problem solving
• Any task where thinking quality matters more than speed

Claude Sonnet — Balanced, general‑purpose intelligence

What it is

• A well‑balanced model that trades a small amount of depth for speed and cost
• Ideal for day‑to‑day professional work

Cost & performance

• Cost: Medium
• Latency: Fast

Reasoning

• Supports reasoning (not as deep as Opus)

Best used for

• Common coding and development tasks
• Documentation creation and editing
• Data analysis and visualization projects
• Content marketing and copywriting
• Image analysis

Claude Haiku — Fast, cheap, high‑volume work

What it is

• The most cost‑efficient and latency‑optimized Claude model
• Optimized for speed and scale, not deep reasoning

Cost & performance

• Cost: Low
• Latency: Fastest

Reasoning

• No advanced reasoning support

Best used for

• Quick code completions and suggestions
• Content moderation and filtering
• Data extraction and categorization
• Language translation

So to summarise – Use Opus when quality of thinking matters most. Use Sonnet when you want the best balance. Haiku should be used when speed and cost matter most.

Side note – Claude Mythos sits outside the normal model-selection conversation. It is a private-preview frontier model aimed at advanced cybersecurity use cases.