CC - Models & Effort
Follow along by running prompts and commands!
Alright, now let's actually start using Claude Code. We're going to run it from the terminal. Open your project in VS Code so you can watch files appear and change as Claude works — but the interaction all happens in the terminal. Any terminal should work.
Your Claude Code subscription also gives you access to the desktop app, which includes additional tools including Cowork, which is like Code but for office work, including editing documents and spreadsheets.
Install Claude Code globally if you haven't already:
npm install -g @anthropic-ai/claude-codeThen start a session from your project directory:
cd your-project
claudeClaude Code has access to whatever directory you're in, so it can read and work with your files directly from the terminal prompt.
Picking a Model
The first thing to know about is the model selector in the Claude Code panel. Claude Code uses Anthropic's Claude models exclusively:
Claude Haiku — the fastest and most affordable, great for quick tasks
Claude Sonnet — the default, best balance of speed and capability (most requests will be fine here)
Claude Opus — the most powerful model, best for complex reasoning and planning
You can switch models at any time using the /model command in the terminal.
✏️ Run this command
/modelUse the [escape] key to exit.
A quick note on cost. Claude Code is billed based on tokens (input + output). Different models have different rates:
Haiku — cheapest per token
Sonnet — mid-range, the sweet spot for most work
Opus — most expensive, but most capable
A good starting point is Sonnet for general work, Opus for planning and complex tasks.
Effort & Ultrathink
Some Claude models also support extended thinking. You can enable this for deeper reasoning on hard problems. You can either set this with the /effort command, or you can enable deeper thinking on a specific prompt by including the keyword “ultrathink”.
A good starting point is Sonnet with “high” or “xhigh” effort. We will talk more about model strategy as we go.
✏️ Run this command
/effortUse the [escape] key to exit.
Usage
Claude Code is still a new product. The way usage is calculated and how much you get has changed recently and seems likely to change again in the future.
Your Claude Code plan has usage limits. To check your current usage, open Claude Code and type /usage.
✏️ Run this command
/usageUse the [escape] key to exit.
Notice that your usage is tracked for the current 5-hour session and the current week. If you hit the max session tokens within the 5-hour window, you won’t be able to use Claude Code again until the session resets at the end of the 5-hour session. Same with the weekly usage. Weekly usage includes all sessions during that week.
If you are interested in seeing more details about your usage, including which models you use the most, pressing the right arrow key will take you to the Stats section. There are a few things to see and I encourage you to experiment with navigating this section with your arrow keys. The ESC key will exit.
Working within your limits
Getting the most out of Claude Code includes monitoring your usage so you don’t have lots of downtime because you ran out of tokens. Usage is a factor of:
Model
Effort
Prompt
Context
We will talk about prompts and context at a later time. For now, know that selecting a cheaper model and/or lower effort will decrease your usage. Haiku is the cheapest model and can be effective for smaller tasks, testing, or carrying out plans designed by Opus or Sonnet. To get the most out of your session, you likely need to create a mix of work that uses different models.
Model recommendations
Haiku - Ignore
Sonnet - Use to implement plans from Opus
Opus - Use for planning tasks and for trickier implementations that Sonnet struggles with