5 Claude Code Commands to Stop Burning Money on Tokens
If you use Claude Code every day and your bill keeps climbing, the problem might not be how much you use it. It might be that you never check what's eating your tokens, or you run every task on the most expensive model because you didn't know you could switch.
I use five slash commands on repeat to keep my context clean and my costs predictable. Each one takes a second to type. Together they stopped the surprise charges and the moments where Claude suddenly "forgets" what we were building.
Why this matters
Claude Code bills by tokens, the chunks of text going in and out of each session. Long conversations, heavy models, and bloated context all add up fast. Most people only notice when the invoice lands or Claude starts giving vague answers halfway through a project. These commands put you back in control before either of those happens.
Track what you're spending
1. /usage to see your session cost at a glance
Type /usage anytime to see what your current session has cost, where you stand against your plan limits, and your recent activity stats. No more guessing until the end of the month. I check this when a session starts feeling expensive or when I'm testing something new and want to know the damage in real time.
2. /context to see what's filling your window
Type /context for a live breakdown of how full your context window is right now. You can see exactly what's taking up space: conversation history, files, tool output, all of it. This is the command I reach for when Claude starts feeling slow or forgetful. Usually the window is nearly full and I know it's time to compress or switch tasks.
Key insight: Claude doesn't get worse over time. Your context window just fills up, and
/contextshows you exactly when that's happening.
Match the model to the job
3. /model to switch models mid-session
Type /model to change which AI model Claude uses, and save your choice as the default for future sessions. Bump up to a smarter model for a big refactor or complex planning. Drop down to a lighter one for quick edits, file organization, or simple fixes. Most people run everything on the top model and wonder why their bill is high. You don't need premium intelligence for every task.
Keep your session lean
4. /rewind to undo mistakes without confusing Claude
Type /rewind to roll the conversation and your code back to an earlier checkpoint. If you make a mistake, don't tell Claude to "revert back to the previous version." That phrasing often confuses it and burns more tokens trying to figure out what you mean. /rewind is explicit. It goes to a saved point in your session and picks up from there.
5. /compact to summarize and keep going
Type /compact to summarize your entire conversation and free up context without starting a new chat. You can add a short note to control what it keeps, like /compact focus on the API changes we decided on. Instead of your context filling up and Claude losing the plot halfway through a project, you compress and keep going. This is my favorite command and the one most people don't know exists.
When to use which
- Want to know what you've spent?
/usage - Claude feels slow or forgetful?
/context, then/compact - Hard thinking vs. quick execution?
/model - Made a wrong turn and need to undo?
/rewind - Deep in a long task but not done yet?
/compact
Five commands. A few seconds each. Your sessions stay focused, recoverable, and a lot cheaper.
Here are some related guides to check out: