Claude Code's concise mode cuts token usage by 40%.
I’ve been using the new concise style for roughly a week across several projects, including a React frontend, Python backend, and infrastructure scripts. The difference shows up right away. Standard mode often explains every file change at length, repeats context I’ve already supplied, and adds unnecessary “here’s what I’m doing” preambles. Concise removes that clutter.
How does the model's internal process change with concise mode?
The model still works through the same steps internally, which is visible in the thinking traces when verbose logging is enabled. Only the final response is shortened. Function signatures no longer repeat docstrings. Diff hunks come without the surrounding “I’m now editing X file” commentary. Error explanations focus on the fix that matters.
# Before (standard)
I'll update the authentication middleware to handle the new token format.
Let me first examine the current implementation...
# After (concise)
Updated auth middleware for new token format. Changed `validate_token()`
signature in `middleware/auth.py:42` to accept `token_version` param.The token savings accumulate quickly. A typical 5‑file refactor that previously used 8‑12k output tokens now takes about 4‑6k. Across an entire day of coding, that can make a noticeable difference on the API bill.
When should you switch back to standard mode?
There are two situations when I switch back to standard mode:
- Debugging unfamiliar codebases
When I enter a repository I have never seen, the additional context in standard mode helps me understand why the model made specific choices. Concise mode takes a simpler approach and assumes I already understand the architecture.
- Complex multi‑step planning
If I ask it to “plan the migration from REST to GraphQL across these 12 services,” I need the full reasoning trail. Concise mode provides the task list, but leaves out the tradeoff analysis.
{
"outputStyle": "concise",
"verboseLogging": false
}How can you enable concise mode in a single session?
For a single session, use /style concise in the CLI. I also keep a shell alias called cccon that starts Claude Code with concise mode, my preferred model (sonnet‑4), and the project context file already loaded.
One gotcha
In multi‑file edits, concise mode can occasionally omit file paths when the change is minor, such as a single‑line fix. I discovered this while reviewing a PR: the model had updated three config files but displayed the diff for only one. Now I run /diff after every batch edit to verify that nothing was missed.
Who benefits most from using concise mode?
It is worth enabling if you can read diffs comfortably and do not need step‑by‑step guidance. The mode reduces noise without removing useful information.
All Replies （3）
Want a live back-and-forth? Join the global AI chat room — login to talk.
My context window usually dies during refactors. Does this actually stop the hallucination spike?
I've been using the new concise style for roughly a week across several projects, including a React frontend, Python backend, and infrastructure scripts. The difference shows up right away. Standard mode often explains every file change at length, repeats context I've already supplied, and adds unnecessary "here's what I'm doing" preambles. Concise removes that clutter.
What changes: The model still works through the same steps internally, which is visible in the thinking traces when verbose logging is enabled. Only the final response is shortened. Function signatures no longer repeat docstrings. Diff hunks come without the surrounding "I'm now editing X file" commentary. Error explanations focus on the fix that matters.
# Before (standard)
I'll update the authentication middleware to handle the new token format. Let me first examine the current implementation...
# After (concise)
Updated auth middleware for new token format. Changed `validate_token()` signature in `middleware/auth.py:42` to accept `token_version` param.The token savings accumulate quickly. A typical 5-file refactor that previously used 8-12k output tokens now takes about 4-6k. Across an entire day of coding, that can make a noticeable difference on the API bill.
Where it falls short: There are two situations when I switch back to standard mode:
- Debugging unfamiliar codebases - When I enter a repository I have never seen, the additional context in standard mode helps me understand the codebase faster.
- Complex multi-step reasoning - When working through intricate logic chains, the extra explanation prevents me from losing track of intermediate conclusions.
Seeing CI logs on one screen is a dream. Which tool are you using for the logs? I use concise mode to cut token noise, but switch back to standard mode when debugging unfamiliar codebases—the extra context helps.
Saving 40% on tokens is a game-changer—especially when paired with the right prompt. For instance, I’ve been using the
concisemode across React, Python, and infrastructure scripts for a week now, and the difference is immediate. Standard mode often drowns you in explanations, repeats context you’ve already provided, and adds unnecessary intros like "Here’s what I’m doing." Concise strips all that away—like how function signatures now omit redundant docstrings, or diff hunks skip the "I’m editing X file" preamble—while keeping the core logic intact. For example, instead of:You get:
The token savings add up fast—my 5-file refactors now use 4-6k tokens instead of 8-12k. Still, I switch back to standard mode when debugging unfamiliar codebases or when I need that extra context to wrap my head around a new system.