THE 5-HR ROLLING WINDOW
Claude counts TOKENS not messages.
Every 5 hrs the window resets.
You unlock EARLY as old tokens age out —
not at a fixed clock time.
Usage shared across web, desktop,
mobile AND Claude Code.
PLAN TOKEN BUDGETS
Pro $20/mo → ~1–1.5M tokens/5hr
Max 5x $100 → ~5x Pro limits
Max 20x $200 → ~20x Pro limits
Model cost matters MORE than plan:
Opus = 5x cost of Haiku
Sonnet = 3x cost of Haiku
HOW TO TRACK RIGHT NOW
Web/Desktop:
Settings → Usage → live dashboard
Claude Code terminal:
/status ← session usage
/cost ← session token cost
ccusage daily ← full breakdown
THE 6 BIGGEST TOKEN DRAINS
Wrong Model
Opus costs 5x Haiku.
Use Sonnet for 90%
of your daily work.
Stale Context
Long threads re-send
full history every msg.
Start fresh between tasks.
Big File Uploads
5 medium files =
30K+ tokens before
you even ask anything.
Extended Thinking ON
Token-heavy feature.
Switch OFF for
routine simple tasks.
Vague Prompts
"Make this better" =
3-5 rounds of back
and forth = 5x tokens.
All Tools Enabled
Every connector adds
tokens to EVERY msg.
Only enable what you need.
8 HABITS THAT CUT YOUR USAGE BY 40–60%
Right Model, Right Task
Haiku → simple Q&A
Sonnet → daily work
Opus → complex only
/clear Between Tasks
In Claude Code: type
/clear when switching
to unrelated work
/compact Long Sessions
Compresses old context
instead of re-sending
everything. Huge saver.
Batch Your Questions
"Do these 3 things:
[1][2][3]" = 1 message
vs 3x the overhead
Use Projects
Projects use RAG —
only loads relevant
context, not everything
Ask for Short Outputs
Add "Max 200 words"
or "Be concise."
Claude won't pad.
Turn Off Ext. Thinking
Toggle OFF for
routine tasks in
Settings or chat
Enable Extra Usage Cap
Settings > Usage >
Set $10–20 cap so you
never hard-block mid-work
See exactly how much your Claude plan is costing you →
Open the Free Calculator