How far each Claude plan really goes
Every plan is worth about 60× its price. A bigger plan buys room, not a better rate.
From our own paid accounts, November 2026. We don’t sell these plans.One week of use, priced at API rates
≈ 0.8B tokens. Opus 5.5 lasts 3–4 days.
About 4B tokens. Opus 5.5 lasts all week.
7–15B tokens. Opus 5.5 lasts all week.
Tap a dot to see whose week it was.
You code a few days a week. Let Opus 5.5 write the plan, then hand it to DeepSeek 4.1 Flash through OpenCode Go.
You code most days. Opus 5.5 on high all week, with a little Fable 5.1, is already very good.
Agents run all day. Only worth it once $100 runs out before the week does.
How to spend it
One main task hands work down. Claude Code lets you mix models freely.
Opus 5.5 on highThe main task, kept small.
Opus 5.5 medium, Sonnet 5.5 highOn $20: DeepSeek 4.1 Flash through OpenCode Go.
A little Fable 5.1Skip it on $20: it runs on usage credits there.
- Stay in one session. Don’t keep opening new ones.
- Keep the main context small. Spin off subtasks and one-off subagents.
- Use DeepSeek alone when it takes over, and let it finish the job.
What other people measured
The four clearest numbers. Reports say Max 20x, the old name for Ultra.
A full week, from $2.31–$4.45 per 1% of the weekly limit.
Fast-and-Pythonic, Oct 25One week on Opus 5.5. 98.96% of it was cache reads.
ganpei777, Oct 18Two October weeks, 6B and 11B tokens, from local logs.
AutoHenry, Oct 27A $200 week held about twice a $100 week.
Yoshinobu-Suzuki, Sep 6Quick answers
Does the $20 plan include Fable?
How much of Max can Fable use?
Does Ultra have a five-hour limit?
Why do the numbers differ so much?
Tokens left over? Make something.
Agent Skills that turn them into something you keep.
