I'm over my token limit... again

I keep finding myself needing to adjust my strategy for token management. What worked yesterday no longer works today.

I primarily use Claude, and the things that affect my usage change constantly.

Each dot is a dated change to Claude models, API prices, tokens used for the same work, or subscription limits. Bars are Claude Code versions published per month (Feb 2025 and Oct 2026 are partial months). Research done with Claude. Sources below.

I watch tokens every day, every hour, and adjust my workflow to try to maximize them.

I'm on the 20X Max plan ($200 a month) and have enough work to do each week that I usually hit the limit, so I need to throttle my usage.

I rarely use Fable, instead opting for Opus 5.5 for important work. Sometimes I can get away with letting sessions delegate to sub-agents using defaults (Opus 5.5 when the session is Opus), and other times I need to specify that they need to be running Sonnet, depending on how quickly tokens are being consumed.

That's my strategy this week. I'll be very interested to reflect back on this in five years and see how much it has changed.

Sources

I used Claude to research the timeline above. The main sources:

• Model launches, prices, and token changes: Claude Platform release notes and pricing
• Plan limits: Anthropic support, Opus 5.5 announcement, TechCrunch, BleepingComputer
• Claude Code updates: version history on npm