Effort level and fast mode are independent Claude controls. Effort scales how many tokens the model spends on text, tool calls, and thinking. Fast mode keeps quality identical and pays a flat latency premium - $8/$40 per million tokens on Opus 5.5 (standard $4/$20) and $10/$50 on Opus 5 and Opus 4.8 (standard $5/$25) - for up to 2.5x speed.
Rule
| Lever | What changes | Cost effect |
|---|---|---|
| Effort | Token spend (and sometimes quality) | Lower effort = fewer tokens |
| Fast mode | Latency only | Flat ~2x premium: $8/$40 on Opus 5.5, $10/$50 on Opus 5 / 4.8 |
- Stack low effort + fast mode for a simple, latency-sensitive task.
- For quality-critical work, drop fast mode and raise effort - paying more per token for identical output makes no sense next to a dial that changes what the model produces.
- Fast mode on Opus 4.7 was removed 2026-07-24; migrate to Opus 5.5, Opus 5, or Opus 4.8.
- Opus 5.5, Sonnet 5.5, and the Fable models cannot disable thinking; on adaptive models, effort is the real depth dial.
Controls: /effort, --effort, CLAUDE_CODE_EFFORT_LEVEL, effortLevel in settings; /fast or "fastMode": true. The first fast-mode message in a conversation pays the full uncached fast-mode input price for the entire existing context - turn it on at session start, not mid-way.
Full playbook: Why fast mode isn’t a lower effort level. Tier choice sits underneath both: Why most agents default to the wrong Claude tier.