Fast mode is an Opus 4.6 configuration that prioritizes token generation speed over cost. It's not a different model and not a version with reduced reasoning -- it's the same Opus 4.6 with the same intelligence, just up to 2.5x faster responses. The tradeoff: significantly higher cost per token.
Personal note: Fast mode just launched at the time of writing. In the medium-sized tasks I've been able to test, the speed difference is noticeable, but I'm not yet sure it justifies paying double (2x) compared to standard Opus. During the 50% discount period (through February 16, 2026), fast mode costs the same as standard mode -- at that price, there's no question.
Don't confuse fast mode with effort level adjustment. Lowering effort level does reduce reasoning quality in exchange for speed. Fast mode sacrifices nothing in quality -- only cost.
Result:
> /fast
Fast mode ON · $15/$75 per Mtok (50% off through Feb 16)
Setup
1. Enable fast mode
Type /fast at any point during your session:
/fast
A ↯ icon appears next to the prompt while fast mode is active.
2. Disable when you don't need it
/fast
The same command toggles it off. When disabled, you stay on Opus 4.6 (it doesn't revert to your previous model).
3. Enable by default (optional)
To keep it always on, add to your user settings:
{
"fastMode": true
}
Reference
| Aspect | Detail |
|---|---|
| Command | /fast (toggle on/off) |
| Model | Opus 4.6 (same model, no quality change) |
| Speed | Up to 2.5x faster output tokens |
| Cost (< 200K) | $30 / $150 per MTok (input / output) |
| Cost (> 200K) | $60 / $225 per MTok (input / output) |
| Discount | 50% off through February 16, 2026 |
| Rate limits | Separate from standard Opus; falls back to standard speed when exhausted |
| Persistence | Persists across sessions |
| Availability | Pro/Max/Team/Enterprise plans with extra usage enabled |
When to use it
- Rapid iteration: code changes where waiting 30 seconds matters.
- Live debugging: interactive sessions where every second counts.
- Deadline-driven work: when speed justifies the extra cost.
For long autonomous tasks, batch processing, or CI/CD, standard mode is more cost-efficient.
Official docs: Speed up responses with fast mode