Claude Opus 5 Fast Mode: 2.5× Speed at Double the Price
Claude Opus 5 offers a fast mode running roughly 2.5× quicker at $10/$50 per million tokens, giving developers independent dials for speed and reasoning depth.
Alongside its standard $5/$25 pricing, Claude Opus 5 ships with a fast mode: roughly 2.5× serving speed at $10 per million input tokens and $50 per million output — double the base price.
Two Independent Dials
Fast mode is distinct from the model's new effort toggle, and the two compose:
- Effort (low/medium/high) controls how deeply the model thinks
- Fast mode controls how quickly tokens are served
A latency-sensitive but genuinely hard task can run high effort in fast mode. A batch job can run low effort at standard speed for minimum cost. Between the two dials, one model ID now spans a 4× price range and a wide latency envelope.
Where the Premium Pays
The 2× surcharge makes sense where a human is waiting:
- Interactive coding sessions in Claude Code
- Customer-facing chat and copilot features with response-time SLOs
- Live document review during meetings
It's wasted on background pipelines, evals, and report generation — the workloads Anthropic instead points at the Message Batches API, which trades latency the other way and unlocks 300K output tokens in beta.
The Precedent
Fast mode continues the pattern Anthropic established with fast mode for Claude Code on earlier Opus models: same model quality, faster output, priced as a premium serving tier. Bringing it to Opus 5 at launch — alongside effort control — turns model configuration into a two-axis decision that used to require choosing between entirely different models.
Practical Guidance
Early adopter advice is consistent: default everything to standard speed and medium effort, then upgrade specific routes where latency measurably affects users. The teams that flip fast mode on globally are paying double for pipelines nobody is watching.