How much does Claude Opus 5.5 fast mode cost?

$8 per million input tokens and $40 per million output tokens, twice the standard $4 and $20 rates.

How much faster is Opus 5.5 fast mode?

Anthropic says up to 2.5 times faster. The standard mode is already about 30% faster than Opus 5.

When is Opus 5.5 fast mode worth it?

When a person is waiting on the response, such as interactive coding or a customer-facing answer. It is wasted on batch jobs, evals and background agents.

Opus 5.5 fast mode economics

Updated September 22, 2026 · first published September 22, 2026

Claude Opus 5.5 has two price lists. Standard is $4 per million input tokens and $20 per million output tokens. Fast mode is $8 and $40, exactly double, for up to 2.5× the output speed. The model is the same. The extra money buys time and nothing else, so the only question is whose time.

What the premium buys

Take a step that produces 5,000 output tokens. At standard rates the output costs $0.10. In fast mode it costs $0.20. If the step takes 60 seconds at standard speed, a 2.5× speedup brings it to about 24 seconds. You pay an extra $0.10 to save 36 seconds.

That works out to about $10 for each hour of waiting removed. Step length does not change the ratio, because the extra cost and the time saved both grow with output tokens. It does depend on your standard-mode speed: this example assumes about 83 output tokens per second, so plug in your own. Input also doubles, so prefix-heavy steps pay a bit more per second saved.

When that is cheap

A developer waiting on an interactive coding agent. At $10 per hour of waiting removed, fast mode costs far less than the time of anyone who is paid to sit there.
A customer-facing answer where latency affects conversion. Price the speedup against drop-off, not against the token bill.
An incident, where every minute has a cost.

When it is pure waste

Background agents, nightly jobs and batch pipelines. Nobody is waiting, so the extra spend buys nothing.
Evals and replays. These are exactly the workloads that multiply token counts and run on a schedule.
Subagents that run in parallel. Wall-clock time is set by the slowest branch and the orchestrator, not by each call.

Standard Opus 5.5 is already about 30% faster than Opus 5, per Anthropic. Some latency complaints that would have justified fast mode on Opus 5 may be solved by the upgrade alone. Measure the standard speed before paying for more.

Controls to put in place

Make fast mode an explicit, per-route decision. It should never be a global default or a flag that an engineer turned on once and forgot.
Tag spend by mode. A fast-mode request at double the unit price looks like a normal request on a dashboard that groups by model only.
Alert on the fast-mode share of output tokens. If it rises on routes where no user is waiting, something was switched on that should not have been.
Budget effort and speed together. Max effort multiplies the token count, and fast mode doubles the price of each token. The combination is the most expensive way to run the model.

Want this applied to your own LLM spend? FinOps LLM runs a free audit of your AI costs and shows where the savings are. Book free audit →

Back to research