お金さえ払えば、Claude Codeは最大2.5倍速くなる──従量課金の「/fast」を調べてみたClaude Code's /fast mode is not included in Pro or Max monthly plans and…

Claude Code, Anthropic's command-line coding assistant, includes a /fast mode that promises to trim waiting time by delivering responses up to 2.5 times faster. The detail that matters most to developers budgeting their tooling, however, is how it is billed: the acceleration is not bundled into the Pro or Max monthly subscription plans. Instead, it runs on pay-as-you-go billing, which effectively turns response speed into a metered, usage-based expense rather than a flat-rate feature.

The observation comes from a Qiita blog post whose author set out to benchmark the mode directly. After preparing test prompts and expecting to simply press /fast to shorten Claude Code's latency, they paused to consult the official documentation and stopped short. The takeaway they surfaced is that /fast is a paid option layered on top of the subscription tiers, so enabling it without understanding the pricing model could produce unexpected charges. That framing is a useful reminder that "faster" and "free" are not the same thing inside modern AI developer tools.

At a practical level, the trade-off is straightforward to state but harder to optimize. In exchange for meaningfully lower latency, users accept additional cost that scales with how much they use the mode. The headline figure of up to 2.5x faster is a ceiling rather than a guarantee, so real-world gains are likely to vary depending on prompt complexity, context size, and the type of task being run. For interactive coding sessions where a developer is waiting on each turn, shaving seconds off responses can compound into a noticeably smoother workflow. For long-running or batch-style automation, the same speed premium may be harder to justify against the cumulative bill.

Understanding where /fast sits requires a little background on how Anthropic structures access to Claude Code. The Pro and Max plans are subscription tiers that grant a bundle of usage under a monthly fee, which appeals to individuals and teams that prefer predictable costs. Pay-as-you-go, by contrast, charges according to consumption and is typically tied to API-style metering. By placing /fast outside the subscription bundles and inside the usage-based lane, Anthropic appears to be treating priority speed as a consumable resource, similar to how cloud providers often price faster compute or lower-latency tiers as add-ons rather than baseline entitlements.

This design is consistent with a broader pattern across large language model services, where inference latency and throughput are increasingly sold as differentiated service levels. Providers commonly offer standard queues alongside priority or expedited options, and the cost of that priority reflects the compute and scheduling required to serve requests more quickly. Framed that way, /fast is less an anomaly and more an example of speed becoming an explicit pricing dimension in the agentic coding market.

The context also helps explain why developers evaluate these choices carefully. Claude Code competes for attention alongside tools such as GitHub Copilot, Cursor, and open-source assistants like Aider, each with its own mix of subscription and consumption-based pricing. In that landscape, latency is a genuine differentiator because agentic coding assistants often chain multiple model calls together to plan, edit, and verify code. Small per-request delays can accumulate across a session, which is precisely why an option that reduces waiting time is attractive, and why its cost structure deserves scrutiny before it is switched on.

For anyone considering the mode, the sensible approach is the one the blog implicitly recommends: read the documentation, confirm the billing implications, and weigh the speed benefit against expected usage. Teams that run Claude Code heavily throughout the day may find the latency reduction worth the metered cost, while lighter or more budget-constrained users may prefer to stay on their existing plan. Because the exact economics depend on individual workloads and current pricing, developers should verify figures against Anthropic's official documentation rather than assume a fixed rate.

The wider point is that AI coding tools are maturing into products with tiered performance, and speed is now something users may pay for separately. Claude Code's /fast mode illustrates that shift clearly, and the practical lesson is to treat the speed-versus-cost tradeoff as a deliberate decision rather than a default.