Claude Opus 5.5 fast mode is now open to everyone on the Claude API. According to Anthropic’s ClaudeDevs account, the research preview is over and the waitlist is gone. You get the same model, up to 2.5x faster, for $8 per million input tokens and $40 per million output tokens.
Here are the facts worth knowing before you flip it on:
- Speed: up to 2.5x faster output, same Opus 5.5 model and quality.
- Price: $8 input / $40 output per million tokens, double the standard $4 / $20.
- Where: the Claude Platform API and Microsoft Foundry on Azure, with Google Cloud listed as “coming soon.”
- Who it helps: developers on paid API accounts. Chat app users see no change.
The short version: you pay double to get long answers streamed much faster, and that trade only makes sense in some workloads.
What Claude Opus 5.5 fast mode actually changes
Fast mode is not a smaller or cheaper model. Anthropic calls it “the same Opus 5.5,” just served in a faster configuration. Your prompts, tools, and results should behave the same.
The speedup applies to how quickly the answer streams once it starts. It does not shorten the wait before the first word appears, and it does not speed up your own tool calls. So a long code rewrite benefits a lot, while a one-line reply barely changes.
If you already use agents that write big diffs or long reports, that streaming speed is where you will feel it. If your slow step is a database query or a test run, it will not help.
Anthropic also says the price is flat across the full 1M-token context window, so long prompts do not push you into a higher tier. That makes cost easy to estimate, but easy to underestimate too, since big contexts add up fast at $8 per million.
What fast mode costs compared with standard speed
The price is a flat 2x on both input and output tokens. This table puts the two modes side by side for Opus 5.5.
| Item | Standard Opus 5.5 | Fast mode |
|---|---|---|
| Input price (per 1M tokens) | $4 | $8 |
| Output price (per 1M tokens) | $20 | $40 |
| Output speed | Normal | Up to 2.5x faster |
| Time to first token | Normal | Not shortened |
| Example: 100K in + 20K out | $0.80 | $1.60 |
The example row is simple math from the listed rates, not an Anthropic figure. It shows the practical point: every request costs exactly twice as much, so the savings have to come from your time or your users’ time.
How to turn on fast mode in the API and Claude Code
In the API, add "speed": "fast" to an Opus 5.5 request. Reports on the launch say the old beta header is no longer needed, though requests that still send it keep working, according to Tech City Authority.
In Claude Code, type /fast. Pro, Max, Team, and Enterprise users pay for it from usage credits, and Team and Enterprise owners have to enable it first. If you want a look at what the tool can do on hard jobs, see our post on how Claude Code wrote a Linux driver for an unsupported scanner.
The catches: cache misses, rate limits, and platforms
Fast mode has a few quirks that can quietly raise your bill or break a pipeline. These are the ones to plan for:
- Switching speeds mid-session causes a prompt-cache miss, so you pay full price to re-read your context. Turn it on at the start of a session.
- Fast mode has its own capacity limits, separate from standard Opus. When you hit them, you get a 429 error and need to retry.
- Falling back to standard speed is your job. Your code must catch the error and resend without the fast setting.
- It does not work with the Batch API, and Google Cloud is not live yet.
The cache point matters most. A team that flips between modes on every request can end up paying far more than 2x.
Is Claude Opus 5.5 fast mode worth paying double for?
It is worth it when a person is waiting on a long answer: live coding sessions, interactive agents, and tools where lag loses users. It is a poor fit for overnight batch jobs, where nobody notices the speed.
If your goal is simply lower cost, look the other way. Cheaper models like the one in our DeepSeek V4.1 Flash pricing breakdown can cover simpler tasks, and you can keep Opus for the hard ones.
Frequently Asked Questions
Is Opus 5.5 fast mode a different model?
No. Anthropic says it is the same Opus 5.5 with a faster serving setup, so quality should not change.
Is the waitlist really gone?
Yes, for Opus 5.5. Anthropic says the research preview has ended. Some reports say older Opus versions may still need access approval, so check your account.
Does fast mode make the first word appear sooner?
No. It speeds up output after the response begins. The wait before the first token stays about the same.
Can I use fast mode in the Claude chat app?
Fast mode is aimed at the API and Claude Code. Chat app users should not expect any change.
A sensible test: run one real task twice, once at each speed, and compare the wall-clock time and the final cost. Ten minutes of testing will tell you more than any benchmark.
Our verdict
Treat Claude Opus 5.5 fast mode like a paid express lane, not a default setting. Use it for interactive work where speed saves real minutes, leave it off for background jobs, and decide at the start of a session so the cache works for you.
Leave a Reply