Opus 5.5 has just been completely unblocked!

  It's the first time in the history of the world.

Just recently, the ultra-fast version of Claude's most powerful model has been completely unblocked!

They have officially announced that Claude Opus 5.5's "Extreme Speed Mode" has completely ended the preview phase of research, cancelled all waiting lists, and is fully open to all developers!

It's hard, it's hard, it's hard, it's hard, it's just a line of code, and it's time for Claude to get off the ground, 2.5 times faster!

One developer exclaimed: "I'm announcing that Opus 5.5 is now my god".

For this ultra-fast version, some commented in one word: "P95 delay is more important than anything else for Agent Cycle".

Next, let's take a closer look at the official documentation, price traps, network configurations, and how to prevent the wallet from being emptied.

Don't get me wrong, the ultrafast model is not a "lightweight version" model that has been filtered by parameters.

This time, we don't have to choose between speed and quality.

According to an Anthropic official,The ultrafast model uses exactly the same model weight and intelligence levels as the standard version.

It is not a "low-intelligence" sorting model, but merely focuses resources on response speeds by optimizing underlying APIs and reasoning configurations.

The result is:The speed of outputs of tokens per second (OTPS) has risen to 2.5 times!

Its use of scenarios, explosive fast-coding, real-time dynamic modeling, and time-sensitive task-heavy boundary thrust.

Currently, the ultra-fast mode only supports Opus 5.5, Opus 5 and Opus 5. Opus 4.8。

Note: Sonnet, Haiku and other models are not supported. The previous version of Opus 4.7 was discontinued in June 2026.)

How do you use it? The method is simple and rough.

All you need to do is add a sentence to the API request:

speed: "fast"

It can be unlocked instantly in high-speed versions.

If you are a Claude Code user, you can enter /fast directly into the CLI and turn it on or off at any time; VS Code users can also switch directly via plugins.

It's 2.5 times faster, isn't it going to go up in price?

It's the opposite! This time, Anthropic is doing another wave of "quick rate cuts!"

Let's take a look at the core price comparison:

Opus 5 and Opus 4.8 ((extremely fast mode): $10 input / $50 output

Opus 5.5 ((extremely fast mode): input $8 / output $40

Yes, you're right, the strongest version is cheaper!

Crater avoidance guidelines

Here are a few crater avoidance guidelines to watch out for.

Free flow, free from resources:Extreme-speed models have independent speed limits.

If you touch the upper limit too hard, it will automatically slide back to normal speed, until the cooling time is over, and then it will automatically recover extremely quickly, not letting you die directly.

In the meantime, we're going to have to do something about it.Prompt Cache cannot be shared between high-speed mode and standard mode.

Therefore, the official suggestion is strong: start the ultra-fast mode at the beginning of the session, otherwise the mid-way switch may cause you to re-pay a non-cache input fee.

This initial fee is charged only once in a conversation, and you will not be charged again even if you switch to ultra-fast mode repeatedly.

Consumption「Use of quotas」

For Pro/Max/Team/Enterprise subscribers, the Extreme Speed Mode is an additional usage quota that is not included in the original subscription quota.

How to check bills

First, use the /status command to see how you've logged in.

1.Pro/Max Personal Subscription Users: Extreme speed mode is not included in your regular subscription limit, but consumes your "use limit". See Settings > Usage at claude.ai.

2.Team/Enterprise enterprise users: deducted from the organization's allowance. Ordinary users can view their spending with /usage and administrators in the background.

3.Claude Console (API client): On the Usage and Cost page in the background, by selecting Speed (Research Preview) from the "Group by" menu, you can see the money for the ultra-fast model separately.

The biggest sincerity of this opening is that there is no application!

For example,API developersYou don't have to email, you don't have to fill out a form, and you can add a parameter to your API request directly: "fast".

Note that the API and Microsoft Azure Foundry are now available, and Google Cloud is about to launch.

For example,Claude Code (CLI) usersIn the terminal, the interaction is instantaneous to take off.

Input /fast, switch on/off with the blank key, and return confirmation.

When you turn it on, the terminal will not only prompt "Fast mode ON", but a flash icon will also appear next to your prompt.

You can even tap /fast while Claude is thinking about the output, and it will answer this round at standard speed, and the next round will take off automatically.

Note that in Claude Code v2.1.280 and above, the ultrafast mode uses Opus 5.5 by default.

You can also write "fastMode" directly in the user profile: true.

The linking logic of the model switching is like this:

Cut it out.If you switch to a model that doesn't support high-speed mode (e.g. Sonnet) with /model after you turn on high-speed mode, the high-speed mode will automatically close.

Cut it out.Opus 5.5: If you switch back to Opus 5.5, it will automatically reset flash icons as long as your saved preferences are turned on.

In a cloud session, you can enter /fast on, but only for current sessions.

When the news came out, many people in the comments section were also worried: "The scariest thing about the default is that half of us may have burned $200 tokens before we knew it".

In order to prevent employees from burning the company's money, Anthropic has provided a highly sophisticated enterprise-level management system.

Forced to open the meeting

The administrator can set "fastModePerSessionOptIn" in settings.json or server-level configuration: true.

In this way, every new meeting, the high-speed mode is closed by default.

This can effectively control "Eat the Beast" employees who simultaneously run dozens of terminal missions.

Speed limits and sleek degradation

Extreme speed mode has its own independent speed limiting pool. If your request is too loud, touch the upper limit: Extreme speed mode will automatically return to standard speed, instead of directly returning the error card dead.

The lightning icon turns gray and enters a cooling state. After cooling, it automatically recovers at 2.5 times the speed.

Well, if your "use limit" flashes in the middle of the session, it will also be lowered to the standard speed to continue running, and the terminal will tell you that the limit is running out, so you won't have to wait.

Having spoken of saving money, it's important to mention that Claude recently went online with another explosive feature: vacuum compression.

In the past, you've been working on a long session of 200,000 tokens, and when you open it and come back, the cue word has expired.

Every time you let an AI write a line of code, it has to "read" those 200,000 tokens over and over again, and a few dollars instantly evaporate.

Right now,Claude has learned.「Automated storage」！ 

It will pause for about 54 minutes before the cache fails.It will automatically compress your long conversations into a refined summary.When you get back to work, just reading the abstract will get you thinking, significantly reducing the reuse of Tokens.

Please note that the following conditions must be met:

1. The terminal session uses an hour of word caching;

2. Up to 200,000 Tokens in the following context;

3. Keep the computer awake;

4. Continuous suspension of approximately 54 minutes (caught up to one hour before the caching failure)

Of course, saving money also comes at a cost.

Some developers on GitHub complained that their seven long sessions were automatically compressed overnight, causing AI to forget the details of some of the underlying technical decisions.

Don't panic, Anthropic listened to the advice and immediately turned off the switch.

You can modify settings.json:idle Compaction: false, or by default trigger thresholds using environmental variables (default is 200,000), and everything is in your hands.

And it's just today.Legend has it that Fable 5.5 has entered an early access stage and is already graying on some companies and individuals!

Currently, Fable 5.5 is only available to the public at the end of testing.

There is widespread speculation that this time point is very reasonable, which allows Anthropic to have more time before the IPO.

However, some have pointed out: "Fable 5.5 must be really, really great to prove that it makes sense to use it instead of Opus in a situation where it consumes so much use".

For most regular subscribers without an API, Fable 5.5 can be incredibly expensive: dropping a few long hints, and the 5-hour allowance runs out.

This means that in the beginning, only those who are willing to pay may be able to do so.More than $50 per million of tokensIt's the business owner who really makes use of this "performance monster".

Anyway, the 2.5 times faster version of Opus 5.5 has arrived.

No, no, no, no, no, no, no, no, no, no.