Claude Opus 5.5 is here, and it costs less to run than Opus 5

Anthropic released Claude Opus 5.5 on 22 September 2026. It is the first model in what the company calls the Claude 5.5 family, and the headline is not really about intelligence. It is about cost.

In Anthropic's own words, Opus 5.5 performs at the level of Claude Fable 5.1 on most work and costs 40 percent less to run than Opus 5. If you build anything on top of these models, that second half of the sentence is the part that matters.

What you pay for it now

These are the list prices on Anthropic's pricing page, per million tokens. Input is what you send. Output is what the model writes back.

• Fable 5.1: 10 dollars input, 50 dollars output
• Opus 5.5: 4 dollars input, 20 dollars output
• Sonnet 5: 2 dollars input, 10 dollars output
• Haiku 4.5: 1 dollar input, 5 dollars output

Read those four lines together and the shape of the market is clear. The top model costs ten times what the cheapest one does, and the model Anthropic now calls its daily driver sits at less than half the price of the one above it while doing, by the company's account, most of the same work.

What it is supposed to be better at

The announcement gives three examples, and they are worth repeating because they say more about the intended use than a benchmark score would.

1. One tester finished a 680,000 line code migration in under a day. Anthropic says that job would have taken an engineering team weeks.
2. Asked to cut load times across every page of a web app, Opus 5.5 succeeded 39 times out of 40. Opus 5 made smaller improvements and changed how the app behaved while doing it.
3. Given a single prompt and told to build a game, it scored higher than any other model the tester tried on graphics and polish.

Notice what those three have in common. None of them is a question with an answer. They are all long jobs with a lot of steps, the kind of work where a model has to keep track of what it has already done. That is where this release is aimed.

The safety part, briefly

Anthropic says Opus 5.5 scores best of any model it has tested on its automated behavioural audit, is less likely than recent models to take actions that are hard to undo, and is more resistant than Opus 5 to prompt injection. It was tested before release by outside evaluators.

Prompt injection matters more than it sounds. If a model reads a web page and that page contains instructions, a weaker model can be talked into following them. Any model you let near your files or your browser needs to be stubborn about that.

Does this change anything if you do not write code

Mostly no, and that is worth saying plainly. If you use Claude through the app, your plan did not change. Free is still free. Pro is 17 dollars a month if you pay for a year up front, which is 200 dollars, or 20 dollars if you pay monthly.

What you get from a cheaper model is indirect. The tools you use that are built on these models get cheaper to run, and some of that reaches you as a better free tier or a longer trial. It is slower than a price cut on your own bill, but it is real.

Where you can use it

Opus 5.5 is available on Anthropic's own platform, where the model name for developers is claude-opus-5-5, and through Amazon Web Services, Google Cloud and Microsoft Azure. If your company already buys cloud from one of those three, you probably do not need a new vendor relationship to try it.

Browse AI tools we have checked

Is it worth switching to

If you are on Fable 5.1 and paying 10 dollars per million input tokens for work that is mostly not at the edge of what a model can do, the arithmetic now argues for moving. If you are on Sonnet 5 and it is doing the job, the cheaper model is still the cheaper model and there is no prize for using a bigger one.

The honest test is boring and it is the only one worth running. Take a real task you already do, run it on both, and compare the bill and the result. Benchmarks are written by the people selling the model. Your own workload is not.

What a million tokens actually is

Token prices mean nothing until you can picture the unit. A million tokens is roughly 750,000 words of English. That is about nine full length novels, or a mid sized codebase read end to end.

So at 4 dollars for a million input tokens, sending Opus 5.5 the equivalent of nine novels to read costs less than a sandwich. The expensive half is output, at 20 dollars, because writing is where the model spends its effort. This is why a chatty prompt that asks for a short answer is cheap, and a short prompt that asks for a long document is not.

• Reading a 40 page contract and answering questions: pennies.
• Rewriting that contract in full: several times more, because every word comes out of the expensive side.
• An agent working for an hour across a codebase: the one that adds up, because it reads and writes in a loop.

How to cut your bill without changing model

1. Use prompt caching. Anthropic charges a reduced rate to read cached input, 20 cents per million on Opus 5.5 against 4 dollars fresh. If you send the same long instructions or document on every call, caching is the single biggest saving available to you.
2. Batch what is not urgent. Anthropic's pricing page offers 50 percent off batch processing. If a job can wait, it should.
3. Ask for less output. Set a length and mean it. Most prompts leave the expensive side of the bill entirely to the model's discretion.
4. Stop sending the whole history. A thread that grows forever is re-read forever.

Common questions

Is Opus 5.5 better than Fable 5.1? Anthropic says it performs at the level of Fable 5.1 on most work, which is not the same as better everywhere. For the hardest end of the work, Fable 5.1 is still the more expensive model for a reason.

Did my Claude subscription get more expensive? No. Plan prices are separate from API prices and did not move with this release.

Do I have to change anything to use it? On the API, the model name is claude-opus-5-5. In the app, model choice depends on your plan and Anthropic rolls new models out over days rather than at once.

Will there be an Opus 6? Anthropic called this the first model in the Claude 5.5 family, which tells you there will be more in this family before there is a new number. Beyond that, anyone who tells you a release date is guessing.

Sources

The release date, the 40 percent claim, the three examples and the safety notes come from Anthropic's announcement of Claude Opus 5.5. The prices come from Anthropic's pricing page. Both were read on 24 September 2026, and prices on that page change, so check it before you plan a budget around it.