Anthropic has released Claude Opus 5.5, the first model in a new series. The real selling point isn't performance, but price: it costs 40 percent less to run than its predecessor. This is immediately relevant for Mac developers.
Anthropic tiers its models by size and price: Haiku is the smallest, Sonnet is in the middle, Opus is the most powerful of the three, and above them are the Fable and Mythos series. Opus 5, then the flagship model, was released at the end of July, followed by a price reduction for Fable 5.1 at the beginning of September. Now, according to Anthropic's announcement, Opus 5.5 is coming, which the company claims performs at the level of Fable 5.1 in most tasks while costing significantly less.
Key Facts at a Glance
- Claude Opus 5.5 is now available, also via Amazon Web Services, Google Cloud and Microsoft Azure.
- Operating costs are 40 percent lower than with Opus 5 when using standard settings.
- Input and output tokens cost $4 and $20 per million, respectively; cache reads cost $0.20.
- The usage limits for the Pro, Max, Team and Enterprise subscriptions are increasing.
- Sonnet 5.5 and Haiku 5.5 are expected to follow in the coming weeks.
The price is the message
Anthropic explains the reduction by stating that the model requires less computing power than its predecessor. The effect is comprised of two parts: cheaper tokens and fewer tokens per task.
| Price per million tokens | Opus 5.5 | Opus 5 |
|---|---|---|
| Input | $4 | 5 dollars |
| Output | 20 dollars | $25 |
| Cache read operations | $0.20 | $0.50 |
| Cache write operations | 5 dollars | $6.25 |
The biggest jump is in cache read operations, at 60 percent. According to Anthropic, these account for the majority of costs in agent-based work and programming, precisely where the model is most widely used.
In addition, there is a faster mode in Claude Code and on the Claude platform, offering speeds up to 2.5 times faster. It costs $8 per million input tokens and $40 per million output tokens, exactly double the standard rate.
In addition to the price reduction, Anthropic is increasing the five-hour limits for subscriptions. Subscribers will also receive a reset of their limit, which can be canceled and redeemed at a time of their choosing.
What this means for developers on the Mac
The connection to Apple is direct. With Xcode 26.3, Apple integrated Anthropic's Claude Agent and OpenAI's Codex into its development environment, as described in the announcement at the time. Since the introduction of Xcode 27, along with new AI frameworks, Xcode has also provided its functionality more broadly via the Model Context Protocol.
The announced efficiency gains are aimed precisely at this way of working. Anthropic cites a tester who reviewed and corrected a codebase of 200,000 lines in under three hours, a task that took Opus 5 over 20 hours and 2.5 times the number of tokens. On Apple's side, this can be combined with tools that have been working together for some time: Claude Code tests iOS apps directly in the simulator.
Anthropic's announcement doesn't specify which model is used in the Xcode agent. Those who want to use Opus 5.5 should continue to access it via the Claude platform or Claude Code until further notice.
Anthropic puts its own figures into perspective
An unusual paragraph addresses the significance of the test results. In their own comparative tests, Opus 5.5 leads in agent-based programming, computer operation, and knowledge work. However, Anthropic notes that differences in comparative tests at this performance level are a less reliable indicator of real-world performance variations. In their own use, the difference between Opus 5.5 and Fable 5.1 is smaller than the test results suggest.
A second reservation concerns the safety testing. Anthropic describes indications that the model is frequently suspected of being under test. This makes it difficult to assess how it performs in actual use.
What is limited in practice
Opus 5.5 launches with security measures that Anthropic otherwise only uses in its top-of-the-line models. This is due to its capabilities in biology and cybersecurity, which the company claims rival those of Mythos 5.1.
In practical terms, this means that errors in your own code can still be found and fixed, while most other cybersecurity tasks are being handed over to the older Opus 4.8 model. Experts can apply for an audit program to gain full access; for biology, there is a corresponding program for certified organizations.
Another change affects all users: Think mode can no longer be turned off in Opus 5.5.
Watermark due to the EU AI Act
For the European market, one detail regarding availability is relevant. Opus 5.5 uses the same watermarking measures as Fable 5.1, which Anthropic explicitly implemented to comply with the EU AI Act. This allows generated texts to be marked as machine-generated.
Anyone working with the model and publishing results in Germany, Austria, or Switzerland should keep this in mind. As with Fable 5.1, it is also possible to use the model without saving data.
A downward price trend
What's remarkable about this release is the direction it takes. A new top-of-the-line model usually comes with a price increase, but here the price drops in every single category. Anthropic isn't selling more performance for the same price, but rather the same performance class at a lower price.
We expect this pattern to continue once the Sonnet 5.5 and Haiku 5.5 are released. Anthropic is announcing the same improvements in performance, efficiency, and security for the smaller models – and it is precisely in these cases that price will determine whether long-term use is worthwhile.
Do you use an AI model directly in Xcode, or do you stick with a separate window alongside it? Tell us in the comments how you handle it.
