Claude Opus 5.5: what EU businesses need to know

Most teams had just got used to the order of things: Opus for everyday work, Fable for the hardest jobs. Claude Opus 5.5 arrived on 22 September 2026 and, according to Anthropic, does most of what Fable 5.1 does at 40 per cent of its price per token.¹ The everyday car has caught up with the racing car on most roads, and the bill shrank on the way.

Claude Opus 5.5 in brief, checked on 9 November 2026

Below you find what Claude Opus 5.5 is, what it costs per task, how it runs for a European team and where it still falls short.

Update 9 November 2026: Fable 5.1 has a successor. Claude Fable 5.2 costs US$8 in and US$40 out per million tokens, double the price of Opus 5.5, and leads on the hardest technical work. Opus 5.5 remains the default for everyday work.

---

What is Claude Opus 5.5?

Claude Opus 5.5 is Anthropic's general model for work that takes time, such as long coding projects, computer use and knowledge work like reports and analysis.¹ You find it in the Claude apps, in Claude Code and in the Claude API as claude-opus-5-5.² Companies that buy AI through a cloud provider get the same model on Amazon Bedrock, Google Cloud and Microsoft Foundry. Sonnet 5.5 and Haiku 5.5, its smaller and cheaper siblings, followed on 28 September and 7 October, and the whole family is explained in what Claude AI is.³

The dial for how much the model thinks is still called effort, and it runs from low to max. Its default moved from high on Opus 5 to medium on Opus 5.5, so a request that sets no level now gets a little less thinking than before.² Thinking can no longer be switched off at all, which means a team that wants the old depth picks high by hand.²

Opus 5.5 is also quicker, writing its answers more than 30 per cent faster than Opus 5. Anthropic cites an early tester that migrated a codebase of 680,000 lines in less than a day, a best case from a launch page, so expect your own code base to take longer.¹

---

How does Claude Opus 5.5 compare with Fable 5.1 and Opus 5?

Every launch page carries a table where the new model wins, a bit like reading a CV written by its owner. Here is Anthropic's own table, with prices from the API documentation:¹²

Terminal-Bench 4.0 measures how well a model handles longer tasks in the terminal, the command line where developers work, and there Opus 5.5 leads Fable 5.1 by more than ten points.¹ OSWorld 2.1 tests whether a model can operate an ordinary computer with mouse and keyboard, and the two sit almost level at 81.8 against 80.7 per cent. A team that wants AI to click through its own systems therefore decides on price, because the quality gap is about one point.

Claude Opus 5.5 in independent tests

References matter for a CV, and Artificial Analysis, an independent testing company, ranks Opus 5.5 first on its combined Intelligence Index with 58 points at max effort.⁴ The figures shrink a little when someone else measures, because Artificial Analysis got 59.6 per cent on Terminal-Bench 4.0 against Anthropic's 66.4.⁴ Count on the independent number when you plan, since your tasks will look more like a stranger's test than like the vendor's best run.

Claude Opus 5.5 in European languages

On Anthropic's Global-MMLU test Opus 5.5 scores 99.1 per cent of its English result in German, 99.3 per cent in French and Spanish and 99.2 per cent in Italian.⁵ A sales team in Lyon or a finance team in Munich can therefore work in its own language with almost no loss. Dutch, Polish and the Nordic languages are missing from that table, so test ten real texts in those first.

---

What does Claude Opus 5.5 cost?

On the Claude API Opus 5.5 costs US$4 per million input tokens and US$20 per million output tokens, where tokens are the word pieces AI counts and charges for.² That sits 20 per cent below Opus 5 and 60 per cent below Fable 5.1, which costs US$10 and US$50, or about €9 and €44 at the ECB rate of 25 September 2026.

The biggest cut lands on cached text, meaning instructions and files the model has already read and can read again, which now costs US$0.20 per million tokens against US$0.50 on Opus 5.⁶ In long sessions that line makes the largest difference, and Anthropic says costs fall about 40 per cent against Opus 5 on typical work at the default settings.¹

The discount disappears on max effort, where Opus 5.5 writes about 119,000 tokens per test task at Artificial Analysis against 73,000 for Opus 5.⁴ Max is like asking a consultant to be really thorough: the work gets careful and the invoice shows it.

On Pro, Max and Team the price cut shows up as time instead of money, because usage limits now last about 25 per cent longer than on Opus 5.⁶ Seat prices stay the same, with Pro at €18 a month and a Team Standard seat at €18 a month on annual billing, and every plan sits in our guide to Claude prices in 2026.⁷ Through satori-claude we run the seats for €490 once and €14 per user a month plus the €22 Standard licence, all excluding VAT and on one invoice in euros.

---

When should a European team pick Claude Opus 5.5 over Fable 5.1?

Almost always, and the reasons fit in two numbers. Pick Opus 5.5 on medium as the default for the whole team, because it costs US$4 in and US$20 out per million tokens and handles most of what needed Fable last summer.²

Move up to high when medium gets stuck on the same task, which Anthropic's own pricing guide estimates adds about 20,000 thinking tokens, roughly US$0.40 per task.⁶ That pays for itself the moment it saves one retry. Fable 5.1 belongs on work where the result matters more than the token price, such as long runs nobody supervises or code with no pattern to follow.⁶

The choice need not be a guess, and a Claude consultant can map your team's most common tasks to the right model so nobody pays Fable prices to write an email. You can do a good part of it alone, though: write down the five tasks your team spends most time on and put Opus, Sonnet or Haiku next to each one with the checklist below.

Claude model checklist: which model for which job

• Emails, minutes and reports: Opus 5.5 on medium, or the cheaper Sonnet 5.5.
• Medium gets stuck on the same task twice: Opus 5.5 on high.
• Hours without supervision, or a code base with no pattern to follow: Fable 5.1.
• Searches, logs and lookups: Sonnet or Haiku, as Anthropic's own pricing guide suggests.⁶

---

Claude Opus 5.5, the GDPR and where the data goes

Opus 5.5 supports zero data retention, meaning the vendor keeps no prompts after processing, which Anthropic switches on per organisation.¹ Fable 5.1 differs here, because Anthropic keeps its prompts and completions for at least 30 days by default, so work with personal data is safer on Opus 5.5.⁸ The GDPR still depends on your contract and on what you send, so check the data processing agreement before customer files go through the model. Our guide to Claude and the GDPR walks through that check.

Anthropic stores data in the US, and its own apps and API have no EU setting.⁹ Teams that need EU processing call Opus 5.5 through Amazon Bedrock's EU cross-region inference, which keeps each request inside six EU regions.¹⁰ The other route is the EU multi-region endpoint on Google Cloud.¹¹ No 5.5 model runs inside one single EU region such as Frankfurt, and the full picture sits in Claude EU data residency.

Opus 5.5 runs safety classifiers for cybersecurity and, new on this model, for biology.² A blocked request comes back as a normal answer with the stop reason refusal, and your code can retry it on another model through Anthropic's fallback setting.² An IT team asking about firewall rules may therefore hit a refusal now and then, so test those prompts before you build them into a tool.

---

What changes in the Claude API with Claude Opus 5.5?

Teams with their own integrations face four changes that can break code which runs on Opus 5 today:²

• Thinking cannot be switched off, and a request that tries returns a 400 error.
• Forced tool use, a tool_choice of any or a named tool, returns a 400 error.
• Thinking blocks are tied to the model and to the conversation that produced them.
• The older computer use tool, computer_20251124, is rejected on the Claude API, while Amazon Bedrock still accepts it.

Search your code for thinking, tool_choice and computer_20251124 before you change the model name to claude-opus-5-5, so the errors show up in your test environment and not at a customer. The text Opus 5.5 writes between tool calls also arrives in thinking blocks now, so an app that streams progress notes goes quiet until you change one display setting.²

---

Summary: Claude Opus 5.5 is the new default

Opus 5.5 is the sensible default model for a European team today, because it scores level with Fable 5.1 on most work and costs 60 per cent less per token. Fable 5.1 still earns its price on long unsupervised runs, while the seat prices in euros sit in how much Claude costs and the coding side in our Claude Code guide.

Paste last week's report into Claude with Opus 5.5 selected and add one line: "Rewrite this so a manager gets the main point in the first sentence, keep every figure and mark anything you are unsure of." Ten minutes later you have two versions side by side, and that comparison tells you more about Opus 5.5 than every table above.

Sources

Footnotes

1. Anthropic (2026). Introducing Claude Opus 5.5, 22 September 2026. https://www.anthropic.com/claude-opus-5-5 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

2. Anthropic (2026). What's new in Claude Opus 5.5. Claude Platform Docs, checked on 9 November 2026. https://platform.claude.com/docs/en/models/opus-5-5/whats-new-opus-5-5 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰

3. Anthropic (2026). Models overview. Claude Platform Docs, checked on 9 November 2026. https://platform.claude.com/docs/en/models/overview ↩

4. Artificial Analysis (2026). Claude Opus 5.5 takes the top spot on the Artificial Analysis Intelligence Index. https://artificialanalysis.ai/articles/claude-opus-5-5 ↩ ↩² ↩³

5. Anthropic (2026). Multilingual support. Claude Platform Docs, checked on 9 November 2026. https://platform.claude.com/docs/en/build-with-claude/multilingual-support ↩

6. Anthropic (2026). What a task costs on Opus 5.5. https://claude.dev/blog/what-a-task-costs-on-opus-5-5/ ↩ ↩² ↩³ ↩⁴ ↩⁵

7. Anthropic (2026). Plans and Pricing, checked from an EU connection on 9 November 2026. https://claude.com/pricing ↩

8. Anthropic (2026). Covered Models. Claude Help Center. https://support.claude.com/en/articles/15425695 ↩

9. Anthropic (2026). Data residency. Claude Platform Docs. https://platform.claude.com/docs/en/manage-claude/data-residency ↩

10. Amazon Web Services (2026). Geographic cross-Region inference. Amazon Bedrock User Guide. https://docs.aws.amazon.com/bedrock/latest/userguide/geographic-cross-region-inference.html ↩

11. Google Cloud (2026). Data residency. Gemini Enterprise Agent Platform documentation. https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/data-residency ↩