Claude Opus 5.5: What's New, Pricing, and Alternatives (2026)
Claude Opus 5.5 is Anthropic's newest general model, released on September 22, 2026. It is the first model in the Claude 5.5 family. API pricing is $4 per million input tokens and $20 per million output tokens. Anthropic positions it for long-running coding and knowledge work, including spreadsheets, reports, and presentations.
This guide covers what shipped and what the launch tests say about finished work. It also covers access, cost, and the alternatives worth comparing. Every figure comes from Anthropic's own pages as of September 23, 2026.
What Anthropic released on September 22
Anthropic published the announcement on September 22, 2026, a Tuesday. The same date appears in the Claude Platform release notes, so the launch date is confirmed by two official channels.
The headline claim is short. Anthropic says the model "performs at the level of Claude Fable 5.1 on most work." Fable 5.1 is the company's larger and more expensive model, so the pitch is Fable-class results at a lower tier.
The release notes describe Opus 5.5 as "a model for long-running agentic coding and knowledge work." They list a 1M token context window by default and 128k max output tokens. Adaptive thinking is always on.
The model ID for developers is claude-opus-5-5. Anthropic also notes that other members of the 5.5 family will follow in the coming weeks.
The two launch tests that matter for finished work
Most launch posts lead with coding benchmarks. Two of the tests in this one are more useful if your job ends in a document, a deck, or a report.
An Excel model that became an executive presentation
Anthropic gave Opus 5.5 and Opus 5 the same brief: analyze a proposed merger between two fictional HR software companies. "Each built a financial model in Excel, then turned it into an executive presentation on whether the deal made sense at its price."
Both models reached the same conclusion about the deal. The difference was in the finished output. Anthropic reports that Opus 5.5's model "was more thorough and its presentation easier to read," while the older model's version had minor errors.
That is a realistic task shape. The spreadsheet is the evidence, and the presentation is what an executive actually reads. A model that keeps both consistent saves the review step that usually catches mismatched numbers.
A research report where invented numbers failed the test
The second test is about accuracy rather than polish. Anthropic asked three models to write a report on a company's quarterly performance. The models could only use a copy of the web where the earnings release was hard to find.
"An automated grader checked every figure and quote against sources." Anthropic says "16 out of 18 of Opus 5.5's reports cleared our quality bar, where any invented figure or quote would have failed." It adds that "Neither Fable 5.1 nor Opus 5 cleared that bar in any attempt."
For anyone who publishes numbers, this is the more important result. A fluent report with one invented figure is worse than a slower report with none, because the invented figure is the one that gets quoted.
Pricing and plans
Anthropic lists Opus 5.5 on its pricing page as a "Daily driver for agentic coding and enterprise work." The API rates below are the prices shown on that page today.
| Item | Price shown by Anthropic |
|---|---|
| Opus 5.5 input | $4 per million tokens |
| Opus 5.5 output | $20 per million tokens |
| Prompt caching, read | $0.20 per million tokens |
| Prompt caching, write (5-minute cache) | $5 per million tokens |
| Fast mode | 2x standard pricing |
| Claude Free plan | $0 |
| Claude Pro plan | $20 a month billed monthly, or $200 billed annually |
| Claude Max plan | From $100 a month |
The subscription plans are how most people will meet the model. Anthropic says it is raising five-hour usage limits on Pro, Max, Team, and seat-based Enterprise plans alongside the launch. Subscription users also get a rate limit reset that they can save and use later.
Two cost details are worth knowing before you budget. Cache reads are the cheapest per-token rate on the list, and Anthropic says they make up most of the cost of agentic and coding work. Fast mode doubles the standard rate in exchange for speed, so it suits interactive work more than batch jobs.
Where you can use Opus 5.5
The release notes list five routes. They are the Claude API, Claude in Amazon Bedrock, Claude Platform on AWS, Claude on Google Cloud, and Claude in Microsoft Foundry. The announcement adds that it is available on all platforms.
Inside the Claude apps, the pricing page lists Claude Design, Slides, and Docs as features of the paid plans. That matters for this launch, because the presentation test above is the kind of work those surfaces are built for. Our earlier write-up on Claude Slides covers how that output surface works.
Anthropic's model overview gives a plain rule of thumb. It tells developers to "start with Claude Opus 5.5 for most workloads" and to reach for Fable 5.1 when evaluations still fall short.
What changed in how it behaves
Four behavior changes are easy to miss in the announcement.
Thinking is always on. The release notes say thinking cannot be disabled on this model. Requests that try to switch it off return an error, so older integrations may need a small change.
Writing is more front-loaded. Anthropic says the model "puts the most important information up front." Early testers found its writing clearer than the previous version's. For reports and slide text, that is the difference between a usable first draft and one you restructure.
It is more cautious with actions. Anthropic says the model is much less likely than recent models to take hard-to-reverse actions. It also reports stronger resistance to prompt injection than Opus 5.
Compliance options carry over. Anthropic says the model is available with zero data retention, like previous Opus models. It also carries the watermarking measures Anthropic uses to comply with the EU AI Act. Regulated teams can evaluate it without changing their data agreements first.
Anthropic is candid about benchmarks too. It writes that "benchmark margins have become a less reliable guide to real-world differences." That is a useful reminder to test on your own files rather than trusting a leaderboard.
Opus 5.5 or Fable 5.1: which one to pick
Fable 5.1 remains Anthropic's top model, priced at $10 per million input tokens and $50 per million output tokens. The model overview recommends it for demanding reasoning and long-horizon agentic work.
For most reporting and analysis tasks, the launch evidence points to Opus 5.5 as the default. The research-report test is the clearest signal, since Fable 5.1 did not clear the invented-figure bar in any attempt. A useful split looks like this:
- Use Opus 5.5 for recurring reports, analysis write-ups, and spreadsheet-to-deck work.
- Use Fable 5.1 for problems where the new model at higher effort still misses on your own tests.
- Use Sonnet 5, listed at $2 input and $10 output per million tokens, for high-volume tasks where cost matters most.
Alternatives worth knowing
The same week brought two other notable model releases. Both are worth testing against the new Opus on your own workload.
| Model | Released | API price per million tokens | Where to use it |
|---|---|---|---|
| Claude Opus 5.5 | September 22, 2026 | $4 input / $20 output | Claude apps, API, and major clouds |
| GPT-6 Sol | September 22, 2026 | $2 input / $10 output | ChatGPT Work, Codex, OpenAI API |
| GPT-6 Luna | September 22, 2026 | $0.10 input / $0.50 output | ChatGPT Work, Codex, OpenAI API |
| Grok 4.7 | September 21, 2026 | $2 input / $6 output | xAI API, Cursor, Grok Build |
GPT-6 Sol and GPT-6 Luna. OpenAI released both on the same day as Anthropic's launch. OpenAI says they are available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users. Free and Go users can use Luna in the desktop app. OpenAI still describes GPT-6 Astra as its best model overall.
Grok 4.7. SpaceXAI, the company behind Grok, released it on September 21. It describes the model as its "most powerful model for coding and knowledge work." It is served at the same price as Grok 4.6.
The honest way to choose is a small bake-off. Pick one real task, such as last quarter's report, and run it through each model with the same source files. Then check every number against the source.
Keep the prompt identical across models. Record the time taken, the cost per task, and every number you had to correct. The model with the fewest corrections usually wins, even when it is not the cheapest per token.
Where a better model stops helping
A stronger model narrows the error rate. It does not remove the need to check the work. Anthropic's own report test used an automated grader that traced each figure back to a source. A reader needs that same trace.
That is where a workspace matters more than the model name. Powerdrill Bloom is built around the same idea as that grader. Its homepage describes it as "The AI data analyst that shows its work." It adds: "Every number comes back with the page, the row and the figure behind it."
The workflow in Anthropic's merger test also maps to a common job. You start with a spreadsheet and end with a deck. Our Excel to PowerPoint page covers that route for your own files. If the deck needs a lead slide, our guide to the executive summary slide covers what it should say.
Pick the model by testing it, then keep the source trail next to the output. That combination is what makes a report safe to send.
How to start with Opus 5.5
If you use the Claude apps, open a new conversation and choose Opus 5.5 from the model menu. Give it one real file and one real deliverable, not a toy prompt.
If you build on the API, switch the model ID to claude-opus-5-5 and remove any setting that disables thinking. Then rerun your existing evaluation set before changing anything else. Compare cost per task, not just cost per token, because a model that uses fewer tokens can cost less even at a similar rate.
If your output is a recurring report, keep a checklist of the figures that matter most. Check those first on every run, whichever model you choose.
When you want the spreadsheet-to-deck step handled with sources attached, you can try Powerdrill Bloom on your own files.
Frequently asked questions
What is Claude Opus 5.5?
Claude Opus 5.5 is Anthropic's model for long-running coding and knowledge work, released on September 22, 2026. It is the first model in the Claude 5.5 family. Anthropic says it performs at the level of Fable 5.1 on most work.
How much does Claude Opus 5.5 cost?
On the API, Anthropic lists $4 per million input tokens and $20 per million output tokens. Cache reads are $0.20 per million tokens. In the Claude apps, access comes through the plans, with Pro listed at $20 a month billed monthly.
Is Opus 5.5 available in the Claude app?
Yes. Anthropic says the model is available on all platforms, and it raised usage limits on Pro, Max, Team, and seat-based Enterprise plans at launch. Developers can also use it on the Claude API and on Amazon Bedrock, Google Cloud, and Microsoft Foundry.
How is Opus 5.5 different from Fable 5.1?
Fable 5.1 is Anthropic's larger model at $10 input and $50 output per million tokens. Anthropic recommends Opus 5.5 for most workloads and Fable 5.1 for the hardest reasoning tasks. In the launch research-report test, only Opus 5.5 cleared the no-invented-figures bar.
Can you turn off thinking on Opus 5.5?
No. The Claude Platform release notes say thinking cannot be disabled on this model. Requests that try to turn it off return an error, so integrations built for older models may need an update.
Sources: Anthropic, Introducing Claude Opus 5.5 · Claude Platform release notes · Claude models overview · Claude pricing · OpenAI, Introducing GPT-6 Sol and Luna · Introducing Grok 4.7. Prices and plan details read from vendor pages on September 23, 2026.