Claude Opus 5.5 (2026): What It Costs, What Changed From Opus 5, and Whether Your Business Should Switch

Claude Opus 5.5: the short answer

The launch-week posts all say "40% cheaper." That is Anthropic's number, and it is true on the right workload, but it is not a 40% price cut. As of September 2026, Claude Opus 5.5 (announced September 22, 2026) costs $4 per million input tokens and $20 per million output on Anthropic's API, 20% under Opus 5's $5 and $25. The bigger cut is cache reads, down 60% to $0.20. Anthropic says it performs at the level of Claude Fable 5.1 on most work. It is on Pro, Max, Team, and Enterprise, and the launch came with higher five-hour usage limits. This page is the full price sheet, where the 40% actually comes from, what changed on your plan, and who should switch, with every number read off Anthropic's announcement, API pricing page, and plan table on September 28, 2026.

What Claude Opus 5.5 is

Opus 5.5 is the first model in what Anthropic calls the Claude 5.5 family. The announcement positions it as the new leading Opus: it "performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5," and generates output more than 30% faster than Opus 5. Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 will follow "in the coming weeks."

The benchmark that matters most for business work is the knowledge-work one. On GDPval-AA v2.1, real professional tasks across 44 occupations, Anthropic reports Opus 5.5 at 1846 Elo, ahead of Fable 5.1 at 1735 and Opus 5 at 1708. The example I would actually show a client: asked to build a merger model in Excel and turn it into an executive deck, Opus 5.5 finished in 63 minutes against 93 for Opus 5, at half the cost, and Anthropic says Opus 5's version had minor errors. Its API ID is claude-opus-5-5, and it is available on the Claude apps, the Claude Platform, AWS, Google Cloud, and Microsoft Azure. Where it now sits in the lineup by job is on best Claude model for business.

Claude Opus 5.5 pricing

From Anthropic's API pricing page, per million tokens, as of September 28, 2026:

Fast mode for Opus 5.5, up to 2.5x the speed in Claude Code and on the Claude Platform, is $8 input and $40 output. That is 2x the standard rate, and still below the $10 and $50 fast-mode rate for Opus 5.

Read the cache column twice. An Opus 5.5 cache read now costs exactly what a Sonnet 5 cache read costs, $0.20, and less than a Fable 5.1 read. For any agent that re-reads the same instructions, documents, or codebase on every call, the frontier model's input cost has collapsed to the mid-tier rate. How caching and batch change the rest of Anthropic's lineup is on Anthropic API pricing.

Where the 40% comes from: the math on a real workload

Anthropic's 40% has three parts: 20% off per-token prices, 60% off cache reads, and fewer tokens per task. Only the first two are on the price sheet, so here is what they do on their own. Same job I use for every model comparison: an agent that runs 1,000 times a month, re-reads a 50,000-token context each run, adds 2,000 tokens of new input, and writes 1,500 tokens. Before cache writes:

• Opus 5.5: 50M cached tokens at $0.20 is $10, plus 2M input at $4 is $8, plus 1.5M output at $20 is $30. About $48 a month.
• Opus 5: $25 cached, $10 input, $37.50 output. About $73.
• Fable 5.1: $12.50 cached, $20 input, $75 output. About $108.
• Sonnet 5: $10 cached, $4 input, $15 output. About $29.

That is 34% off Opus 5 from the price sheet alone, and most of it is the cache line. Now run the same job with no caching: 52M input tokens at $4 plus the output is about $238 on Opus 5.5 against about $298 on Opus 5. That is exactly 20%. So the honest version of the headline: if your workload caches, you get most of the 40% from prices alone; if it does not, you get 20% from prices and have to earn the rest from the fewer-tokens claim, which only your own bill can confirm. Anthropic's testers report that part is real (Box says a third of the tokens on its evaluations), but measure it on your job before you budget on it.

What changed on your Claude plan

Anthropic's plan table lists Opus on Pro, Max, Team, and Enterprise, and not on Free. Pro is $20 a month ($17 billed annually), Max starts at $100, Team is $20 per seat billed annually or $25 monthly, and Enterprise is $20 per seat plus usage at API rates. The full plan-by-plan picture is on Claude AI pricing.

Two plan changes shipped with the model. First, Anthropic said it is "increasing five-hour usage limits on Pro, Max, Team, and seat-based Enterprise plans." Second, subscribers get a rate-limit reset they "can now save and use whenever you choose," which is new: until now, a limit hit meant waiting out the window. What Anthropic did not publish is a number. The pricing page still describes limits as a rolling five-hour window plus weekly caps with "no fixed message count," so the only place you will see the increase is Settings > Usage. How the windows work, and what to do when you hit one, is on Claude usage limits.

One more plan-table change worth knowing: on September 13, the Team column listed Fable as not included. As of today, it reads "50% of weekly limits" on premium seats. If your team has premium seats, the top model is now in reach without Enterprise.

Should your business switch to Opus 5.5?

• If you build on the API with Opus 5: yes, this week. It is cheaper per token, much cheaper on cached input, faster, and Anthropic reports better results. Change the model ID in a test environment, run your real task, compare the output and the bill.
• If you are on Fable 5.1 for cost-sensitive agent work: test Opus 5.5 first. Anthropic says the gap between them is narrower than the benchmarks suggest, and Opus 5.5's output is 40% of Fable's price.
• If your team uses claude.ai for daily writing and email: keep Sonnet as the default. Sonnet 5 is half the per-token price and lighter on limits, and Sonnet 5.5 is coming in weeks. Opus 5.5 is for the long jobs.

The mistake I watch teams make every launch is switching everyone's default to the new flagship and then running out of limits by Wednesday. Higher limits make that less likely this time. They do not make it free.

What to do today

Pick the longest task anyone on your team ran in Claude last week: the report, the spreadsheet model, the code change that took four follow-ups. Run it once on Opus 5.5 and count the prompts it takes, not just the quality of the answer. That number is the real saving, and it is the one Anthropic's 40% depends on. And if you want Claude set up properly across the company, with the right model on the right job instead of everyone guessing at the dropdown, that is the Claude for small business install, from $4,500 in about two weeks; The Shop ($397) is the do-it-yourself version.