← Back to blog

Guides

Claude Opus 5.5 is 40% cheaper. Will your company see it?

Harriet Team · · 4 min read

On September 22, 2026, Anthropic released Claude Opus 5.5 and said it costs about 40% less to run than Opus 5 on typical workloads. That is a real cut in the price of the best model most companies use. Whether it shows up on your company’s bill is a separate question, and the answer depends on how you pay.

What Anthropic changed

The numbers below are from Anthropic’s announcement and its pricing pages on September 25, 2026. They will move again, so check the source before you budget on them.

  • Opus 5.5 input tokens are $4 per million and output tokens $20 per million, 20% less than Opus 5.
  • Cache reads are $0.20 per million tokens, 60% less than Opus 5. Anthropic says cache reads make up the majority of agentic and coding costs, which is why the combined saving is larger than either cut on its own.
  • Anthropic estimates a 40% lower running cost at default settings on typical workloads, so your own mix of tasks and settings will land somewhere around that, not exactly on it.
  • Five-hour usage limits went up on Pro, Max, Team and seat-based Enterprise plans.
  • Sonnet 5.5 and Haiku 5.5 are due in the coming weeks with similar efficiency gains.

Update, September 30, 2026. Sonnet 5.5 arrived on September 28, and who sees that saving follows the same logic.

Anthropic’s own wording on the Claude Code blog is the important part. The saving applies “for usage billed by the token”. Not every plan is.

On Claude Team, the bill does not move

A Claude Team seat is a flat fee, $20 a month on annual billing for a standard seat, with usage bundled in. The seat price did not change on September 22.

What a Team customer gets is more room inside the same fee, because the five-hour limits went up. That is worth having. It is not a saving, and it is not something finance will see on an invoice. If your company has been on Team for a year, every model price cut in that year has arrived as headroom, never as a lower bill.

That is the trade a flat seat makes. Predictable, and immune to good news.

On Claude Enterprise, the usage line falls

Claude Enterprise is priced differently. The seat fee covers access only, and every token your people use is billed at API rates on top. Anthropic’s help pages are explicit that there is no included allowance.

So an Enterprise customer does see this cut, on the usage line, for whatever share of their work runs on Opus 5.5. The seat fee stays where it was. The saving is largest for long, context-heavy sessions, which is where cached tokens dominate, and smallest for quick chat. If most of your usage is Sonnet or Haiku already, wait for the 5.5 versions of those before expecting much.

The catch is the same one that makes Enterprise expensive in the first place. Usage at API rates with no cap is what lets the price cut through, and it is also what makes the bill hard to predict. Anthropic lets admins set spend limits per organization and per user, and on this plan you should.

The shape of bill that always benefits

There is a pattern here that outlasts any single model release. Frontier models get cheaper every few months. A bill built on seat fees never falls when that happens. A bill built on usage does, automatically, the day the price changes.

Three things make sure the next cut reaches you too.

  1. Pay for usage, not for seats. Seats are a fee for access. Usage is a fee for work. Only one of them tracks the price of the model.
  2. Run the models on your own API keys where you can. When Anthropic drops a price, your provider invoice drops with it, with nobody in between.
  3. Route everyday work to cheaper models. Most tasks in a company never needed Opus. If the frontier model is reserved for the people and tasks that earn its price, a 40% cut on it is welcome, and a smaller share of your bill than the headline suggests.

Harriet is built around the third shape. Every model runs behind one policy with a budget per team and your own keys if you want them. Before you move a team to the new model, eval mode runs the same task on Opus 5.5 and your current default side by side, with cost and quality next to each other, so the switch is a decision rather than a reflex.

If you are on Team and wondering whether Enterprise is the answer, Claude Team vs Claude Enterprise covers the moment teams outgrow the flat seat. If you already have a quote, compare it. And for the full picture of what routing and caps do to an AI bill, start with cost control.

Common questions

Does the Claude Opus 5.5 price cut lower a Claude Team bill?
No. A Claude Team seat is a flat monthly fee with usage bundled in, so the seat price stays the same. What changes is headroom. Anthropic raised the five-hour usage limits on Team, so each seat gets more work done for the same money. The invoice does not fall.
Does the price cut lower a Claude Enterprise bill?
Partly. Enterprise charges a seat fee for access plus usage at API rates. The seat fee is unchanged, and the usage line falls for work that runs on Opus 5.5, because Anthropic's per-token prices dropped. How much depends on how much of your usage is on Opus and how much of it is cached context.
What did Anthropic actually change on September 22, 2026?
Anthropic released Claude Opus 5.5 with input at $4 and output at $20 per million tokens, 20% below Opus 5, and cache reads at $0.20 per million, 60% below Opus 5. Anthropic estimates a 40% lower running cost on typical workloads at default settings. It also raised five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans. Sonnet 5.5 and Haiku 5.5 are due in the following weeks.
How does a company make sure model price cuts reach its AI bill?
Pay for usage rather than for seats, run the models on your own API keys where you can, and route everyday tasks to cheaper models so the frontier model is only paid for where it earns its price. A bill built on seat fees never falls when a model gets cheaper. A bill built on usage does.