Claude Opus 5.5 Ships: Cheaper, Faster, Sharper
Anthropic shipped Claude Opus 5.5 on September 22, 20% cheaper per token and over 30% faster, with typical tasks costing 40% less to run.
Anthropic released Claude Opus 5.5 on September 22, its new top-tier model. Input and output tokens both cost 20% less than Opus 5, output generates over 30% faster, and Anthropic says a typical task now costs 40% less to run end to end.
What shipped
Anthropic shipped Claude Opus 5.5 on September 22, 2026, the new flagship in a fresh Claude 5.5 line. Two numbers matter, and they are not the same measurement.
The plain price cut: input tokens are $4 per million, down from $5. Output tokens are $20 per million, down from $25, both a flat 20% cheaper, list price against list price. Cache reads, most of the cost on coding and agent work, drop 60%, from $0.50 to $0.20 per million tokens.
The bigger number, 40% less to run, is a separate measure: a typical task on Opus 5.5 costs 40% less end to end than on Opus 5, because the model also finishes using fewer tokens. That is the price cut plus a token-efficiency gain, not a second discount on the sticker price. A fast mode runs up to 2.5x faster at $8 input and $40 output per million tokens, and output overall generates more than 30% faster than Opus 5.
Anthropic reports GDPval-AA v2.1, a test of real-world work across 44 occupations, moving to 1846 Elo from 1708. Opus 5.5 is live now on Claude, the API, Claude Code, AWS and Google Cloud. Sonnet 5.5 and Haiku 5.5 follow in the coming weeks.
Why an owner should care
Most owners paying for Claude are not tracking benchmark scores. They are tracking a bill and a wait, and this moves both: a lower price per token, fewer tokens per task, faster answers. If Opus 5.5 was already doing your heaviest lifting, it now costs less to run the same jobs, no new plan required.
What it looks like inside a shop
An AV or smart home shop running proposals, service notes, or client-portal code through Claude gets the same output at a lower bill by picking the model, nothing else. A shop already using Cowork inside Claude for scopes and quotes runs that same workflow on a cheaper, faster engine this week.
Any service business sees the same saving: an electrician running estimates, a design firm drafting proposals, a bookkeeper reconciling notes. Same task, same accuracy bar, lower cost, shorter wait.
What to do this week
- Confirm which Claude model your account or integration defaults to, and switch to Opus 5.5 if you are still on Opus 5.
- Re-check your monthly Claude spend against the new $4 input / $20 output pricing before assuming last month's bill still applies.
- Test the fast mode against your normal setting for agent or coding workloads, to see if $8/$40 per million tokens is worth the speed.
- If you route proposals or scopes through Claude for AV or trades work, see how the same drafting fits an AI proposal engine built for AV integrators.
Questions owners ask
Is Opus 5.5 actually 40% cheaper, or 20% cheaper?
Both are real, but they measure different things: list price per token is 20% lower, and the 40% figure is Anthropic's estimate of a typical task's full cost once token efficiency is added in.
Do I need to change anything to get the lower price?
No. If your account already points at Opus 5.5, the new pricing applies automatically. If it still points at Opus 5, you switch the model yourself.
Is this worth switching for a small operation?
If you run proposals or client-facing drafts through Claude today, yes. Anthropic's own benchmarks show equal or better quality at lower cost, no new account needed. Source: Introducing Claude Opus 5.5, Anthropic, September 22, 2026
