Claude Opus 5.5 20% Price Cut Comparison

Engineer holding a tablet and inspecting an open server rack in a server hall

If you are budgeting for API use, Claude Opus 5.5's September 22, 2026 launch rates are US$4 per million input tokens and US$20 per million output tokens. Both are 20% below Opus 5. Anthropic's estimated 40% task savings depend on usage, rather than applying to every bill.

 

What lower rates mean for a task

Tokens are the units used to process content: input is sent to the model, and output is generated. With 1 million input tokens and 1 million output tokens, you would pay US$24 instead of US$30 at standard rates. That saves US$6, excluding caching and separate tool costs.

Cache reads, which reuse stored inputs, cost US$0.20 per million tokens, down from US$0.50. The 60% reduction covers reads only; creating the cache costs extra.

A developer checks a tablet at a test bench lined with connected phones.

 

Compare Claude Opus 5.5 under the stated conditions

In the original article on Injoys, you will find pricing tables, benchmark conditions and a migration checklist. Its useful distinction is between cheaper tokens and cheaper completed work: the Terminal-Bench 4.0 score rose from 52.3% to 66.4% using xhigh, while the task-cost estimate uses default settings.

Read the full article on Injoys

 

A practical check for repeated coding work

If you reuse instructions for coding tasks, separate the cache check from the completion check. Confirm that stored inputs were reused, then include failed attempts and retries when calculating the cost of work that passes review. A lower read charge alone cannot establish that the whole workflow costs less.

Keep human review time separate. That makes it easier to see whether a cheaper run also leaves less work to correct.

 

Who needs to check settings first

These rates concern token-billed API use; they do not establish lower monthly subscription fees. If you maintain an existing integration, check more than the model name: settings that disable thinking are rejected, and thinking tokens count toward output charges.

The Claude API identifier is claude-opus-5-5. AWS, Google Cloud and Microsoft Azure support was announced at launch, but identifiers and tool support may differ by platform.

Start by checking existing requests against Anthropic's migration guide, including thinking settings, forced tool calls and response block handling.

Read the full article on Injoys

This post was written with assistance and published on a channel run by Injoys (injoys.com). The original article and its sources are at the link above.

Comments

Popular posts from this blog

Bitcoin Rally Claims Reviewed: U.S. Liquidity, Stablecoins, Reserves and Mining

Medicare Coverage 2027: When to Change Plans

Daedong Law 1651: Chungcheong Reform and Kim Yuk's Role