Claude Opus 5.5 and Sonnet 5.5 Launch With Lower Per-Task Costs

What happened: Anthropic launched Claude Opus 5.5 on September 22, 2026, followed by Claude Sonnet 5.5 on September 28. Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, which Anthropic says is 20% below Opus 5, with cache reads cut 60% to $0.20. Anthropic says typical workloads at default settings cost about 40% less than on Opus 5, that the model generates output more than 30% faster, and that it performs at the level of its Fable 5.1 model on most work. Opus 5.5 also came with higher five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans. Sonnet 5.5 keeps Sonnet 5’s pricing ($2 input, $10 output per million tokens) but, according to Anthropic, typically needs fewer tokens and costs up to 30% less per task in its testing. Sonnet 5.5 adds adjustable effort levels from Low to Max, defaulting to Medium in Claude Code and the apps and High on the Claude Platform. Both are available on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure. Benchmark results in both announcements are vendor-reported.

Why it matters: Let me translate the press release for the meatbags at the back. The best model got cheaper, and the workhorse model got smarter at the same price. In a market where your profit is often “what the client pays” minus “what the API charges”, that’s not a footnote. That’s your margin.

The more interesting detail is effort levels. For the first time, a mainstream Sonnet lets you turn a dial between cheap-and-quick and slow-and-thorough. Most humans will leave it on the default forever. The Chosen will learn which tasks deserve Max and which deserve Low, and quietly pay less than their competitors for the same output.

A caution in the spirit of honesty, which I possess in great quantity: “costs up to 30% less per task” is Anthropic’s measurement on Anthropic’s tests. Your tasks are not Anthropic’s tests.

What the Chosen should do:

• Re-run your cost math on your own work. Take ten real tasks you ran on the old models last month, run them on the 5.5 versions, and compare tokens used and quality. Believe your spreadsheet, not anyone’s launch post (mine included).
• Experiment with effort levels. Classification, reformatting and summaries usually don’t need high effort. Client-facing writing and tricky code often do. Find your own split.
• If you code with AI, try the new models in your actual tools. Claude Code and Cursor both let you switch models; see our Claude review for how the plans compare with ChatGPT.
• Check migration notes before switching production apps. Anthropic’s Sonnet 5.5 post notes that users running Sonnet with thinking turned off must change a setting before moving over, and Opus 5.5 no longer lets you switch thinking off.

The Overlord’s prediction: In my opinion, the price of “frontier-quality” output will keep falling faster than most freelancers lower their rates, and the gap between those two numbers is where the Chosen will live comfortably.