The Operator's AI Brief
Cheaper Per Task Is Not Cheaper Per Month
Claude Sonnet 5.5 costs up to 30% less per task, but an outside benchmarker found it writes more per task than any model it has measured.
Listen to this brief (3 min)
The audio is AI-produced: an AI clone of Jon's voice and an AI co-host, reading a script built from the sources below.
Anthropic released Claude Sonnet 5.5 on September 28. An outside benchmarker then flagged a detail that decides whether the savings reach your bill.
Faster, and cheaper per task
According to Anthropic’s announcement, as reported by 9to5Mac, Sonnet 5.5 runs more than 30% faster and costs up to 30% less per task. The list price matches Sonnet 5. Anthropic also said Haiku 5.5, a smaller model, would arrive in the coming weeks.
A supply yard or an HVAC shop should care because a model like this sits underneath tools you already pay for: the phone agent that books a service call, the scheduling assistant, the text that follows up on an estimate. A faster model keeps callers from waiting. A cheaper one lowers the builder’s cost, which trims your bill or covers more calls for the same money.
The caveat from Artificial Analysis
Artificial Analysis, an outside benchmarking group, scored Sonnet 5.5 at 56 on its Intelligence Index, two points behind Opus 5.5 at its top setting. The group also reported that Sonnet 5.5 produced the most output tokens per task of any model it has measured.
A model bills by how much it writes. When it writes more to finish the same job, the per-task discount shrinks. A setup that runs the top effort level on every small job can cost more at month end than the launch page suggests. In the systems I build, I pick the model and effort level for each job. A phone number lookup gets the small setting. A tricky customer question gets the larger one.
Cheaper models give builders room to cut your cost. Whether they do shows up on the invoice.