If you use ChatGPT or Claude, you’ve probably noticed these companies keep releasing different models, and they’re not all supposed to do the same thing. Some are built to be extremely powerful but expensive, while others are designed to be faster and cheaper. Anthropic released the latest version of its middle model on Monday, called Sonnet 5.5, and it’s getting surprisingly close to the expensive one.
Anthropic says Sonnet 5.5 is more than 30% faster and can cost up to 30% less to complete the same task. Importantly, it hasn’t actually lowered the price. The model has simply become more efficient, getting the same work done using fewer tokens. One investment firm tested it across 2,441 finance tasks and found it performed better than the previous Sonnet while using roughly a quarter of the tokens per answer.
Closing the gap
The obvious question is whether making a model more efficient means sacrificing how good it is. Anthropic’s own testing suggests that gap is getting pretty small. On GDPval-AA, a test covering real-world work across 44 occupations and nine industries, Sonnet 5.5 scored 1,844 points compared with 1,846 for Opus 5.5, Anthropic’s flagship model.
Anthropic does stress that Opus remains stronger at complex, open-ended work that requires sustained judgment. But it also costs twice as much per token. Sonnet is instead designed for more clearly defined everyday work, including fixing bugs and creating documents, presentations and
Why efficiency matters
And that’s becoming a big part of where AI is going. Building smarter models is one thing, but for businesses actually using them, getting the same quality of work done faster and for less money can be just as important.
The more efficient these models become, the cheaper it becomes to use AI across thousands or even millions of tasks. Sonnet 5.5 is a pretty good example of that shift: its price hasn’t changed, but Anthropic is trying to make each dollar spent on it go further.
