Claude Sonnet 5.5: 30 percent faster at the same price

What's new

On September 28, 2026, Anthropic released a new version of its most widely used AI model: Claude Sonnet 5.5. It replaces Sonnet 5 and is built for everyday tasks such as coding, writing and documents in office programs. According to Anthropic, it is the fastest Sonnet model the company has ever made, and it is available now in Claude's own products as well as via the API from Anthropic and the cloud providers AWS, Google Cloud and Microsoft Azure.

What's clever

The clever part is that Sonnet 5.5 solves tasks with fewer "tokens," the chunks of text the AI model thinks and writes in. That makes it more than 30 percent faster than its predecessor, and because it needs fewer tokens and tool calls per task, the bill often ends up lower, even though the price per token is unchanged. The model has also gained new security features against cyberattacks on par with the much larger Opus 5.5 model, and it is the first Sonnet version subject to the same strict safety requirements as Anthropic's large models.

Cheaper or better?

The price in the API, meaning what developers pay to use the model, is unchanged: $2 per million input tokens and $10 per million output tokens, plus 20 cents for tokens re-read from the cache and $2.50 for writing to the cache. Because the model is more efficient, the total bill for a typical task can, however, end up as much as 30 percent lower. On the Terminal-Bench 4.0 benchmark, a test of coding skills, the score jumps from 10.3 percent for the old model to 70.6 percent for Sonnet 5.5. On the OSWorld 2.1 test, which measures the ability to use a computer autonomously, it goes from 57.0 to 80.1 percent. Compared with the more expensive Opus 5.5 model, Sonnet 5.5 scores slightly lower on independent tests, but actually wins on some coding tasks because it can run more AI agents at once without blowing the budget.

What it's good at

Sonnet 5.5 is intended as a versatile workhorse for everyday use: fixing bugs in code, writing and summarizing texts, and creating spreadsheets and other office documents. Anthropic highlights that the model has become better at keeping track of long, compound tasks without losing the thread along the way. For an ordinary business, this means in practice that you can get more AI help with more tasks for the same or less money than before.