The release became the default for Free and Pro users, while adding a million-token context window in beta.
What happened
Anthropic retained Sonnet 4.5's starting API prices and described better instruction following and consistency. Its announcement includes computer-use evaluations and early-access feedback comparing the new model with earlier Claude versions.
Why it matters
Bringing stronger capabilities to a frequently used tier can matter more than raising a flagship benchmark. Computer use also opens workflows in applications that lack a convenient API.
The bigger picture
Our take. A successful benchmark run does not remove mistakes or malicious instructions from real websites. Measure the model on a complete workflow, including how it handles ambiguous screens and when it asks the user to intervene.
Source checked November 18, 2026. Company claims are attributed; paragraphs marked “Our take” are editorial interpretation.