Claude Sonnet 5.5 puts token efficiency beside response speed

A new Sonnet update

Anthropic introduced Claude Sonnet 5.5 on September 28, 2026. The announcement emphasized coding, document work, collaboration, and efficiency compared with the preceding Sonnet model.

The company described improvements in cost per task rather than relying only on a lower listed token price. That distinction is useful for enterprise evaluations.

Fewer tokens can mean several things

A model may reach a correct result with less deliberation, produce a shorter answer, or skip work that was actually needed. Only the first two are beneficial when the deliverable still satisfies the requirements. Usage savings need to be checked alongside completeness.

For a professional artifact, the review should examine the file itself, including calculations, citations, layout, and behavior in its intended application.

Compare accepted work

Run the same task set with the same permissions and a comparable reasoning budget. Count revisions and human interventions before a result is approved. Separate simple daily tasks from the difficult cases that justify a larger model. This gives the team a more realistic picture of efficiency than a single speed figure or a price table without workload context.