The release aims to make capabilities associated with larger models practical for more frequent tasks.
What happened
Sonnet 5 launched across Claude plans and through the Claude API. Anthropic describes performance approaching Opus 4.8 on some tasks, with effort settings that adjust the cost-performance tradeoff.
Why it matters
Agent workloads can involve many model calls, so a smaller price per successful task can have a large effect. The comparison needs to include retries and human correction, not just token rates.
The bigger picture
Our take. Start with representative work at a modest effort setting, then increase effort where it improves outcomes. Provider benchmark results help identify candidates for that test.
Source checked November 18, 2026. Company claims are attributed; paragraphs marked “Our take” are editorial interpretation.