Anthropic has released Claude Sonnet 5.5, positioning it as a faster, more economical model for well-scoped coding, document creation, computer use, and other agent-driven workflows.
Anthropic says the model generates output more than 30% faster than Sonnet 5 and can cost up to 30% less per task by completing work with fewer tokens. Token pricing remains unchanged at $2 per million input tokens and $10 per million output tokens.
The performance gains are substantial. Sonnet 5.5 scored 70.6% on Anthropic’s Terminal-Bench 4.0 evaluation, compared with 10.3% for Sonnet 5. On GDPval-AA, which measures professional work across multiple industries, it scored 1,844—just two points behind the more expensive Opus 5.5. These are vendor-reported results and should still be tested against each team’s own workloads.
Independent testing supports the value proposition. In its September 28 release evaluation, Vals placed Sonnet 5.5 second among 66 evaluated models, 0.47 percentage points behind Opus 5.5 while costing less than two-thirds as much per test. Benchmark versions and leaderboards can change, so the dated result is best treated as an external snapshot rather than a permanent ranking.
Why it matters
For builders, Sonnet 5.5 could become the practical default for high-volume agent workloads. It offers near-frontier performance without requiring Opus-level spending, making it particularly attractive for coding agents, support automation, document pipelines, and multi-step business workflows.
That creates a clearer routing strategy alongside Claude Opus 5.5: reserve Opus for ambiguous work that demands sustained judgment, while directing well-scoped implementation and production volume to Sonnet. The important metric is not token price alone, but the total cost of reaching an acceptable result.
The model is available through Anthropic’s API as claude-sonnet-5-5, as well as through AWS, Google Cloud, Microsoft Azure, and Claude’s applications.