Teams running AI in software workflows know the bill has more than one line. Quality matters, but every retry, tool call and generated token affects both cost and waiting time. A model that finishes with fewer steps may improve the equation even when its API rate stays the same.
That is the case Anthropic makes for Claude Sonnet 5.5, introduced on September 28. The company says it generates output more than 30% faster than Sonnet 5 and costs up to 30% less per task in its tests. The token price itself has not changed: it remains $2 per million input tokens and $10 per million output tokens. The claimed savings largely come from needing fewer tokens to finish the work.
What the scores show
In Anthropic's published results, Sonnet 5.5 scored 70.6% on Terminal-Bench 4.0, compared with Sonnet 5's 10.3%. On CursorBench 4.0, it reached 55.5%, versus 34.1% for Sonnet 5 and 57.8% for Opus 5.5. These figures point to a substantial improvement in the coding tasks measured there. They are still vendor-published evaluations, and effort settings and test conditions differ across benchmarks. They cannot establish the same improvement for every production codebase.
Anthropic draws an important line itself: Opus 5.5 remains stronger on complex, open-ended work that requires sustained judgment. For engineering teams, the practical decision is therefore less about choosing the highest-scoring model for everything. Sonnet may suit recurring, bounded tasks, while Opus can be reserved for work that genuinely needs deeper investigation or review.
A per-token rate is only one part of the bill. The useful measure is the cost of completing a task to an acceptable standard.
Effort changes the economics
The effort setting matters too. Anthropic says Claude Code and its apps default to Medium, while Claude Platform defaults to High. Lower settings usually mean faster answers and fewer tokens; higher settings allow the model to work longer. Any fair comparison should therefore record effort, retries, tool calls and the acceptance standard for the output.
Sonnet 5.5 is available through Anthropic's platforms, AWS, Google Cloud and Microsoft Azure, under claude-sonnet-5-5 on Claude Platform. Anthropic also says some higher-risk cybersecurity requests can fall back to Sonnet 5 while it develops verified access to more advanced capabilities. That condition matters when comparing behavior across categories of work.
Our reading at MnzAI Labs is that this release deserves a controlled test against a team's actual work: choose bugs, reviews or documents with checkable outcomes, then compare total elapsed time, tokens, rework and final quality. If a faster first answer needs more corrections, the claimed savings may vanish. If it meets the same bar in fewer iterations, the model switch has measurable value.
Source: Anthropic's Claude Sonnet 5.5 launch and results, published September 28, 2026.