Anthropic released Claude Sonnet 5.5 on September 28. It pitches Sonnet 5.5 for coding, debugging and other well-defined work. Opus 5.5 stays its pick for open-ended problems that need sustained judgment. It says Sonnet 5.5 writes output more than 30% faster than Sonnet 5 and can beat Sonnet 5's best scores at about a tenth of the cost per task. The API price stays at $2 per million input tokens and $10 per million output tokens, half of Opus 5.5. We graded it the same day as the successor to Claude Sonnet 5.
What the results show
- It does defined work well. Claude Sonnet 5.5 moves from tenth to sixth on the board. It fixes more real issues than Sonnet 5 and now writes plans as good as Claude Opus 5.5 and Claude Fable 5.1.
- It needs a clear brief. On one-line bug reports, where the agent has to work out what is wanted, it improved on Sonnet 5 but stays far behind Opus 5.5. That is the split Anthropic describes.
- It is light on the plan and the fastest agent on the bench. At the same monthly price a run of our suites takes about 40% of Sonnet 5.5's five-hour window, against about 90% for Opus 5.5 and about half for GPT-6 Sol. It answers in under a minute per task. On the same real issues that is about six times faster than Sonnet 5 and twice as fast as Opus 5.5.
Our take: this is a Sonnet done right. Sonnet 5 sat too close to Opus in plan cost and too far behind it in quality. Sonnet 5.5 is far lighter than Opus and much closer to it on defined work. Give it a clearly defined task and it does the work fast on very little of the plan. Leave the problem open and Opus 5.5 is still the one to use. We would now pair the two, with Sonnet 5.5 taking the long list of smaller tasks. We are already using it on our own project and are extremely happy with it.
Grades are comparable within the same suite version. See the full Claude Sonnet 5.5 scorecard or follow the family over time on Trends.