Anthropic has released Claude Sonnet 5.5, the second model in the Claude 5.5 series after Opus 5.5. According to the official line, it outputs over 30% faster than the previous Sonnet 5 and can save up to 30% of the cost for the same tasks. These two numbers are the core selling points of the entire launch, and also the part this article must repeatedly flag: they come from Anthropic's official statements and lack independent measurement and definition of methodology.
To understand the weight of this release, one must first look at Anthropic's tiering of models. Claude's publicly available models fall into four tiers, from strongest to weakest: Fable, Opus, Sonnet, Haiku, with lower tiers being cheaper. According to Anthropic's positioning, Fable handles large projects that run for hours across an entire codebase, Opus 5.5 handles complex work requiring repeated trade-offs, and Sonnet 5.5 excels at well-scoped daily tasks such as fixing bugs, writing documentation, making slides and spreadsheets. Haiku 5.5 will be released within a few weeks.
In other words, Sonnet 5.5's official persona is 'the cheap tier for daily work.' But what truly sparked discussion about this release is that on several benchmarks it no longer looks like a mid-tier model, but sits right next to Opus 5.5. This raises a question worth asking: when the cheap tier's benchmark scores approach the flagship tier, why does Anthropic still emphasize that Opus 5.5 is stronger?