Anthropic released Claude Sonnet 5.5 on Sept. 28 as the newest model in its Claude 5.5 family, and the company says the upgrade comes without a price increase. On Anthropic’s launch page, Sonnet 5.5 is described as a clearer, faster successor to Sonnet 5, with output that is more than 30% faster and task costs that can be up to 30% lower for most work. The model docs list it as “Active (latest)” and show the same $2 per million input tokens and $10 per million output tokens as Sonnet 5. Anthropic launch page Claude docs
What actually changed
The headline is not just that Sonnet 5.5 is faster. Anthropic is trying to move a mainstream model closer to the performance of its higher-end Opus line while keeping the economics closer to a workhorse tier. The company says Sonnet 5.5 is strongest on well-scoped everyday tasks such as bug fixing, document drafting, slides and spreadsheets, while Opus 5.5 remains the better fit for complex work that requires sustained judgment. That split matters because it tells customers how Anthropic wants the models used: Sonnet for the routine, Opus for the sprawling. Anthropic launch page
Anthropic’s own rationale is that the model does more of the same work with fewer tokens and less waiting. The company says Sonnet 5.5 generates outputs 30%+ faster than Sonnet 5 and, in its testing, can cost up to 30% less per task even though the per-token prices are unchanged. That distinction is important for teams that budget around completed workflows rather than a raw token bill: the listed rate can stay flat while the total bill falls if the model needs fewer steps or shorter answers to finish a job. Anthropic launch page
Why the benchmarks matter
Anthropic says the performance jump is especially visible in coding and other agentic work. On Terminal-Bench 4.0, which Anthropic describes as an agentic coding evaluation, Sonnet 5.5 scores 70.6% versus 10.3% for Sonnet 5. On GDPval-AA, a benchmark Anthropic says covers real-world work across multiple occupations, Sonnet 5.5 scores 1,844 compared with 1,449 for Sonnet 5 and sits just two points below Opus 5.5. The launch page also says Sonnet 5.5 is strong on long-horizon work and image understanding, including being the first Sonnet model to beat Pokémon Red using screenshots only. Anthropic launch page
Those numbers are vendor-reported, so they should be read as Anthropic’s case for the model rather than independent confirmation. Still, the pattern is clear enough to explain the practical consequence: for many coding, drafting and office-work tasks, a lower-tier model is now closer to the company’s flagship than previous generations were. Anthropic also says that on several benchmarks Sonnet 5.5 at lower effort settings beats Sonnet 5’s best score for about a tenth of the cost per task, which is the kind of spread that could change default model selection inside a product team. Anthropic launch page
The boundary that still matters
Anthropic is careful not to claim that Sonnet 5.5 replaces Opus 5.5 across the board. The launch materials say Opus remains clearly stronger for complex, open-ended work that requires careful judgment over long sessions. That limitation is the real decision boundary for buyers: Sonnet 5.5 is positioned as the model to start with for ordinary work, but not as a universal substitute for higher-end reasoning. For organizations using Claude in production, the question is not only whether Sonnet 5.5 is good enough, but whether it is good enough at the effort setting and latency profile your workflow needs. Anthropic launch page
The other boundary is operational. Anthropic’s docs say five breaking changes affect code already running on Sonnet 5, including changes to up-front thinking, forced tool use, thinking blocks, the older computer-use tool on the API and Google Cloud, and the advisor tool. The docs also note that text between tool calls can now come back in thinking blocks, which can make an app appear to go quiet unless it is updated to handle that response shape. Sonnet 5.5 keeps a 1M-token context window and 128K max output, but teams that stream tool-call text or depend on older tool identifiers should treat this as a migration, not a drop-in alias swap. Claude docs
What to do next
For developers and product teams, the useful next step is a compatibility check before any rollout: verify whether your app depends on Sonnet 5’s tool-calling behavior, thinking-block handling or the older computer-use tool, then compare Sonnet 5.5 on the same workload and effort setting. If your use case is routine coding, drafting or spreadsheet-style work, Anthropic’s own materials suggest Sonnet 5.5 is the model to evaluate first; if it is a multi-hour, ambiguous or judgment-heavy task, the company still points to Opus 5.5. For a Sonnet 5.5 upgrade, start by checking whether your app depends on Sonnet 5 tool-calling or streamed thinking blocks; those are the changes most likely to break production code.
