Pin Sonnet 5.5 for well-scoped decks, sheets and reviews when speed and cost matter. Keep Opus 5.5 when the job still needs open judgement. Keep Fable 5.1 for a named hard task that already earns the switch.
Anthropic's performance, cost and GDPval-style scores are its own, and the early-tester quotes are ones it chose. Effort settings and cyber or biology safeguards can change day-to-day behaviour on the same model. The choosing-a-model page still says most workloads start with Opus 5.5, while the Sonnet 5.5 launch page positions Sonnet for well-scoped everyday work. Plan limits and the claude.ai picker default can differ by account. Sources checked 1 October 2026.
Usable results on representative scoped tasks and open-judgement tasks comparing Sonnet 5.5 with Opus 5.5, including total cost or subscription limits, correction effort and any safeguard fallback. One strong task may justify that use; a wider default needs repeated evidence.
For ordinary Claude work after Sonnet 5.5, pin Sonnet 5.5 when the job is well scoped and speed or cost matters: decks, spreadsheets, reviews and similar. Keep Opus 5.5 when the work still needs open judgement. Keep Fable 5.1 for a named hard task that already earns the switch.
This follows Anthropic's Sonnet 5.5 and Opus 5.5 positioning and its choosing-a-model guidance, checked 1 October 2026.
What the earlier rules said
The live opinion Start with Opus 5 unless a named hard task needs Fable 5.1 argued for Opus 5 as the default, with Fable 5.1 kept for a defined demanding job. After Opus 5.5 shipped on 22 September 2026, ordinary work looked better on Opus 5.5, with Fable still the named hard-task exception. The question hasn't changed: which model do you open first, and when does a harder one earn a named switch? Sonnet 5.5 changes the everyday half of the answer.
What Anthropic said on 28 September
Anthropic's Claude Sonnet 5.5 page, dated 28 September 2026, presents Sonnet 5.5 as a faster, lower-cost complement to Opus 5.5. Anthropic says it runs 30%+ faster than Sonnet 5, costs up to 30% less for most work at the same list prices of $2 per million input tokens and $10 per million output, and is strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides and spreadsheets. Where Opus 5.5 is built for complex work that needs careful judgement, Sonnet sits on the scoped side of that split. The GDPval-style scores and early-tester quotes on that page are Anthropic's.
How that sits with Opus and Fable
Anthropic's choosing a model page still says most workloads start with Opus 5.5, and lists Sonnet 5.5 for everyday coding, agent and enterprise workloads where speed and capability both matter. Its Opus 5.5 page keeps the judgement-heavy side of the family. The docs still point to Fable 5.1 when evals on Opus at higher effort fall short on demanding reasoning or long-horizon agentic work. Haiku 5.5 hadn't shipped when sources were checked on 1 October 2026. Staying on an Opus 5 setup that already works is fine if changing the picker costs you more than it saves.
How to pin it on your own work
If the job has a clear brief, a known template and a finish line you can check, start on Sonnet 5.5 and judge it by correction effort, speed and the cost or limit you actually hit. If the job still needs you to decide what matters, what to leave out or how to weigh conflicting sources, start on Opus 5.5. Keep Fable 5.1 for a named hard task where you already have a reason to use it, and compare on the same inputs before you change that. Effort settings change behaviour within a model, so note the setting when you compare.
Open the model picker, run one scoped job you know well on Sonnet 5.5 and one open-judgement job on Opus 5.5, then keep the pin that needed less fixing for the cost or limit you saw. Widen either default only when that result repeats.
Sources
Anthropic, Claude Sonnet 5.5 (28 September 2026). Anthropic, Claude Opus 5.5. Anthropic, Choosing a model. cliffinkent, Start with Opus 5 unless a named hard task needs Fable 5.1. Checked 1 October 2026.