Claude Sonnet 5.5: Faster Work and a November Deadline

If your business runs anything on Claude Sonnet 4.5, you now have a date in the diary: 30/11/2026.

What happened

On 28/09/2026 Anthropic released Claude Sonnet 5.5, the second model in its new 5.5 family and the cheaper, quicker partner to Claude Opus 5.5, which arrived a week earlier on 22/09/2026.

Sonnet 5.5 is available on every Claude plan, from Free through Pro, Max, Team and Enterprise, as well as through the Claude API and on Amazon Web Services, Google Cloud and Microsoft Azure. Anthropic says it runs more than 30 per cent faster than Sonnet 5 and costs up to 30 per cent less for most work.

That saving needs reading carefully. The list price has not moved: it is still 2 US dollars per million input tokens and 10 US dollars per million output tokens, roughly £1.51 and £7.54 at the 01/10/2026 exchange rate. The saving comes from the model needing fewer tokens to finish the same task, so your bill only falls if your work behaves like the work Anthropic measured.

Two days later, on 30/09/2026, Anthropic told developers that Claude Sonnet 4.5 will be retired from the Claude API on 30/11/2026, with Sonnet 5.5 named as the replacement. After that date, any request still pointed at the old model will fail.

Why it matters for UK businesses

Most small firms we work with do not choose models by hand in the chat window, and they do not need to. Where the model choice matters is in the things you built once and stopped looking at: a Make or Zapier scenario that drafts quote replies, a script that summarises inbound enquiries, a custom tool a contractor wired up last year. Those often have a model name typed into a settings field, and that name does not update itself.

The benchmark that stands out for office work is GDPval-AA, which tests models on real tasks across 44 occupations and nine industries. Sonnet 5.5 scored 1844 against 1449 for Sonnet 5, and sits almost level with Opus 5.5 at 1846. Anthropic also describes giving the model a company's quarterly results, call transcripts and a slide template, then asking for a ten slide operating review; two experts judged the first draft ready to send.

For a business paying per token through an API or an automation platform, a model that does Opus grade office work at Sonnet prices is the bit worth paying attention to.

How Adevious AI sees it

We would make Sonnet 5.5 the default for routine automated work straight away: email triage, document summaries, first drafts of proposals and reports. On the evidence published, there is little reason to pay Opus rates for that kind of task any more.

We would keep Opus for the jobs where being wrong is expensive, such as long contract reviews, multi step reasoning over financial data, or code that touches payments. On harder coding tasks the gap is still real: 46.2 per cent for Sonnet 5.5 against 54.4 per cent for Opus 5.5 on FrontierCode.

A few cautions before you switch. First, release day numbers come from the vendor, and they lean on coding and agent tests rather than the messy spreadsheets and half finished Word files most firms actually hand over. Run your own sample before you trust them. Second, the 30 per cent saving is a claim about token counts, not a price cut, so check your usage report after a fortnight rather than budgeting for it up front. Third, effort settings now matter: in the Claude apps and Claude Code the default effort is Medium, while the API defaults to High. A workflow moved across without checking that setting can come out slower and dearer than you expected, or thinner than you wanted.

Finally, a model swap is a change to a live process. Outputs will shift in tone and length. If a downstream step parses the reply, for example pulling a figure into a spreadsheet, test it before you flip the switch.

What to do this week

Make a list of every place your business calls Claude outside the chat window. Check automation platforms, internal tools, website chat widgets and anything a freelancer built for you. Search each one for the text sonnet-4-5.

Where you find it, change the model to claude-sonnet-5-5, run ten real examples through it side by side with the old output, and check the effort setting. Book that work before mid November so you are not doing it in the last week, when any problem leaves you with no fallback.

If you are not sure where Claude is wired into your systems, that is a useful thing to find out in its own right.

Talk to us

Adevious AI builds and maintains Claude powered tools for UK businesses. If you would like a hand auditing your automations before the 30/11/2026 cut off, or deciding which jobs belong on Sonnet and which on Opus, get in touch and we will take a look.

Sources: https://www.anthropic.com/claude-sonnet-5-5 https://www.anthropic.com/news https://platform.claude.com/docs/en/about-claude/model-deprecations https://support.claude.com/en/articles/12138966-release-notes

Acronyms: AI: Artificial Intelligence API: Application Programming Interface UK: United Kingdom