Tech X Telco

Claude Sonnet 5.5: faster work, API costs and what to check

Anthropic has launched Sonnet 5.5 for everyday work. Its speed claims, API pricing and migration changes need different checks.

Michel Elijah ·

AI-generated editorial image of routine document work beside a laptop; not a Claude interface.

Anthropic released Claude Sonnet 5.5 on 28 September for everyday work. For people already using Claude, the practical question is whether it reduces the time spent getting a usable result.

What you need to know

Anthropic says Sonnet 5.5 generates output at least 30% faster than Sonnet 5 and can cost up to 30% less per task. Those are company findings, not a promised subscription discount. Compare the finished work and the time you spend correcting it.

Where can you use it?

Anthropic’s Sonnet availability page lists Sonnet 5.5 on Claude.ai through the web, iOS and Android. Its developer documentation lists the Claude API and supported cloud platforms, including Amazon Bedrock, Google Cloud and Microsoft Foundry. Check the model shown in your own account rather than assuming every existing conversation has changed.

What does “lower cost” mean?

Anthropic attributes lower task costs to fewer tokens, with unchanged unit rates. Task cost and token price are different measures. The announcement does not establish a reduction in your Claude subscription charge.

The standard Claude API rates are US$2 per million input tokens and US$10 per million output tokens. Opus 5.5’s corresponding rates are US$4 and US$20. These are USD developer prices, separate from consumer subscription pricing.

As an illustrative calculation, 100,000 input tokens and 10,000 output tokens cost US$0.30 with Sonnet 5.5: US$0.20 for input plus US$0.10 for output. The same volumes at the standard Opus 5.5 rates cost US$0.60. This example excludes caching, batch discounts, tool charges, other pricing modifiers and tax; it is not a measured real task.

Try a job with a clear finish line

Our suggested comparison is a short piece of work you can assess yourself. For an email, supply the facts, audience and length. For a spreadsheet, provide a small set of figures with a known total. For a presentation, specify the number of slides and the decision each should support.

Keep the brief and source material the same when comparing models. Record how long the first draft takes, how many corrections you request and whether the final output meets the brief. A quick answer that needs extensive repair can cost more of your working day than a slower usable draft.

Anthropic positions Opus 5.5 for more complex, open-ended work. Compare that claim with your own tasks; we have not run a hands-on comparison.

Comparing the two models? Read our Opus 5.5 report on access, limits and API prices.

Developers need to check the migration details

The API identifier is claude-sonnet-5-5. The migration documentation describes breaking changes: a previous disabled-thinking configuration needs the new between_tools setting, and forced tool selection can return an error. Read the migration guidance before swapping the model identifier in a working application.

A sensible trial should include the requests your application actually sends, its tool handling and the output your users see. Decide what would count as a failed upgrade before the trial. Keep a known working configuration available while you assess the new one.

Sources and further reading