Reviewer says Claude's newest AI model finishes jobs in one try

A reviewer at MakeUseOf says Claude Opus 5.5, the newest model from Anthropic, is much better at finishing a job from a single request. Writer Yadullah Abidi says earlier assistants often needed repeated corrections and reminders. In his testing, Opus 5.5 got to a finished result far more often without that back-and-forth.

What changed

Abidi tested it on smart home dashboards, CAD and 3D design projects, coding tasks and general research. His clearest examples were two Home Assistant dashboards. One covered his main home. The other tracked AC costs and rent splits, and he gave it very little background. Each came back in one shot, with no string of fixes.

He also says the way he writes requests has changed. Instead of listing every step, he describes the result he wants, sets limits, and says what must be checked before the job counts as done. He notes that work still happens behind the scenes. The difference is that one request can start the whole process.

What the tests show

Some numbers come from outside the review, and some come from the vendor. The vendor figures are less independent.

• CodeRabbit found 51 of 80 known bug patterns, against 49 for its baseline. Abidi calls that a modest gain.
• It also caught a bug involving overlapping database updates that the baseline missed.
• In an Anthropic merger test, it built an Excel model and a slide presentation in 63 minutes. Opus 5 took 93.
• In another Anthropic test, an automated grader passed 16 of 18 research reports that were screened for invented figures or quotes.

What to keep in mind

The merger test was run by the vendor and used fictional companies. Abidi says that means it is not about to replace financial analysts. The research result also came from an internal test. He still checks every source an AI model uses, and he treats Opus as a capable reviewer, not a reason to skip testing.

For everyday users and small businesses, the takeaway is practical. A tool that needs fewer fixes can save time, but its output still needs a human check, especially for money, code and facts.

What is not known yet

The material does not say what Opus 5.5 costs or which plans include it. It also does not say how it performs outside the reviewer's own projects and the tests above. The review text provided was cut off, so any further conclusions from Abidi are not covered here.