This article is one chapter of my book Delphi in all its glory – AI-assisted development, the fifth book of the series.
According to Anthropic (therefore, take these numbers as a direction, not as a promise), Claude Opus 5.5 works at the level of Fable 5.1 on most jobs, and costs 40% less to run than Opus 5.
Input and output tokens are 20% cheaper. Cache reads are 60% cheaper, and Anthropic says cache reads make up most of the cost of agentic and coding work. Opus 5.5 also uses fewer tokens per task. Lower prices plus fewer tokens: that is how Anthropic gets to the 40% saving. The output also comes more than 30% faster.
Anthropic says the model is “particularly good at long and sprawling jobs like codebase-wide migrations and audits”. Think of upgrading an old Delphi project. One early tester audited and fixed a 200,000-line codebase in under three hours. Opus 5 needed more than 20 hours and 2.5 times as many tokens.
Deloitte reported that Opus 5.5, at its lowest effort, caught 72% of the known bugs in their code reviews. Opus 5, at high effort, caught 56%.
It follows your writing rules (says Anthropic)
Later in this book (see the chapter “Fixing Claude’s Output”) I show how to make Claude write plainly with an “output style”: a file with your writing rules.
Anthropic now says that Opus 5.5 “puts the most important information up front, is less likely to use jargon or idiosyncratic phrases, and follows the writing rules you give it”.
But test this on your own rules before you believe it.
The leash
Anthropic gives Opus 5.5 cybersecurity safeguards “similar to Fable 5.1’s”. Anthropic says you can still find and fix bugs in your own code, and even scan your source code for vulnerabilities. But “most cybersecurity tasks will be re-routed to Opus 4.8”. Anthropic’s examples: exploit generation, binary-based vulnerability scanning and penetration testing.
The re-routing does not happen silently. You see a notice that the model switched, and the answer is labeled with the model that wrote it. But the rest of the conversation stays on Opus 4.8, until you switch back yourself. In Claude Code you can turn off the automatic switching (Config > MODEL & OUTPUT > “Switch models when a message is flagged”). A flagged request is then stopped instead. Anthropic says that the conversation pauses and lets you switch to the other model, or edit the prompt and retry. When the stop came in the middle of Claude’s work, Claude Code only printed a notice, and Claude carried on without the stopped answer.
This was one chapter of a whole book
You just read one chapter of Delphi in all its glory – AI-assisted development: more than 400 pages about Claude Code and AI for Delphi programmers. Written by a Delphi programmer, with the failures documented next to the wins.