What is Opus 4.5?

As of mid-2026, Opus 4.5 failed 11% and GPT-5.4 failed 25% of automated prompt injection attacks, with real-world human attacker success rates far higher.

“Opus 4.5 and GPT-5.4 still failing 11% and 25% of the time respectively against a set of automated attacks and real-world vulnerability against adaptive human attackers was much higher.”

Carlos García on scrapping a working, profitable system to bet on newer AI models.

“We decided to like destroy everything we had been building for two years that was working that brought us to profitability.”

Claude Code's multi-agent code review feature, which traverses an entire codebase simultaneously, only became production-reliable with Opus 4.5, Opus 4.6, and Sonnet 4.6.

“It was only with like Opus 45 and 46 that we and Sonnet 4.6 that we felt like okay we are now able to like run multiple code review agents simultaneously to traverse the entirety of the codebase and to synthesize a set of like real issues that an engineer needs to address before merge.”

Claude (Anthropic) crossed a usefulness threshold for research-level mathematics around Opus 4.5 or 4.6, roughly catching up to ChatGPT after a prolonged period of being unfit for the task.

“For a long time the claw model models were just like not useful for research math and then I think maybe around opus 4.5 or opus 4.6 they like more or less caught up.”