I tested Opus 4.5 to see if its really the best in the world at coding - and things got weird fast

▤ Summary

This article tests Anthropic’s Claude Opus 4.5 on several coding tasks and finds mixed results. The model fails on file handling, a WordPress plugin build, and a JavaScript validation fix, while it performs better on a more complex cross-app workflow. The author concludes that Opus 4.5 looks unreliable in the basic chatbot interface, even though it performs strongly inside Claude Code for agentic coding. The piece also highlights a gap between Anthropic’s marketing claim and the model’s real-world behavior. It ends by noting that the model may still improve, but it does not yet feel ready for prime time in this setup.

▦ Linked Companies

Anthropic

$10M to $25M