AI Search tests Claude Opus 5 using Claude Code to build a Windows-style browser interface, reconstruct a room from an image and produce a financial presentation video. The reviewer shows useful interactive elements and visual results but also describes incomplete features, hardcoded behavior and inaccuracies. The financial narration is generated material, not independently verified financial reporting.
Further demonstrations use Blender to create and animate an X-wing model and a digital audio workstation to arrange a techno track with downloaded virtual instruments. The reviewer reports long runs and substantial token use. The examples establish what this particular workflow produced, not that the model reliably completes every comparable creative task.
The review also reports failures to find a concealed frog and correctly identify tumors in a small image test. These demonstrations are not clinical validation or medical advice. The reviewer distinguishes the model's willingness to answer biomedical prompts from the accuracy of its answers.
AI Search compares reported benchmarks, context size, pricing and restrictions, noting that leaderboard differences may be uncertain where confidence intervals overlap. The final recommendation favors cheaper or faster alternatives for the reviewer's own work and reserves Opus for selected difficult coding or creative tasks. This is a subjective purchasing assessment, not a universal performance conclusion.
Watch on YouTube
