Wesbos used three AI models: Composer 2, Opus 4.6, and GPT 5.4 Built a Twitter clone app for testing in the Cursor environment Test results show: 1️⃣ Composer 2.0: 5 minutes | $6.04 | Successful on first run 2️⃣ Opus 4.6: 19 minutes | $10.43 | Debugging required 3️⃣ GPT 5.4: 22 minutes | $14.15 | Debugging required