A new leaderboard built specifically to judge AI models on vibe coding has crowned Claude Fable 5 as the model to beat, posting a 90.35% overall application pass accuracy score on Vibe Code Bench v1.1, according to BenchLM.ai’s July 2026 snapshot. Claude Opus 4.8 came second at 82.72%, with Claude Sonnet 5 third at 81.33%. The benchmark asks models to build full web apps wired up to real services including Supabase, Stripe test mode, email and file editing. For Australian developers, the result matters because it is the first widely cited benchmark built around finished, working software rather than isolated code snippets, and it lands as local search interest in vibe coding tops 9,900 monthly queries even as the broad term itself starts to cool.