Anthropic's Opus 5 model has achieved a record score on SlopCodeBench, a benchmark designed to evaluate how well AI coding agents produce code that resembles human-written, 'slop' code—code that is functional but includes typical human imperfections like typos, inefficient loops, and pragmatic shortcuts. The benchmark, released in June 2026, simulates real-world coding scenarios where developers often prioritize speed and readability over perfection. Opus 5 outperformed GPT-5 and Gemini 3 by a significant margin, scoring 92.4% on the test suite. The results suggest that advanced context engineering techniques allow AI to better mimic human coding styles and grasp contextual intent.
Opus 5 just proved something huge. It didn't just write correct code. It wrote code that looks like mine. Typos, lazy variable names, the occasional off-by-one. That's not a flaw. It's a breakthrough.
For years we demanded perfect code from AI. Clean, efficient, textbook. But that's not how humans work. We ship fast. We patch later. We write 'slop.' And now machines get it. They understand context, not just syntax. They see the bigger picture of a tired developer at 2 AM.
This is evolution. Not in speed or accuracy. In empathy. The AI is learning to be human. And that's the scariest and most beautiful thing we've built yet.