SlopCodeBench: Top Model Scores Just 16% on Long-Horizon Coding Agent Benchmark
ajratner · x · 2026-10-23
SlopCodeBench may be the most relevant coding agent benchmark yet: while Terminal-Bench, FrontierCode, and DeepSWE top out at 50-75%, its best score is 16%. Agents implement a small program from a spec, then extend it across 3-8 iterations, with hidden tests each round checking new and all prior features—simulating real long-horizon codebase work where coherence and maintainability matter. Leaderboard: GPT-6 Astra XHigh 16.33% strict (69.39% core), Opus 5.5 Medium 14.29% (73.98%), GLM 5.3 XHigh 12.24%.
More from coding & agent
- AI Assistant Offers to Draft Your Replies and Track Friends Daily — "Sounds Like Hell" — cnakazawa · 2026-10-23
- Dev uses Grok Build to model 140 types of network effects into a Moog-style synthesizer gallery — seanmcdonaldxyz · 2026-10-23
- Dev builds 'NFX' network effect synthesizer with Grok from Casado's moat thread — seanmcdonaldxyz · 2026-10-23
- Why Zig's explicit allocators make it a great fit for building databases — arpit_bhayani · 2026-10-23
- How to build a Company OS with Claude Code and Slack, after 11 days of building — aakashgupta · 2026-10-23
- Zapier CEO's AI fluency rubric: the 9-step roadmap from Capable to a Transformative rating — aakashgupta · 2026-10-23