Claude Fable 5: The harness matters more than the model
SMRTR summary
The same Claude Fable 5 model scored poorly under Claude Code (59.8% FuncPass, 19% SecPass) but topped the leaderboard under Cursor (72.6% FuncPass, 29% SecPass). The agent framework surrounding a model, not the model itself, largely determines how well AI fixes security vulnerabilities — yet even the best combination still misses roughly 70% of security flaws.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original article