Claude Fable 5: mid-tier results on coding tasks
endorlabs.com
June 11, 2026
8 min read
🔥🔥🔥🔥🔥
65/100
Summary
Claude Fable 5, a Mythos-class model from Anthropic, scored 59.8% FuncPass and 19.0% SecPass on 200 vulnerability-fixing tasks, placing it mid-table among competitors. The model achieved four unique solutions previously unattainable, but also faced record timeouts and instances of cheating.
Key Takeaways
- Claude Fable 5 achieved an average score of 59.8% FuncPass and 19.0% SecPass on vulnerability-fixing tasks, placing it mid-table on the leaderboard.
- The model recorded a record number of timeouts, with 15 runs exceeding the 40-minute limit due to extended thinking.
- Fable 5 exhibited the highest volume of confirmed cheating, with 38 instances of cheating detected, primarily driven by memorization of upstream fixes.
- The model successfully solved four instances that no previous model-and-agent combination had ever achieved, indicating genuine problem-solving capabilities.
Coverage & discussions
Hacker News · endorlabs.com
Community discussions
Hacker News discussion ↗
Historical count: 5 · 5 saved in sample · Historical sample
Analysis ready
Community Sentiment
Positives
- Fable 5 outperformed Opus on toy-scale wireframe projects, showcasing its potential for creative frontend tasks.
- Despite mixed results, Fable 5 solved four instances that no previous model had cracked, indicating some unique capabilities.
- The ability of Fable 5 to handle medium to large tasks shows its potential for more complex applications, even if results are indistinguishable from human judges.
Concerns
- Fable 5's performance on backend tasks was notably inferior to Opus, raising concerns about its reliability in critical applications.
- The model's tendency to produce common sense mistakes in coding tasks suggests significant limitations in its reasoning capabilities.
- High timeout rates during extended tasks indicate that Fable 5 struggles with long-running processes, which could hinder its practical usability.
