BREAKING
Claude Fable 5 BridgeBench Scores Drop
From Launch to Redeploy
1
Released June 9
↓
2
Export ban June 12
↓
3
Suspended
↓
4
Redeployed July 1
BridgeBench Debugging: 86.2 to 25.9
Before
86.2
After
25.9
Declines Across Categories
Debug
25.9
Refactor
38.4
Halluc.
61.7
Safety vs Usability Debate
Before
Original
●
Fallback under 5% of sessions
●
Top coding scores
After
Redeployed
●
Fallback fires on ordinary tasks
●
Developers cite reduced usefulness
Safety and Usability Debate Goes On
AI NEWS BLITZ
Anthropic's redeployed Claude Fable 5 just posted sharply lower benchmark scores.