Performance Reports Raise Eyebrows
Claude Fable 5's recent relaunch on July 1 has sparked controversy regarding its performance. Initial assessments show a stark decline in metrics according to BridgeBench, which recorded a drastic fall in debugging scores from 86.2 to 25.9, while refactoring scores also dropped significantly. This decrease has been attributed to a new safety classifier that rerouted many tasks to the older Opus 4.8 model, rather than indicating a true decline in the AI's capabilities.
FABLE 5 RETURNED NERFED. The new guardrails are excessively restrictive across many tasks.
— BridgeMind (@bridgemindai) July 2, 2026
Contrasting Evaluations
In contrast, Arena.AI's findings suggest that Fable 5's performance remains stable since its last version update, with some improvements in specific areas like document processing. Arena utilized a blind human-preference survey method, reporting only minor changes that fell within normal variation.
The community has been curious about how Claude Fable 5 performs pre- and post-deployment.
— Arena.ai (@arena) July 2, 2026
Impact on Users
The backlash from users has been significant, with many remarking on social media that the tool feels 'broken' and 'nerfed.' Users engaged in creative writing and document analysis reported minimal changes to performance, but developers noted substantial rerouting when engaging with security-related tasks. The discrepancies between BridgeBench and Arena likely stem from the types of tasks evaluated, with BridgeBench's prompts triggering the new safety measures more frequently.




