Community benchmark group BridgeMind said on July 2, 2026 that the relaunched version of Claude Fable 5 performs dramatically worse on its BridgeBench coding suite than the model that shipped in early June.
According to the group's published leaderboard, the July 1 build scored 25.9 on BridgeBench's debugging benchmark, down from 86.2 for the original release. Refactoring fell from 73.6 to 38.4, and the hallucination benchmark, which measures how often a model fabricates false claims when analyzing code, dropped from 75.9 to 61.7. BridgeBench describes its debugging test as TypeScript repair scoring, and its hallucination test as 30 tasks across six clusters with 175 questions verified by code execution and ground truth.
"Fable 5 came back nerfed," the group wrote on X, arguing that Anthropic owes users an explanation.
Get the ICD Newsletter
Subscribe for source-forward cyber news, OSINT notes, breach updates, and analysis. Have evidence or a lead? Send it to ICD.
