Claude Fable 5 came back online July 1, and the internet decided it was broken. Nerfed. Lobotomized. The benchmarks seemed to agree — until you looked closer.
BridgeBench’s numbers looked brutal. Debugging dropped from 86.2 to 25.9. Refactoring fell from 73.6 to 38.4. But here’s the catch: of 12 TypeScript debugging tasks, only three actually reached Fable 5. The other nine were intercepted by Anthropic’s new safety classifier and rerouted to Claude Opus 4.8. BridgeBench scores every fallback as zero because the wrong model answered.
So the model didn’t get dumber. The gatekeeper got more aggressive.
Arena.AI ran thousands of blind human-preference votes and found something different. Fable 5’s performance was mostly flat compared to the June version. In document and expert text categories, it actually improved after reinstatement.
Both benchmarks are correct. It just depends what you’re measuring.
Anthropic deployed the new classifier as a condition of Fable 5’s reinstatement after US export controls were lifted. It was trained to block a specific jailbreak technique that got Fable 5 to identify software vulnerabilities. It works. It also catches a lot of things it shouldn’t. Debugging TypeScript looks enough like “security work” to trigger the fallback constantly.
Anthropic has acknowledged the false positives and says the system will be refined. No timeline yet. For now, if you’re using Fable 5 for coding, you might be talking to Opus 4.8 more than you realize.
