Feels like an end of era, ordinary people will probably never again get upgraded frontier models.
Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-07-02
Rohan Paul argues that Claude Fable 5's return with a safety-routing classifier that falls back to Opus 4.8 marks the end of unfiltered frontier model access for ordinary users, citing BridgeBench scores that collapsed across debugging, refactoring, and hallucination tasks.
Appears in
Extraction
Topics: ai-safety-routingmodel-performance-degradationfrontier-model-accessanthropicai-policy
Claims
- Claude Fable 5 was re-released with a classifier-based safety guardrail that routes flagged requests to Opus 4.8 rather than serving them on the frontier model.
- BridgeBench scores on the new Fable 5 dropped catastrophically: Debugging from 86.2 to 25.9, Refactoring from 73.6 to 38.4, and Hallucination resistance from 75.9 to 61.7.
- The guardrail is not a simple refusal layer but a routing classifier that silently degrades responses on many ordinary tasks.
- Ordinary users now receive 'permissioned intelligence' rather than direct access to frontier model capability.
- Anthropic has not publicly explained the guardrail change, and the model returned is materially different from the one that was originally banned.
Key quotes
Now we only permissioned intelligence. The cost of putting a gatekeeper inside intelligence.
Fable 5 came back, but the old promise did not. End of an era.
The new guardrails are kicking in on way too many tasks and falling back to Opus 4.8. This is not the model that got banned. Anthropic owes everyone an explanation.