Fable 5 user says the model got dumber after Anthropic bundled it into subscriptions

A widely shared post claims Anthropic’s flagship Fable 5 model degraded noticeably once the company made it permanently available inside subscription plans. The author reports the model simply “felt dumber” and, unable to pin down why from experience alone, turned to measurement: across five separate metrics, August generated far fewer reasoning (“thinking”) tokens than July. Because Fable 5 runs with thinking always on and its reasoning depth scales with an effort setting, a sustained drop in thinking tokens is a plausible mechanism for the perceived quality loss — less deliberation per request would show up exactly this way.

The claim is notable mainly because of its timing. Pinning a reasoning decline to the moment a frontier model is folded into consumer tiers feeds a recurring suspicion that providers quietly throttle expensive reasoning to control serving costs as usage scales. The token-count evidence gives the complaint more substance than the usual vibe-based “it got nerfed” posts.

That said, this is a single user’s self-reported analysis, not an audited benchmark. The methodology behind the five measurements isn’t detailed, thinking-token usage varies heavily by task mix, and Anthropic has not confirmed any change to how Fable 5 allocates reasoning effort. The correlation with the subscription rollout is suggestive rather than conclusive — worth watching, but short of proof that the model was deliberately downgraded.