Fable 5.1 burns through Max users' 5-hour limits in minutes; here's what's really going on

APPSO · wechat · 2026-09-02

APPSO's deep dive into Fable 5.1: on launch day Max users reported absurd quota burn — one Max5x user went 0% to 100% in 17 minutes, and a single planning-stage prompt ate most of a 5-hour window. Anthropic reset everyone's quota, and post-reset testing returned to normal, suggesting a metering bug.

Why it burns quota:

Real capability: after reset, it delivered a solid AI research workbench demo with good visualization and agent dialogue; writing quality notably improved.

Pricing reality: API prices unchanged ($10/$50 per million tokens); only CacheRead got cheaper, cutting typical task costs 25% and heavy agentic workloads up to 45%. But community estimates put Max20x weekly quota at only 2x Max5x, so cheaper per-call doesn't mean cheaper overall. Best reserved for hard, long-horizon tasks starting at Low/Medium with capped agent counts.

Related event: Claude Fable 5.1 tops benchmarks but costs rise(32 posts)→

Original post →

More from coding & agent

coding & agent channel →