Zvi says Fable 5.1 looks strong, but hard to rank

What the source reports

Mowshowitz reads the launch as a strange two-model moment. Fable 5.1 and GPT-6 Astra both look strong, while early public tests still have large error bars. He reports lower cache-read costs and zero data retention for eligible customers. The numbers are the story. Cache reads fall from $1 to $0.25 per million tokens. Boris Cherny says typical Claude Code sessions can be up to 38% cheaper. Biology safeguards intervene on benign requests 85% less often, and Claude Code users should see about 60% fewer cyber interventions. Zvi’s own practice is modest: use both models on hard questions while the evidence settles.