Fable 5.1’s cheaper cache meets its appetite for tokens

Fable 5.1 earns a positive review for writing, discussion and broad capability, with a recurring qualification: give it more reasoning budget and it tends to find more work.

The launch left headline input and output prices unchanged while cutting cache reads from $1 to $0.25 per million tokens. That can make repeated context much cheaper, but the saving depends on how much of a session is cached and how much new output it generates.

Zvi Mowshowitz also points to less intrusive safeguards and changes in data-retention eligibility as practical improvements. His enthusiasm does not erase the operating-cost question. A more capable model with cheaper cached input can still make a routine task expensive by pursuing it at excessive depth.