What's this about?
oran_ge tells his own users the awkward version: Fable 5.1's blended price is about a third lower than Fable 5 for the same tokens — but Artificial Analysis testing shows it consumes roughly 1.5× the tokens in Max mode, so in Max mode the cost goes *up*, not down. His advice: keep watching your token consumption. And a warning: tomorrow a more expensive model, probably GPT-6, is likely to arrive, so watch your wallet.
Why do price cuts keep failing to reduce bills?
Because the price is per token and the model decides how many tokens to use. A cheaper, more thorough model can cost more than an expensive, terser one, and the effort setting is the multiplier. This is the third distinct instance this week of a headline price improvement that does not translate: cheaper cache reads that do not affect subscription metering, an introductory rate that doubles at year end, and now a per-token cut cancelled by higher consumption.
Why is it worth noticing?
- It is a vendor telling its own users the unflattering version, which is worth noting.
- "Max mode costs more overall" matches dotey's independent 1.5× measurement exactly.
- The right metric is cost per completed task, and almost nobody publishes that — which is what makes FrontierHarness useful.