Frontier AI Labs Raise Prices on New Models as Usage Costs Climb

Frontier AI labs have released new models that cost more per token than prior versions. Fable 5, described as Mythos 5 with added guardrails, charges twice the input-token price of GPT 5.5 while producing comparable output. Anthropic announced that Fable series models will move to usage-based billing outside existing subscriptions.

The change signals a broader industry shift away from flat-rate pricing toward per-token charges.

Companies can switch to open-source models that trail the latest proprietary systems by roughly one year. For many business tasks, last year's frontier performance remains sufficient. Fine-tuning an open-source model on internal data can eliminate the need for lengthy system prompts. The instructions become part of the model weights, removing ongoing prompt-caching requirements.

Fine-tuning requires an initial investment and ongoing infrastructure management. The approach replaces a recurring API bill with GPU operations that may not justify the effort at low usage volumes. At higher scale, the same method has produced cost reductions exceeding 90 percent, according to the reported analysis.

Labs continue to subsidize usage in part to collect data that can later train smaller, cheaper models.