July 12, 2026
Metered (cost-weighted) tokens
- Plan allowances now count metered tokens: each token category counts in proportion to its real AI-provider cost — cached reads count less, output counts more. The exact weights are published in the docs and come from the same rate table the live meter uses.
- For typical builds (mostly cached context) this means more effective headroom than raw counting; it also guarantees no usage pattern can cost more than a plan includes.