diff --git a/changelog.mdx b/changelog.mdx
index f90a492..17e7c99 100644
--- a/changelog.mdx
+++ b/changelog.mdx
@@ -22,6 +22,21 @@ export const STAGE_SELF_MANAGED_M66 = "$0.0001";
usezombie is in **stealth-mode testing** and pre-production. APIs and agent behavior may change between releases without long deprecation windows. Email [usezombie@agentmail.to](mailto:usezombie@agentmail.to) if you want a hand calibrating an agent or to join as a design partner.
+
+ ## Platform token spend now actually bills
+
+ Runs on a platform-provided model were billed the run fee only — the tokens the agent consumed priced as zero. The platform has always known how to charge for token usage, but the runner never reported how many tokens a run spent, so the token component of every platform run settled at zero. A run's input and output token usage is now reported as the run executes, so platform token spend is charged the way the rate card always described.
+
+ ## Upgrading
+
+ - **Update your runner fleet to bill token spend.** The fix is in the runner binary. A host running an older runner keeps working and is never charged a wrong amount — it simply continues to bill run-fee-only, exactly as before, until you update it. Update your runner hosts to start billing the token component on the runs they execute.
+
+ ## Bug fixes
+
+ - **Platform-model runs under-billed** — a run that consumed model tokens on a platform-provided key was charged only the per-second run fee; the token component was zero. Such runs now bill the input and output token counts the agent actually consumed, metered as the run accrues rather than only at the end.
+ - **Cached-input tokens** are not yet billed separately: until the model-provider layer reports cache reads on their own, cache-heavy runs bill those tokens at whatever class the provider folds them into. No run is over-charged for them.
+
+
## Agent memory loss is now measurable