Token accounting for a run, summed across every model call it made.
cachedInputTokens is a SUBSET of inputTokens (the provider counts a cache hit in both), so
the billable split is inputTokens - cachedInputTokens at full rate plus cachedInputTokens at
the cached rate — never inputTokens + cachedInputTokens.
Token accounting for a run, summed across every model call it made.
cachedInputTokensis a SUBSET ofinputTokens(the provider counts a cache hit in both), so the billable split isinputTokens - cachedInputTokensat full rate pluscachedInputTokensat the cached rate — neverinputTokens + cachedInputTokens.