Calculated from tokens, not read from a bill — but reconciled against
one. A controlled 52.5s call priced at ₹5.05 here and billed ₹5.13 in
AI Studio,
1.6% apart.
scale = prompt_token_count ÷ (detail_text + detail_audio)
billed_text = detail_text × scale ·
billed_audio = detail_audio × scale
cost = Σturns (billed_text × rate_text + billed_audio × rate_audio)
The per-modality breakdown falls ~18% short of the reported total, so the
reported total is apportioned across it — otherwise those tokens are
priced at zero. Rates are per model; using one model's card for another
is a silent 2× error.
Only calls made through this server appear here. Scripts hitting the same
API key bypass the ledger, so any comparison against an account-level
total will under-read.