Methodology · Mensura
Every figure on this site traces to a published upstream rate plus one flat fee. Nothing blended, nothing hidden.
Model rates come from the live provider catalog, refreshed regularly with context windows and per-token prices as published — never edited to flatter us.
$1 buys 1,000 credits, always. A provider price of $X per 1M tokens is X credits per 1K tokens — the math is exact, not rounded to flatter us.
Consumer price = provider cost × 1.20. The 20% platform fee pays for the gateway, the ledger and support, and is returned on every API response as margin_bps.
Usage (prompt/completion tokens) comes from the upstream response itself, and every response carries sonnely.credits_charged (what you paid) alongside margin_bps (the fee that produced it). Failed or empty calls are not charged; cancelled streams stop the meter.
Compare tables price 1M input + 250K output tokens as a neutral yardstick. Your workload differs — the calculator prices yours.
RITE · SHELL
$ call → empty response
error: no content · charged: 0
$ call → cancel mid-stream
meter stops · billed to last token
# nothing succeeds silently