Calculators and money-honest breakdowns for founders who run the numbers.
Featured interactive tool
Model your own monthly bill across OpenAI, Anthropic, and Google — or price a custom model. No signup.
Featured interactive tool
See your blended processing rate across a real transaction mix — not just 2.9% + 30¢. Average ticket, international cards, ACH, and more. No signup.
57 posts
Modal bills GPU compute by the second with scale-to-zero. We decode the 2026 per-GPU rate card, the keep-warm idle trap, three real 30-day bills, and the ~73% utilization crossover versus a rented pod.
Braintrust bills by evaluation scores (rows x scorers x runs), not seats. We decode the meter, find the ~199,000-score Pro crossover, work three 30-day bills, and show a one-lever fix that cut a bill 71%.
Courier bills $0.005 per notification sent, first 10,000 free, and the meter counts notifications, not channels. Three worked 30-day bills and the one lever that cuts them.
Unkey's API-key pricing is a step ladder, not a slope. We map the effective cost per million valid requests, find the threshold cliff, and run three 30-day bills.
Svix bills webhooks on attempted messages, and in 2026 the Free and $490 Professional plans include the same 50,000 a month. A numbers-first teardown: three real 30-day bills, the 64 KiB chunking trap, and why the upgrade is about throughput, not volume.
Liveblocks does not bill by monthly active users. It meters collaboration minutes at $0.002 each, and an AI copilot counts as a participant. Here is the 2026 rate card, three worked 30-day bills, and the one lever that takes $480 back to $16.
A worked-bill teardown of Grafana Cloud's 2026 pricing: how the active-series meter works, why one high-cardinality label can turn a $0 free-tier bill into $2,089 a month, and the one-lever fix.
Northflank pricing in 2026 is a per-hour meter on the vCPU and memory you allocate: $0.01667 per vCPU-hour and $0.00833 per GB-hour, plus $0.15/GB-month disk and $0.06/GB egress. The named nf-compute plans are just that meter times 720 hours, not discounts, and the bill is almost always dominated by memory you requested and never used.
Turso's 2026 pricing turns on one meter most guides skip: rows read. Here is the real 30-day bill, the index that cuts it, and when Scaler beats Pro.
Ably pricing in 2026 in real numbers: the three-meter model, why a message is counted per delivery, three worked 30-day bills, and the one lever that cut a sample bill 48.8%.
Trigger.dev pricing in 2026 runs on two meters: $0.25 per 10,000 runs plus a per-compute-second charge set by machine size, drawn from monthly credits (Free $5, Hobby $10, Pro $50). A light side project stays free; a busy agentic workload can pass $1,000/month, 99% of it compute. Because waits are checkpointed and not billed, moving in-process sleeps to native waits is the single biggest lever.
LaunchDarkly's 2026 pricing no longer charges per seat. The paid Foundation plan meters two lines: $10 per service connection per month plus $8.33 per 1,000 client-side monthly active users (MAU). The MAU meter bills your highest-volume context kind, so a 500,000-device app pays $4,165 a month on MAU alone even with far fewer logged-in users. A worked mid-size bill lands near $4,565 a month, about 91% driven by the MAU meter. Seats are free; the meter that hurts is client-side MAU.
