LLM API Cost Calculator
Estimate what the same workload costs on 35 verified models — plus more behind an explicit opt-in. Enter your monthly volume once; the table re-sorts itself, cheapest first, with the footnotes that change bills left visible.
Models
Tick up to four, or use “Compare all models” above. † marks a model with a scheduled price increase — see the footnotes under the results.
You can pick up to 4 models — or use “Compare all models” for the full table.
| Model | Input cost / mo | Output cost / mo | Total / mo | Notes |
|---|
How to read these numbers
An LLM bill is just two multiplications — requests times input tokens, requests times output tokens, each priced per million — but the headline rate card hides three traps, and this calculator is built to surface all of them. The first is peak pricing: DeepSeek's published rates are off-peak rates, and traffic during its weekday peak windows costs exactly double, so a calculator that quotes the headline understates a real weekday bill by up to half. The toggle above reprices DeepSeek at peak so you can budget for the worse case. The second trap is scheduled increases: several current prices are introductory or promotional with a dated rise already announced. Models carrying one are marked † and their footnotes spell out the future rate, because a migration decision made on a promo price expires with the promo. The third trap is unverified numbers: prices that circulate through third-party trackers without appearing on an official page. Ours are badged “Price pending verification”, kept out of the default comparison, and only included when you explicitly opt in — a budget deserves to know which of its inputs are solid.
A few usage notes. Averages matter more than precision: take your input and output token averages from a week of real traffic rather than a guess, and remember that long system prompts and retrieved context inflate input tokens fast — cached-input rates, shown on the pricing leaderboard, can cut that part sharply for repeated prefixes. Reasoning models may bill hidden reasoning tokens as output, so treat their output average generously. And if two models land within a few percent of each other, choose on quality and latency — at that distance, the cheaper model is the one your users do not make you retry.
Every price here carries its official source link and the date it was last verified, and the full dataset is the same one behind the leaderboard. For the complete table — including cached-input prices and context windows — see LLM pricing.
MyGateway — one endpoint for every LLM API, bring your own keys. Join the waitlist.
MyGateway is a bring-your-own-keys gateway for LLM APIs: one endpoint, one key format and one usage view across OpenAI, Anthropic, Google, DeepSeek and more — while the provider keys, and the bills, stay yours. It is currently in private testing. Join the waitlist and we will email you when public invites open.
You're on the list — we'll email you when invites open.
Something went wrong. Please check the email address and try again.