Prices checked

AI API cost calculator

What your AI workload costs per month, and the cheapest model in the list for the same tokens. Every price comes from the vendor's own pricing page, linked below. Runs in your browser: no sign-up, no tracking.

1. Your workload

Pick an example or type your own numbers. Example token sizes are my estimates, not vendor data.

Tasks per month (API calls) Input tokens per task (prompt, context, history) Output tokens per task (includes thinking tokens) Cached share of input (optional, % billed at cache hit price)

Use announced future prices (Gemini 4 from Dec 1, Gemini 3.8 Flash from Jan 1) Cheapest alternative only among my picks (off: search all models) DeepSeek off-peak (half price outside peak hours)

2. Models to compare

3. Monthly cost

Model	Per month	Per 1,000 tasks	Per task	Switch to cheapest saves

Full ranking: all models for this workload

#	Model	Per month	In / cached / out per 1M	Source

How the math works, and what it leaves out

Cost per task = fresh input x input price + cached input x cache hit price + output x output price, all per 1M tokens. Monthly = cost per task x tasks.
Price is not quality. The cheapest model may fail more tasks. Check your own pass rate before you switch.
Long context cliffs: some models bill the whole request at higher rates once input crosses a line (Claude Haiku 5.5 over 100K, Meta Muse Ember over 200K, OpenAI GPT-6 and GPT-5.6 lines over 272K). The calculator applies this per task.
Tokenizers differ. The same text can be more tokens on one model than another. Anthropic says Claude 4.7 and later produce about 30% more tokens for the same text.
Not included: cache write fees, cache storage fees, batch and flex discounts, fast or priority modes, web search and tool fees, regional or data residency uplifts, taxes, volume deals.
Standard pay as you go API prices in USD. GPT-5.6 Sol's price is a promo until Nov 30, 2026; OpenAI has not published what comes after.