AICost positions itself as "an independent price desk for every AI model on the market," with the core purpose of bringing together LLM API prices scattered across different vendors' pricing pages in one place for comparison. The website emphasizes independence and verifiable sources, and labels each price with the date it was read.
Purpose AICost is used to compare API pricing for mainstream AI models, including input cost and output cost per million tokens, and further provides blended rates that can be converted into actual bills, subscriptions, AI agents, and a calculator. It is aimed at developers, teams, or procurement decision-makers who need to evaluate model invocation costs.
Main capabilities
- Pricing comparison: covers API token prices from major labs and provides a "key comparisons" section.
- Blended-rate calculation: the website states that its blended rate assumes 92% input / 8% output and 82% cached load, and this assumption is applied to all effective rate figures on the site. The raw input and output rates are read from each vendor's own pricing page. If a vendor does not publish a cached input rate, no cache discount is applied, so that model's blended figure is an upper bound rather than a fully like-for-like comparable value.
- Model and tool pages: it provides per-model pricing pages for 23 frontier providers. The first calculator, "LLM API cost," is already live and uses the same verified data set. The site mentions that a side-by-side comparison feature is coming soon.
- Live log and summaries: it records every price change, model release, and policy change in the AI market and promises to record them within hours rather than weeks; it provides a weekly summary.
How to use Users can browse each model's pricing page directly through the website, use the LLM API cost calculator, and refer to the key comparisons. The website states that pricing data is manually verified against each vendor's official pricing page, without using crawlers and without relying on third-party aggregators; every number is accompanied by a verification date. If a number is wrong, the page is corrected and entered into a public correction log recording the date, page, incorrect content, and the current correct content.
Use cases Suitable for developers, technical teams, and cost-sensitive users who need to understand LLM API input/output token costs, estimate API bills, track model price changes, and follow the impact of cache discounts and blended rates.
Limitations The website explicitly states that it does not claim to cover content it has not yet covered: coverage is 23 frontier providers, and the side-by-side comparison feature is not yet live. The blended rate is based on fixed assumptions (92% input / 8% output, 82% cached load), and when a vendor has not published a cached input rate, thatfigure is an upper bound rather than like-for-like comparable. In addition, the website page text mentions several specific claims about pricing and billing (such as 429 being free, retry costs, batch API discounts, Claude Code caching, etc.), but the materials do not provide their detailed data or verification results, so no further expandable details are available at this time.
