CostPerPrompt tracks live pricing across 232+ AI models

A Show HN post introduces CostPerPrompt, a website that tracks live pricing for more than 232 AI models, refreshed automatically, alongside calculators meant to answer a concrete question: what a given chatbot, agent or API workload will actually cost per month. The site's own explainer walks through how AI API billing works: providers charge per token, where one token is roughly three quarters of a word, with separate rates for input tokens (what a user sends) and output tokens (what the model generates). Output tokens usually run 3 to 5 times more expensive than input tokens. As a worked example, a model priced at $5 and $25 per million tokens costs $5 for every million tokens sent to it and $25 for every million tokens it writes back. The site argues that two discounts change the real cost picture substantially: prompt caching, which cuts the cost of repeated input by up to 90% and matters most for chatbots that resend prior conversation history with every turn, and batch processing, which takes about 50% off the price when a workload can tolerate waiting for results. According to the post, its calculators account for both of these discounts, whereas most existing cost-estimate articles do not, which the site says leaves their estimates running 2 to 3 times too high or too low. As of the post, the flagship model pricing table shown on the site was last updated on 2026-08-02. The Show HN submission had drawn 13 points and 5 comments at the time this was written; the text does not name the person, team or company behind the project.

Key facts

CostPerPrompt lists live, auto-refreshed pricing for 232+ AI models.
Output tokens typically cost 3 to 5 times more than input tokens across providers.
A worked example prices a flagship model at $5 per million input tokens and $25 per million output tokens.
Prompt caching can cut repeated input costs by up to 90%, and batch processing cuts costs by about 50% for workloads that can wait.
The site says most cost-estimate articles ignore these two discounts, making their figures run 2 to 3 times too high or too low.

Why it matters

AI API pricing is fragmented across dozens of providers and hundreds of models, each with its own per-token input and output rates that change over time. Anyone budgeting a chatbot, an agent, or a production API integration has to translate raw per-token prices into a realistic monthly bill, and the post argues that most attempts to do this by hand or from generic articles miss two discounts, prompt caching and batch processing, that can swing the estimate by a factor of 2 to 3.

Who it affects

Developers and teams building chatbots, autonomous agents, or any product that calls AI APIs at volume, since they are the ones who need to forecast token spend before committing to a provider or model.

How to use it

The site, at costperprompt.com, presents a live pricing table covering 232+ models, updated automatically, with the flagship pricing table shown as last refreshed on 2026-08-02. Its calculators take a workload description, such as chatbot, agent, or API usage, and convert per-token input and output rates into a projected monthly cost, factoring in prompt caching (up to 90% off repeated input) and batch processing (about 50% off) where they apply. No pricing or account requirement for using the site itself is mentioned in the text.

How solid is it

This is a Show HN self-submission with 13 points and 5 comments at the time of writing, so outside verification is thin. The text does not disclose who built the site, what company or team stands behind it, or the mechanism by which pricing data is 'refreshed automatically', whether that means live provider APIs, scraping, or manual updates.

Risks and caveats

The post gives no author or company attribution and no methodology for how pricing stays current, so the accuracy and maintenance of the 232+ model count cannot be checked from the text alone. The only concrete pricing figure given is a generic $5 / $25 per million token example rather than a named model or provider, and the claim that other cost articles run 2 to 3 times off is the site's own framing rather than an independently verified comparison.

“Output is usually 3-5× more expensive than input.”

— CostPerPrompt's pricing explainer