AI//COST

AI//COST is an independent pricing desk for people who need to understand what large language models actually cost to use. It brings model rate cards, provider pages, comparisons, calculators, and pricing history into one place. That matters because a headline price per million tokens rarely tells the whole story. It also helps keep consumer subscription prices separate from usage-based API rates. Input and output have different rates, cached input may be much cheaper, batch processing can change the bill, and reasoning models may charge for tokens a user never sees. The site turns those moving parts into comparable figures without asking readers to build a spreadsheet for every provider.

The main pricing index covers hundreds of models across dozens of providers. Each model page can show input, output, and cached rates, a context window, status information, and the date the price was last checked. Provider hubs make it easier to scan one vendor's lineup, while two-way comparisons put competing models beside each other. AI//COST also keeps price history, so a current rate isn't presented as though it has always been the rate. Search runs across the index, comparisons, and supporting research, which helps when a reader knows a model name but not where it sits in the broader market.

The calculators make the directory more useful than a static table. The LLM API cost calculator estimates monthly spend from request volume, token counts, cache hit rates, batch tiers, and reasoning tokens. A token counter shows how the same text can produce different counts and input costs across models. There are also live tools for comparing subscriptions with API usage and for estimating text to speech costs. Scenarios can be adjusted rather than accepting one hidden workload assumption, and the main calculator supports shareable scenarios and CSV export. Several other calculators are clearly marked as planned, so readers can distinguish shipped tools from the roadmap.

Its methodology is unusually visible. AI//COST says rates are read manually from each vendor's canonical pricing page, recorded with a verification date, and propagated through dependent pages and calculators. The formulas behind calculated figures are published, and default assumptions are shown in the interface. Token counts carry accuracy labels that separate exact browser-side counting from calibrated or estimated results. Benchmark scores come from named third-party leaderboards rather than being presented as the site's own testing. Missing data is left blank, changes can enter a public corrections log, and the methodology itself is versioned.

The natural audience is developers, founders, finance owners, and platform teams evaluating AI spend. It can help a developer shortlist affordable models before a proof of concept, or help a team test how caching and response length change a production budget. It is also useful when comparing a consumer subscription with pay-as-you-go API access, since those purchasing models aren't directly interchangeable. Editorial comparisons, capability answers, agent cost pages, and a glossary add context for readers who need more than a raw rate, though anyone making a purchase should still confirm the linked vendor page before committing.

What sets AI//COST apart is the connection between the source rate, the calculation, and the editorial explanation. A price is tied to a vendor page and a date, then reused by calculators and comparisons instead of being copied into separate pages by hand. That structure reduces the chance that two parts of the site quietly disagree. The desk also states its limits. Prices use a USD baseline, country-specific tax and localized pricing aren't currently published, and neutral benchmark coverage can be incomplete. Those caveats make the figures easier to interpret because readers can see what has and hasn't been measured.

Access is free. The calculator page says there is no signup and no paywall, and the main product is the public pricing desk and its browser-based tools. The site discloses affiliate links and visually marked sponsorships, while saying those relationships don't determine editorial rankings. Readers get substantial value without creating an account, but they should pay attention to each page's last verified date because provider prices and model availability can change quickly. For teams that want a transparent starting point for model cost research, AI//COST offers a strong public reference with enough detail to move from a list price to a workload estimate.