Pricing Methodology

Data Sources

Every price in the AIModelCalc calculator is sourced directly from the official pricing documentation of each provider. We do not use automated scrapers or third-party price aggregators as primary sources, because those introduce a lag and an additional point of failure.

For open-source models like Llama, there is no single canonical price because different inference providers charge different rates. We use a representative mid-market rate from the major inference hosts and note this in the calculator.

How We Handle Caching Discounts

Several providers offer prompt caching — where repeated portions of a prompt (like system instructions) can be cached and reused at a reduced per-token cost. The discount structure varies significantly by provider:

• Anthropic (Claude): Cache reads are priced at 10% of the standard input token rate
• OpenAI: Cache reads are priced at 50% of the standard input token rate
• Google Gemini: Context caching has a separate pricing tier with storage costs per hour

The AIModelCalc calculator includes a caching toggle that applies provider-appropriate cache discounts. The cache savings shown assume a cache hit rate of 100% on cached tokens — real-world hit rates will vary depending on how consistent your prompts are. Use the caching estimate as an upper bound on potential savings.

What Our Calculator Does Not Include

There are parts of AI API costs that our calculator doesn't model, either because they're too variable or because the data isn't publicly available:

• Enterprise volume discounts. Most providers offer negotiated pricing at high volumes. We don't have visibility into those rates and don't attempt to model them.
• Free tier usage. Several providers offer free monthly credits. Our estimates assume you've exhausted any free tier.
• Batch processing discounts. OpenAI's Batch API offers a 50% discount on requests that don't require real-time responses. Google and Anthropic have similar offerings. We note this where relevant but don't build it into the default calculation.
• Storage costs for context caching. Google charges separately for cached context storage per hour. For short-duration workloads this is negligible; for long-running context it can add up.
• Egress and infrastructure costs. Our calculator covers API token costs only — not the compute, network, or storage costs of running your application.

Update Frequency

We monitor provider pricing pages and developer communications for changes. AI pricing is volatile — providers change rates, deprecate models, and introduce new pricing tiers frequently. Our typical update timeline:

• Major pricing changes: Updated within 24–48 hours of official announcement
• New model launches: Added after stable pricing is confirmed (not during preview periods with unclear pricing)
• Model deprecations: Kept in the calculator with a deprecation note until the model is no longer accessible via API
• Corrections from readers: Addressed within 24 hours, typically faster

Reporting Errors

If you spot a pricing error — whether a stale rate, a wrong per-million figure, or a missing model — please report it through our contact page. Include a link to the provider's official pricing page showing the correct rate. We update quickly and appreciate the help keeping this accurate.