AI Automation Cost Calculator

Prompt caching: if the same large block of context (e.g. a long contract template or policy document) is sent repeatedly, Anthropic caches it and charges a fraction of the normal input rate for subsequent calls. Real costs could be lower than the calculator shows for high-repetition workflows.

Batch API: asynchronous batch jobs (not real-time) are discounted by 50%. Relevant if your automation doesn't need instant responses.

The Gemini estimates above use standard published rates. Google applies additional charges in certain situations. Requests involving very long documents (above roughly 200,000 tokens) are billed at a higher rate; the Deep Think reasoning mode carries a premium on top of standard output rates; and the Search Grounding feature - which lets the model pull in live web results - adds a small per-query fee.

Context caching, which can reduce costs for repeated use of the same large document, is not reflected here either. For straightforward business workflows of the kind this calculator covers, standard rates are a reasonable starting point, but if you expect to use any of these advanced features regularly, check Google's current pricing page for the full breakdown.

Cached inputs: OpenAI applies automatic prompt caching in a similar way to Claude. Repeated context portions are cheaper.

Batch API: same 50% discount as Claude for non-real-time processing.

The model fees come straight from each provider's published rates, with FX rates listed in the assumptions panel. The token sizes per task are based on typical real-world examples for each workflow and complexity level. Treat the result as a strong starting estimate, not a contract. Real invoices move with actual usage, but the order of magnitude is reliable.

Real usage is rarely the same every month. The range reflects normal variation around the typical token sizes for the workflow and complexity you've chosen. The per-task figure shows the midpoint, useful for back-of-envelope thinking.

Only the model fee. Not included: prompt caching discounts, batch API discounts (typically 50% off for non-real-time work), Gemini's premium features such as Deep Think, Search Grounding and long-context escalation, third-party data sources, hosting and infrastructure, retries, and any human-review step in the workflow.

The seven workflows shown cover the patterns we see most often. If your requirement is broadly similar in shape and volume, the closest match will still give you a useful estimate. If it is genuinely different, please get in touch and we'll be happy to discuss your project with you.

The verified date shown below the calculator tells you when the rates were last checked against each provider's published pricing page. Rates are reviewed on a regular cadence and the calculator republished. If the date looks old, the underlying providers may have moved since.

The Recommended tag flags the model that best fits the complexity you've chosen, per provider. Beyond that, the right choice depends on factors the calculator cannot see, including your existing setup, your data security requirements, and the level of judgement the work demands. The calculator surfaces the cost. The choice stays yours.

These are the three commercial providers we see most often in real-world deployments. Open-source models (Llama, Mistral) and other commercial options are real choices, but they have different cost structures (typically infrastructure-based rather than per-token) and are not directly comparable in this format.

Cheapest is often the right answer for genuinely simple, high-volume tasks. For work that requires judgement, multi-step reasoning, or nuance, the savings on a smaller model can disappear in extra retries, escalations, or output you have to fix.

Each workflow tier maps to a realistic input and output size for that level of complexity. A Quick Review of a short contract uses fewer tokens than a Comprehensive Review of a multi-document agreement. The full breakdown is in the assumptions panel under the calculator.