How Claude billing works
Claude charges for input tokens and output tokens at separate rates, with output priced higher. Anthropic also offers prompt caching, which can reduce the cost of repeated prompt content, and a batch option for asynchronous workloads. Rates vary by model and change over time, so we link the official pricing page rather than reprinting figures that could go stale.
The mental model: your bill ≈ (input tokens × input rate) + (output tokens × output rate) × monthly requests. Claude's long context makes it easy to send large inputs — which is powerful, but also where costs can quietly grow.
Choosing a Claude model
Anthropic offers a spread of Claude models, generally from smaller, faster, lower-cost options up to larger, more capable ones. As with any provider there is no single "best" Claude model — match the model to the job:
- Small, fast models: strong fit for high-volume, latency-sensitive tasks and simple transformations.
- Mid-tier models: a common default for assistants, drafting, and summarisation where quality matters but budget still counts.
- Largest models: worth their premium for the hardest reasoning, analysis, and long-document work where mistakes are expensive.
Long context and prompt caching
Claude's large context windows let you pass entire documents or long histories in one call. That is convenient, but every token in the window is billed as input, so a "just send everything" habit can be the biggest hidden cost in a Claude deployment.
- Retrieve, do not dump: send only the relevant passages instead of whole documents where you can.
- Use prompt caching for stable prefixes (system prompt, policy text, reference material) that repeat across requests.
- Cap output length so long-form generations do not balloon the pricier output side of the bill.
Worked planning scenario (illustrative)
Illustrative assumptions — not a quote: a contract-review assistant processing 4,000 documents/month, each ~8,000 input tokens of contract text plus a ~1,500-token instruction prompt, returning ~800 output tokens of structured findings.
That is roughly 38M input and 3.2M output tokens per month. The instruction prompt repeats on every request, so it is an ideal caching candidate; the contract text does not repeat, so the lever there is retrieval and chunking rather than caching. Plug your own figures into the AI API Cost Calculator to see the monthly range with current Claude rates.
How to confirm real numbers
Measure the real token counts your prompts produce during a prototype, then multiply by current rates from Anthropic's official pricing page. Because long-context habits and traffic both vary, plan a budget range rather than a single number and revisit it as your prompts evolve.
Pricing and capabilities change. Provider rates, model line-ups, and features are updated by the providers over time. Any figures on this page are clearly labelled illustrative planning assumptions, not quotes. Official sources are checked on a recurring schedule. Unusual or structural changes are held for human review. See the Pricing Sources & Data Status page for verification dates, and always confirm current rates on the official provider pages below.
Official sources
Confirm current pricing and capabilities directly with each provider: