Claude API Cost Calculator

How the Claude API cost calculator works

This is a monthly bill estimator, not a token counter. You describe a workload (how many requests per day, how many input and output tokens each request uses, and roughly how much of the input is served from the prompt cache) and the calculator multiplies it out against Anthropic's published per-token pricing. As of July 2026 that is $5 input / $25 output per million tokens for Claude Opus 4.8, $3/$15 for Claude Sonnet 5 (with an introductory $2/$10 rate through August 31, 2026), $1/$5 for Claude Haiku 4.5, and $10/$50 for Claude Fable 5. Our Claude API key guide covers the full pricing table and rate-limit tiers.

The math is shown line by line: input cost plus output cost minus caching savings. Prompt caching bills repeated context at roughly a tenth of the input rate, which is why the cache hit rate slider moves the estimate so much. Coding assistants and long-running agents that resend the same system prompt and conversation history on every request routinely cache 60 to 90 percent of their input tokens; one-off requests cache nothing. Treat the hit rate as an approximation and check your real numbers in the Anthropic Console after a week of traffic.

The verdict block is deliberately honest. If your estimate is high and your usage is interactive, a flat-rate subscription such as Claude Max or Claude Pro is often the better deal, because the API bills per token with no ceiling. If your workload is programmatic, subscriptions are not an option and the levers are caching, the Batch API, and picking the cheapest model that still succeeds at the task; our Opus vs Sonnet comparison covers that decision in depth.

Frequently asked questions

How is the monthly cost calculated?

Requests per day times 30 gives monthly requests. Input tokens are billed at the model's input rate, output tokens at the output rate, and the cached share of input tokens is billed at roughly 0.1x the input rate. The total is input cost plus output cost minus caching savings. Batch API discounts (50 percent for asynchronous jobs) are not applied automatically, so batch-heavy workloads may come in lower than the estimate.

What do Claude models cost per million tokens?

As of July 2026: Claude Haiku 4.5 costs $1 input / $5 output per million tokens, Claude Sonnet 5 costs $3/$15 with an introductory $2/$10 rate through August 31, 2026, Claude Opus 4.8 costs $5/$25, and Claude Fable 5 costs $10/$50. Prompt caching and the Batch API reduce real-world bills substantially.

Is the API or a subscription cheaper?

It depends entirely on usage pattern. Interactive use (you, in Claude Code or claude.ai) is usually cheaper on a flat-rate plan once your API estimate passes roughly $100/mo, which is where Claude Max starts. Programmatic use (products, automation, agents) can only run on the API. The two billing worlds never mix: a $200 Max subscription includes zero API tokens, and API credits grant no app subscription.

What is a realistic prompt-cache hit rate?

For coding sessions and long-running agents that resend a stable system prompt plus growing conversation history, 60 to 90 percent of input tokens are commonly cache reads. For stateless request and response APIs with unique prompts, the rate is close to zero. When in doubt, start at 50 percent and refine with real Console usage data.

How accurate is this estimate?

It is a planning estimate, not a quote. Real bills vary with actual token counts, cache behavior, tool-use overhead, and retries. The per-token prices match Anthropic's published pricing as of July 2026, but always confirm current rates in the Anthropic Console before committing to a budget.