Claude API Pricing Calculator

1. Pick a model. The tool fills the input, output, cache write and cache read prices from the official list. Every price stays editable.
2. Enter your input tokens and output tokens per request, then your requests per month. Take these from the usage numbers your own API responses report.
3. Optionally set a cached share, a cache hit rate, Batch or US-only. Read the monthly cost and the saving against list price.

Here is what each result means:

Claude API pricing is a pay-per-token model: you pay one price for the tokens you send and a higher price for the tokens the model writes back. Prices are quoted in US dollars per million tokens, which the vendor writes as MTok. There is no monthly fee for API access itself.

A token is a small piece of text. Anthropic says a token is about 3.5 English characters, and that the exact number varies by language. That figure is an approximation, not a billing rule. Your bill uses the token counts reported in each API response.

This is a different thing from the app plans. The pricing page lists monthly app plans in one section and API token prices in another. This tool covers API token prices only. This page is an independent estimator and is not affiliated with Anthropic.

The cost of one request is input tokens times the input price plus output tokens times the output price, divided by one million.

1. Multiply input tokens by the input price per million tokens.
2. Multiply output tokens by the output price per million tokens.
3. Add the two and divide by 1,000,000 to get the cost of one request.
4. Multiply by requests per month. Halve the result for the Batch API. Multiply by 1.1 for US-only inference where it is offered.

With prompt caching, the cached part of the input is priced differently. A request that reads the cache pays the cache read price. A request that misses pays the cache write price. The tool splits the cached tokens by your hit rate: cached tokens x (hit rate x read price + miss rate x write price).

Official Claude API Prices, Read 2026-10-02

These are the list prices in USD per million tokens for models sold on the Claude API on that date. The cache write column is the 5-minute price.

The official table also lists two limited-availability models and four retired models. They are left out of the picker. Use the Custom option to price any model by hand.

A support chatbot sends 2,000 input tokens and gets 500 output tokens per request. It runs 100,000 requests a month on a model priced at $2 input and $10 output.

Calculation: input = 2,000 x $2 / 1,000,000 = $0.004. Output = 500 x $10 / 1,000,000 = $0.005. One request costs $0.009. The month costs $0.009 x 100,000 = $900, and the year costs $10,800.

Output is only a fifth of the tokens, yet it is 55.6% of the bill. That is the effect of the higher output price.

Add caching: cache 75% of the input with a 90% hit rate. The blended price of a cached token is 0.9 x $0.20 + 0.1 x $2.50 = $0.43 per million, so the 1,500 cached tokens cost $0.000645 per request. One request now costs $0.006645 and the bill falls to $664.50. That is a saving of $235.50, or 26.2%. Switch the same job to Batch and it drops to $332.25.

Model Choice

The model sets both prices. On the list read on 2026-10-02, input prices run from $1 to $10 per million tokens and output prices from $5 to $50. The same workload can cost ten times more on the top model than on the cheapest one.

Output Length

Output tokens cost five times the input price on every current model in the table. Shorter replies cut cost faster than shorter prompts.

Prompt Caching

Caching reprices repeated input. A cache read costs 0.1 times the input price on most models, 0.05 times on Opus 5.5 and 0.025 times on Fable 5.1. A 5-minute cache write costs 1.25 times and a 1-hour write costs 2 times. A prefix shorter than the model minimum is not cached at all. The minimum is 512 to 4,096 tokens depending on the model.

Batch Processing

The Batch API gives 50% off input and output tokens. Requests run asynchronously. Anthropic says most batches finish in under one hour, and a batch expires after 24 hours.

Tokenizer Version

Anthropic states that Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text. A prompt that counted 2,000 tokens on an older model counts roughly 2,600 on a newer one. Recount before you compare models.

Where Inference Runs

US-only inference adds a 1.1 times multiplier on Claude 4.6 and later models. Global routing is the default and uses standard prices. Regional endpoints on partner clouds carry their own pricing.

The model is the largest single cost choice. This table prices one workload on six models: 2,000 input tokens, 500 output tokens and 100,000 requests a month, at list price with no caching or batch.

Token counts differ between tokenizer versions, so a real migration will not match this table exactly. Other charges sit outside token prices. Web search is listed at $10 per 1,000 searches, and fast mode on Opus 5.5 is listed at $8 input and $40 output. To weigh the bill against what the feature earns, use the ROI calculator.

Before You Build

Price the feature before writing code. Estimate tokens per request, pick a model and check that the monthly cost fits the plan. The break-even calculator shows how many paying users cover it.

When You Choose a Model

Run the same workload through two or three models. A cheaper model that passes your tests can halve the bill. Speed matters too, and the token generation speed simulator shows how long a reply takes to stream.

When You Review a Bill

Enter last month's real token totals and compare the estimate with the invoice. A gap usually points to cache writes, tool use or a model you forgot was in use. Add the API line to your wider software spend with the AI tool stack cost calculator.

1. Forgetting the System Prompt

The system prompt, tool definitions and chat history are input tokens on every request. Cutting 1,000 tokens from a prompt sent 100,000 times a month saves $200 at $2 per million.

2. Treating Caching as Free

The first request pays the write price, which is above the normal input price. At a 10% hit rate, the worked example costs $940.50 instead of $900. A 5-minute cache breaks even near a 21.7% hit rate, and a 1-hour cache near 52.6%.

3. Guessing Tokens from Words

Character and word ratios are rough. Use the token counting endpoint, which Anthropic says is free to use, or read the usage numbers in real responses.

4. Reusing Counts from an Older Model

Newer models tokenize the same text into about 30% more tokens. Old counts understate the new bill.

5. Mixing Up Plans and API Prices

Monthly app plans and per-token API prices are listed separately on the pricing page. A plan price tells you nothing about an API bill.

6. Ignoring Tool Tokens

Tool names, descriptions and schemas count as input tokens. The pricing page also lists a tool use system prompt of about 290 tokens on recent Opus models.

The arithmetic is exact for the prices and token counts you enter. The estimate is only as good as those inputs.

It depends on the model. On the official list read on 2026-10-02, input prices range from $1 to $10 per million tokens and output prices from $5 to $50. Haiku 4.5 is $1 and $5, Sonnet 5.5 is $2 and $10, and Opus 5.5 is $4 and $20.

Claude Haiku 4.5 has the lowest list price at $1 input and $5 output per million tokens. The cheapest model for you is the lowest-priced one that still passes your own quality tests.

The pricing page lists per-token prices for every model and no free model. Token counting is the one listed endpoint that is free to use. Check your console for any credits on your own account.

There is no fixed number. Anthropic says a token is about 3.5 English characters, and newer models produce about 30% more tokens for the same text. Count your real text with the token counting endpoint for a figure you can budget on.

A cache read costs 10% of the input price on most models, so a fully reused prefix costs up to 90% less. The first write costs 25% more for a 5-minute cache. Anthropic notes that a 5-minute cache pays off after one cache read.

Yes. The pricing page states that caching multipliers stack with the Batch API discount. In the worked example on this page the two together cut a $900 month to $332.25.

No. The pricing page lists monthly app plans and per-token API prices in separate sections. This calculator covers the per-token API prices only.

Real requests vary in size, cache hits depend on timing, and tools such as web search add their own charges. Compare the token totals in your usage report with the numbers you entered here.

No, not on current models. The pricing page says Claude 4.6 and later models include the full 1M token context window at standard pricing.

The pricing page lists every price in USD. The currency result here multiplies the USD cost by a rate you type, so it is a planning figure and not an invoice amount.

No. The maths runs in your browser and nothing is sent to us. Save keeps a result only in this browser.

• Claude API pricing: model, caching, batch and data residency prices (Anthropic documentation, read 2026-10-02).
• Claude pricing page (Anthropic, read 2026-10-02; API prices per million tokens).
• Prompt caching: cache lifetime and minimum cacheable length (Anthropic documentation, read 2026-10-02).
• Batch processing: 50% discount and 24-hour expiry (Anthropic documentation, read 2026-10-02).
• Token counting: free endpoint and the newer tokenizer (Anthropic documentation, read 2026-10-02).