Claude Sonnet API cost: the short answer
For the Anthropic Claude API, Claude Sonnet 5.5 is listed at $2 per million input tokens and $10 per million output tokens. Anthropic lists those rates in USD, with input and output billed separately. Its pricing documentation lists Sonnet 4.6 at $3/$15 per million input/output tokens, so a cost estimate needs to name the Sonnet version rather than treating “Claude Sonnet” as one rate. These are Anthropic API list rates, not a quote for every provider or cloud channel. Anthropic pricing documentation
For Sonnet 5.5, Anthropic’s Sonnet page reports a 1.1× price multiplier for US-only inference. Other providers and cloud channels may have different billing terms, so check their rate cards separately. Anthropic’s Sonnet page
Claude Sonnet 5.5 API rates
Anthropic’s pricing documentation lists these Sonnet 5.5 rates in USD per million tokens (MTok). Anthropic pricing documentation
| Billable usage | Price per 1 million tokens |
|---|---|
| Standard input | $2.00 |
| Standard output | $10.00 |
| 5-minute cache write | $2.50 |
| 1-hour cache write | $4.00 |
| Cache read | $0.10 |
A cache write and a cache read are separate billable categories. If you use prompt caching, price each category separately; don’t apply the standard input rate to every input token. Anthropic’s Messages API reports cache creation and cache-read token counts in response usage data. Messages API reference
How Sonnet version affects the estimate
| Claude API model | Input per MTok | Output per MTok |
|---|---|---|
| Sonnet 5.5 | $2 | $10 |
| Sonnet 5 | $2 | $10 |
| Sonnet 4.6 | $3 | $15 |
| Sonnet 4.5 | $3 | $15 |
Check that your request uses the intended model and that your API channel bills at Anthropic’s rates before using these figures for a budget. Anthropic pricing documentation
Estimate the cost of a request
For ordinary, uncached input and output, use:
Request cost = (input tokens ÷ 1,000,000 × input price) + (output tokens ÷ 1,000,000 × output price)
For Sonnet 5.5, that is:
(input tokens ÷ 1,000,000 × $2) + (output tokens ÷ 1,000,000 × $10)
Worked example: one request and a monthly scenario
Suppose one request uses 3,000 input tokens and returns 800 output tokens, with no caching:
- Input: 3,000 ÷ 1,000,000 × $2 = $0.006
- Output: 800 ÷ 1,000,000 × $10 = $0.008
- Total: $0.014 per request
If every request had that same token usage, 10,000 requests would cost $140, before any eligible credits or other provider charges. This is an illustrative scenario, not a forecast: actual cost depends on request volume, token counts, model, and billing channel.
Worked example: repeated context with prompt caching
Suppose an application sends the same 50,000-token prefix on 20 requests, and the later requests reuse a valid cache entry. Pricing just that repeated prefix:
- Without caching: 20 × 50,000 = 1,000,000 input tokens; 1,000,000 ÷ 1,000,000 × $2 = $2.00.
- With a 5-minute cache write and 19 cache reads: write: 50,000 ÷ 1,000,000 × $2.50 = $0.125; reads: 950,000 ÷ 1,000,000 × $0.10 = $0.095; total = $0.22.
That is $1.78 less, or 89% lower, for the repeated prefix in this example. It does not price any other input or output tokens. The result depends on the prefix being cacheable and reused while the cache entry is valid. Anthropic says the default cache lifetime is five minutes and that reuse refreshes the entry without a separate refresh fee; cache-read tokens are still billed at the cache-read rate. Prompt-caching documentation
Measure production usage before forecasting
Use representative requests from your workload, not an assumed average alone. The Anthropic Messages API reference documents the POST /v1/messages endpoint. For a workload using only five-minute caching, record the model, input and output counts, cache_creation_input_tokens, and cache_read_input_tokens from each response, then price each category at its matching rate. Anthropic documents cache_creation_input_tokens as the total of the values in the cache_creation object. Messages API reference · Prompt-caching documentation
This measurement recipe is scoped to five-minute caching. If your workload uses one-hour caching or mixes cache lifetimes, don’t apply one aggregate cache-creation count to a single write rate; use the documented lifetime-specific breakdown, and verify how its subtotals relate to the aggregate before calculating. If you can’t establish that breakdown, don’t produce a mixed-lifetime estimate from the aggregate alone.
For a monthly estimate, multiply measured per-request costs by the expected number and mix of requests; use separate scenarios if traffic or token lengths vary substantially. A calculator that ignores cache writes or reads can understate a cache-enabled workload. Compare your calculation with your organization’s billing records.
Does a Claude subscription cover API use?
Anthropic documents monthly Claude API credits for eligible Max, Ultra, and Team subscribers: $100 for Max, $200 for Ultra, and $20 per Team Standard seat or $100 per Premium seat, pooled up to $500 per month for a Team organization. New subscribers can claim credits after seven days on an eligible plan. Confirm your plan’s eligibility and the current claim or activation requirements before subtracting credits from a budget. Anthropic’s API-credit documentation
The credits apply to Claude API use in the Claude Console, not to Claude Platform on AWS, Amazon Bedrock, Google Cloud, or Microsoft Foundry. Credits refresh each billing cycle (monthly for annual plans) and expire at the end of that cycle without rollover. Estimate token-metered API costs first, then subtract only credits your organization is eligible to claim for its actual channel. Anthropic’s API-credit documentation
Bottom line
For the Anthropic API, the listed standard rates for Sonnet 5.5 are $2 per MTok input and $10 per MTok output. Price cache writes and reads separately when caching is active, and use a lifetime-specific breakdown if you mix cache lifetimes. Measure representative production requests, confirm the model and billing channel, and check Anthropic’s pricing and API documentation before committing a budget because rates and terms can change.
Sources
Pricing — Claude Platform Docs
Messages — Claude API Reference
Prompt caching — Claude Platform Docs
API credits for Max, Ultra, and Team plans — Claude Platform Docs