Anthropic publishes detailed Claude API pricing and feature matrix

Anthropic’s developer documentation now lays out a fuller price sheet for Claude API access, showing not just headline token rates but the add-on charges and discounts that affect production bills. The company’s pricing page, available in its Claude Platform documentation and crawled on May 22, says Claude models are billed in U.S. dollars per token category, with separate charges for input and output and additional pricing for prompt caching and batch processing. The pricing page says Claude Opus 4.7 starts at $5 per million input tokens and $25 per million output tokens. The same documentation lists Claude Sonnet 4.6 at $3 and $15 per million tokens, respectively, and Claude Haiku 4.5 at $1 and $5. Anthropic’s Opus 4.7 product page repeats the Opus 4.7 pricing and says the model is available through the Claude API. (platform.claude.com) ### Which Claude models are on the current pricing sheet? Anthropic’s documentation names Opus 4.7, Sonnet 4.6 and Haiku 4.5 in the current pricing matrix. The page presents them as the main model lineup in the API docs and breaks out pricing by base input tokens, cache-write categories, cache hits and output tokens. Anthropic’s Opus 4.7 page says the model is aimed at developers building higher-end AI applications and links back to the same pricing framework. (platform.claude.com) That page says Opus 4.7 is also offered on Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry, though the Claude Platform documentation distinguishes between Anthropic-operated and partner-operated cloud pricing. ### How does Anthropic say developers are billed? (platform.claude.com) Anthropic’s pricing page says all prices are in USD and separates billing into standard input, output and prompt-caching categories. The documentation says the “Base Input Tokens” column covers standard input pricing, while “5m Cache Writes,” “1h Cache Writes,” and “Cache Hits & Refreshes” apply to prompt caching. (anthropic.com) The same page says a cache hit costs 10% of the standard input price. Anthropic says that means caching pays off after one cache read for the five-minute duration, where writes are priced at 1.25 times standard input, or after two cache reads for the one-hour duration, where writes are priced at 2 times standard input. ### What other cost levers does the documentation spell out? (platform.claude.com) Anthropic’s product page for Opus 4.7 says prompt caching can reduce costs by as much as 90% and batch processing can cut costs by 50%. The pricing documentation also references batch processing and cloud-platform-specific pricing pages, indicating that the listed rates are part of a wider billing structure rather than a single flat menu. (platform.claude.com) The pricing page also says Opus 4.7 uses a new tokenizer compared with previous models and may use up to 35% more tokens for the same fixed text. Anthropic attributes that change to improved performance across a range of tasks. ### Where are the rate and feature details published? Anthropic’s pricing information appears on the Claude Platform documentation page titled “Pricing - Claude API Docs.” The page says developers seeking the most current pricing information should also check Claude’s broader pricing page, and it links to related documentation for prompt caching and partner cloud platforms. (anthropic.com) (platform.claude.com) May 22 search results show the pricing page live in Anthropic’s documentation, and Anthropic’s Opus 4.7 page points developers to use `claude-opus-4-7` through the Claude API. For developers comparing deployment options, Anthropic’s docs also direct readers to separate pricing pages for AWS and Microsoft Foundry. (platform.claude.com)