Claude API pricing per million tokens

Claude API prices for current models

Anthropic API prices for the four current Claude models are below, in USD per million tokens (MTok), with prompt caching rates.

All four have a 1M-token context window and 128K max output. Haiku 5.5's prices step up once a prompt goes over 100,000 tokens. For how Sonnet and Opus differ beyond price, see Claude Sonnet vs Opus.

Older Claude models still on the API

If you're still pinned to an earlier model, here are its rates per million tokens.

Sonnet 5's $2 input and $10 output introductory price is now its standard price, because the increase to $3 and $15 scheduled for September 1, 2026 will not happen. Opus 4.1 and Opus 4 are retired on the direct API: Opus 4.1 remains on Amazon Bedrock and Google Cloud (Google-managed), and Opus 4 only on Google Cloud (Google-managed). Opus 5.5's per-token prices are 20% lower than Opus 5's, and Anthropic says it costs 40% less than Opus 5 on typical workloads.

Prompt caching prices

• A 5-minute cache write costs 1.25x the model's base input price.
• A 1-hour cache write costs 2x the base input price.
• Cache reads cost 0.1x the input price on Haiku 5.5 and on earlier models.
• Cache reads drop to 0.05x the input price on Opus 5.5 and Sonnet 5.5, and down to 0.025x on Fable 5.1.

On Opus 5.5, for example, reading cached input tokens costs $0.20 per million instead of the standard $4 per million input rate.

Batch, fast mode and regional pricing

• The Batch API is 50% cheaper than standard requests, and you can combine batch discounts with prompt caching discounts.
• Fast mode for Opus 5.5 runs up to 2.5x faster and bills at $8 per million input tokens and $40 per million output tokens.
• US-only inference (when inference_geo is set to us) is billed at 1.1x the standard rate.
• On Amazon Bedrock and Google Cloud, regional and multi-region endpoints cost 10% more than global endpoints for Sonnet 4.5, Haiku 4.5, Opus 4.5, and later models.
• Claude 4.7 and later models use a tokenizer that produces about 30% more tokens for the same text, which raises the cost of running the exact same prompt compared to older versions.

Tool and agent charges

• Web search costs $10 per 1,000 searches.
• Web fetch has no extra charge beyond the tokens used.
• Code execution is free when used with web search or web fetch, or $0.05 per hour per container after a free allowance when used on its own.
• Claude Managed Agents bill tokens at the model's standard rates plus $0.08 per session-hour of runtime.

API usage is paid from prepaid credits in USD unless your organization is invoiced through Anthropic sales; the Claude API key guide covers setup. New users receive a small amount of free Claude API credits to test the API before paying.

API billing or a subscription

The direct Claude API bills for every token you send and receive. If you pay for a consumer plan, Claude Max includes $100 (on Max 5x) or $200 (on Max 20x) a month in API credits, while Claude Pro includes none (see our full Claude API vs subscription cost breakdown).

If you'd rather not pay per token, Plan2API turns a Claude Pro or Max plan you already pay for into an OpenAI-compatible p2a_ key. It never bills per token; your plan's own usage limits apply instead (see Claude subscription API). It works with the Chat Completions format only (POST /v1/chat/completions, not Anthropic's /v1/messages), with model ids such as claude/claude-opus-5-5 and claude/claude-sonnet-5. You can connect an account on the Free plan with no card.