Claude API Pricing 2026: Current Models, Caching & Batch.

Quick Answer: Anthropic charges for the Claude API per million tokens (MTok), with separate prices for input and output. As of October 7, 2026: Claude Haiku 4.5 is $1 in / $5 out, Claude Sonnet 5.5 is $2 / $10, Claude Opus 5.5 is $4 / $20, and Claude Fable 5.1 is $10 / $50. The Batch API halves both prices, and cached input is billed at a fraction of the normal input price. Prices below come from Anthropic's official pricing page, read on October 7, 2026.

We build software on the Claude API every week, including our own Claude integration case study, so we keep this page current for clients who need to budget an AI feature before they build it. Prices change, so check the date at the top and the official page before you sign off a budget.

Claude API prices for current models

These are Anthropic's first-party API rates. They also apply to Claude on Microsoft Foundry. Amazon Bedrock and Google Vertex AI set their own prices.

MTok means one million tokens. A token is a small chunk of text, a word or part of a word, so a million tokens is a lot of text. Every request is billed for the tokens you send (input) and the tokens Claude writes back (output), and output costs five times as much as input on every model in the table.

How prompt caching changes the bill

If many requests start with the same text, such as a long system prompt, product catalog or set of instructions, you can cache it. The first request pays a little more to write it to the cache. Later requests that reuse it within the cache window pay the much lower cache-read price instead of the full input price.

Anthropic's multipliers, relative to the normal input price:

• 5-minute cache write: 1.25×
• 1-hour cache write: 2×
• Cache read: 0.1× on most models, 0.05× on Claude Opus 5.5 and 0.025× on Claude Fable 5.1

Because a cache read costs a tenth of the input price (or less), a 5-minute cache pays for itself after one reuse.

Batch API: half price if you can wait

The Batch API processes requests asynchronously, not in real time, and takes 50% off both input and output. It suits work nobody is waiting on, like overnight document processing, bulk classification or back-filling summaries. Batch and caching discounts can be combined. Batch prices per model:

Other things that show up on the invoice

• Long context costs no extra. Claude 4.6 and later models bill a 900,000-token request at the same per-token rate as a 9,000-token one.
• Web search is $10 per 1,000 searches, plus the tokens the results add.
• US-only processing (the inference_geo setting) multiplies all token prices by 1.1× on Claude 4.6 and later models. The default global routing uses the standard prices.
• Fast mode on Claude Opus 5.5 is $8 in / $40 out, twice the standard price, for faster responses.
• Free credits: new API accounts get a small amount of free credit to test with.

What real workloads cost

Each example uses the prices above. Change the numbers to match your own volume.

A customer-support chatbot on Claude Sonnet 5.5. 10,000 conversations a month, each sending 2,000 input tokens and getting 500 output tokens back:

• Input: 20 million tokens × $2 = $40
• Output: 5 million tokens × $10 = $50
• About $90 a month. If 1,500 of those 2,000 input tokens are the same system prompt on every request and are cached, they bill at $0.20 instead of $2: 15M × $0.20 + 5M × $2 + $50 = about $63 a month, plus a small charge for writing the cache.

Summarizing 1,000 long documents on Claude Haiku 4.5 through the Batch API. Each document is 20,000 input tokens and produces a 1,000-token summary:

• Batch input: 20 million × $0.50 = $10
• Batch output: 1 million × $2.50 = $2.50
• $12.50 total, versus $25 at the normal real-time price.

An AI agent on Claude Opus 5.5. 500 tasks a month, each reading 50,000 tokens and writing 5,000:

• Input: 25 million × $4 = $100
• Output: 2.5 million × $20 = $50
• About $150 a month. The same workload on Claude Sonnet 5.5 is about $75. Whether the cheaper model holds up is something to measure on your own tasks, not assume.

Which model should you budget for?

Price per token is only half the picture. What matters is the cost per finished job: a cheaper model that needs more retries or more human review can cost more overall. A practical way to start:

• Claude Haiku 4.5 for high-volume, simple work: tagging, routing, short extractions.
• Claude Sonnet 5.5 for most business features: chat, drafting, everyday coding and tool use.
• Claude Opus 5.5 when the task is complex and mistakes are expensive: long multi-step agents, hard reasoning.
• Claude Fable 5.1 for the most demanding work, at 2.5× the Opus 5.5 price.

Run a sample of real requests through two models, compare the quality and the measured cost, and pick from data.

Need help putting Claude to work?

We design, build and run AI features on the Claude API, from a single integration to full agents, and we can estimate the running cost with you before anything is built. If you'd like that done for your product:

• AI integration services: add Claude to the CRM, helpdesk or database you already run.
• AI development services: custom AI agents and assistants, built and supported.
• AI automation agency: put agents on repetitive back-office work.

Want to see what this looks like in a real product? Read our Claude integration case study, or learn more about the people who built Claude.

Prices from Anthropic's official pricing page, read October 7, 2026. Prices can change at any time; the official page is the source of truth.