Claude API Pricing (October 2026): Per-Model Token Costs

Short answer: the Claude API charges per million tokens (MTok), with separate prices for input and output. As of October 17, 2026, the current models cost Haiku 5.5 $0.10 / $0.50 (prompts up to 100K tokens), Sonnet 5.5 $2 / $10, Opus 5.5 $4 / $20 and Fable 5.1 $10 / $50. You can cut costs with the Batch API (50% off) and prompt caching (cache reads cost 2.5%-10% of the input price). There's no subscription or minimum spend.

This page is for developers who call the API directly. If you want the Claude chat apps or a coding agent with a flat monthly price, see Claude pricing and Claude Code pricing.

Current model prices

All prices are per million tokens, in USD, from Anthropic's pricing documentation. Mythos 5.1 and Mythos 5 are available only to organizations verified through Anthropic programs such as the Cyber Verification Program.

Older models still on the API

Sonnet 5's $2 / $10 price was announced at launch as an introductory rate through August 31, 2026. Anthropic has since made it the standard price, and the planned increase to $3 / $15 on September 1, 2026 won't happen. Opus 4.1, Opus 4, Sonnet 4 and Haiku 3.5 are retired on the Claude API. Some remain on Bedrock or Google Cloud, at $15 / $75 for Opus 4.1 and Opus 4, $3 / $15 for Sonnet 4 and $0.80 / $4 for Haiku 3.5.

How each pricing modifier works

Prompt caching

Caching stores a repeated prompt prefix, such as a system prompt, a document or the conversation so far, so later requests read it cheaply. A 5-minute cache write costs 1.25x the base input price. A 1-hour write costs 2x. A cache hit costs 0.1x on most models, 0.05x on Opus 5.5 and Sonnet 5.5, and 0.025x on Fable 5.1 and Mythos 5.1. Each hit also refreshes the cache. With the standard 0.1x hit price, a 5-minute write pays for itself after one cache hit, and a 1-hour write after two. You can turn on caching automatically with one top-level `cache_control` field, or place breakpoints yourself.

Batch API

The Message Batches API processes requests asynchronously at 50% off both input and output. It stacks with caching and data residency multipliers. Use it for jobs that don't need an instant answer, such as classification, evaluations, summarizing archives and backfills. Fast mode isn't available in batch.

Long context

Claude 4.6 and later models (except Haiku 5.5) include the full 1M-token context window at standard rates. Anthropic's example is that a 900K-token request costs the same per token as a 9K-token request. Haiku 5.5 is the exception. Any request whose prompt is over 100,000 tokens pays the higher prices ($0.50 / $2.50). The prompt length includes cache reads and writes, and the higher rate applies to the whole request.

Fast mode

Fast mode gives significantly faster output (Anthropic's pricing page says up to 2.5x) at premium rates: Opus 5.5 $8 / $40, and Opus 5 and Opus 4.8 $10 / $50 (research preview). Caching and data residency multipliers stack on top. It runs on the Claude API, and for Opus 5.5 on Microsoft Foundry. It isn't available on Claude Platform on AWS, Bedrock or Google Cloud.

Data residency

Setting `inference_geo: "us"` on Claude 4.6 and later models applies a 1.1x multiplier to every token category, including cache reads and writes. For example, Sonnet 5.5 becomes $2.20 / $11. Global routing, the default, uses standard prices. On older models, the parameter returns an error.

Tokenizer change

Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text. The exact increase depends on the content. Before migrating, compare costs with the token counting endpoint rather than comparing per-token prices alone.

Tool and platform fees

Worked cost examples

Each example uses list prices from the tables above. Formula: (tokens / 1,000,000) x price per MTok. The volumes are illustrative, so plug in your own.

1. Support chatbot: Haiku 5.5 vs Sonnet 5.5

1,000,000 conversations a month, each with 1,500 input tokens and 300 output tokens, gives 1,500 MTok of input and 300 MTok of output. Haiku 5.5: (1,500 x $0.10) + (300 x $0.50) = $150 + $150 = $300/month. Sonnet 5.5: (1,500 x $2) + (300 x $10) = $3,000 + $3,000 = $6,000/month. Test whether Haiku's answers are good enough before paying 20x more.

2. Document Q&A with prompt caching (Sonnet 5.5)

You send a 50,000-token reference document with each of 10,000 questions a day. Without caching: 10,000 x 50,000 = 500 MTok x $2 = $1,000/day in document input. With a 1-hour cache refreshed by steady traffic, assume one write per hour: 24 x 0.05 MTok x $4 = $4.80. The other 9,976 requests are hits: 9,976 x 0.05 MTok = 498.8 MTok x $0.10 = $49.88. Total: about $54.68/day, roughly 95% less. Questions and answers are billed at normal rates on top.

3. Overnight batch job (Opus 5.5)

Summarizing 100,000 reports at 3,000 input and 500 output tokens each gives 300 MTok in and 50 MTok out. Standard: (300 x $4) + (50 x $20) = $1,200 + $1,000 = $2,200. Batch API: (300 x $2) + (50 x $10) = $600 + $500 = $1,100.

4. Haiku 5.5 just over the 100K line

A 90,000-token prompt with 2,000 output tokens costs (0.09 x $0.10) + (0.002 x $0.50) = $0.009 + $0.001 = $0.010. A 150,000-token prompt pays the higher rates on everything: (0.15 x $0.50) + (0.002 x $2.50) = $0.075 + $0.005 = $0.080. That's 8x more for under 2x the tokens. If you can split long inputs into chunks under 100K tokens, Haiku stays cheap. Otherwise, compare against Sonnet 5.5, which has no long-context surcharge.

5. Agent session on Managed Agents

This is Anthropic's own example: a one-hour Opus 5.5 session with 50,000 input and 15,000 output tokens costs $0.20 + $0.30 + $0.08 runtime = $0.58. If 40,000 of those input tokens are cache reads, it costs $0.04 + $0.008 + $0.30 + $0.08 = $0.428.

6. Migrating from Sonnet 4.6 to Sonnet 5.5

Text that is 1 million tokens on Sonnet 4.6 becomes about 1.3 million on Sonnet 5.5 because of the newer tokenizer. Input cost changes from 1 x $3 = $3.00 to 1.3 x $2 = $2.60. The same logic applies to output: 1 x $15 = $15 becomes 1.3 x $10 = $13. The move still saves money, but less than the headline price drop suggests.

Which model should you use?

• Haiku 5.5: high-volume classification, extraction, routing and short chat, as long as prompts stay under 100K tokens.
• Sonnet 5.5: Anthropic's recommendation for most production workloads, including coding and agents, at $2 / $10.
• Opus 5.5: the most complex reasoning and agentic coding. It costs 2x Sonnet 5.5 but is cheaper than older Opus models ($5 / $25).
• Fable 5.1: long-running agents where capability matters more than cost. Its 0.025x cache hit price helps a lot in agent loops with large repeated context.

How to cut your Claude API bill

1. Route easy requests to Haiku and hard ones to Sonnet or Opus, instead of sending everything to one large model.
2. Cache stable prefixes such as system prompts, tool definitions and reference documents. Use the 1-hour TTL when traffic has gaps of more than five minutes.
3. Move anything that can wait up to a day into the Batch API for 50% off.
4. Cap output with max_tokens and ask for concise formats. Output costs 5x input on every current model.
5. Use web fetch (no fee) instead of web search ($10 per 1,000) when you already know the URL.
6. Set workspace spend limits in the Claude Console, and track usage on the Usage page.
7. For high volumes, ask Anthropic sales about volume discounts, which are negotiated case by case.

Free credits and payment

Anthropic's pricing FAQ says new users get a small amount of free credit to test the API. The amount isn't published. After that, you buy prepaid credits under Settings > Billing in the Claude Console by credit card. Enterprise customers can arrange invoicing. Max subscribers get $100 or $200 a month in API credits, and Team plans get $20 per Standard seat and $100 per Premium seat, pooled and capped at $500 a month. These credits expire at the end of each billing cycle. For a step-by-step setup, see how to get an Anthropic API key. Startups backed by a partner VC may get credits through Claude for Startups; direct applications are closed as of October 2026. Our free AI API keys guide lists providers with ongoing free tiers.

Cloud platforms

Amazon Bedrock and Google Cloud (Vertex AI) set and invoice their own Claude prices. From the Claude 4.5 generation onward, their regional and multi-region endpoints cost 10% more than global endpoints. Claude Platform on AWS and Claude in Microsoft Foundry are run by Anthropic and bill standard Claude API rates through the AWS or Azure Marketplace, in Claude Consumption Units: 100 CCU = $1, with no prepaid credits. Monthly subscriber API credits don't apply on any of these platforms.

Claude vs OpenAI and Gemini API prices

Competitor prices come from our OpenAI API pricing and Gemini API pricing pages. Gemini also has a free API tier on some models, which Claude doesn't offer. List prices are close across providers, so tokenizer differences, caching rules and long-context surcharges usually matter more than the headline rate.