Home / Blog / Pricing

Pricing

Claude API Pricing in 2026: Every Model, Caching and Max Credits

Claude API pricing per 1M tokens, from Haiku 5.5 at $0.10 input to Fable 5.1 at $10, with prompt caching rates and when Max plan credits beat the API.

By Engineer Foodie Desk ✓ Sources checked Oct 19, 2026

7 min read

Anthropic’s pricing page is a spreadsheet with feelings. Six models, two Haiku bands, cache writes that cost more than the tokens you’re caching, and a Max plan that tosses in API credits like a tip jar. Here’s the plain version: every rate in US dollars per 1 million tokens, pulled from Anthropic’s official docs and checked on October 18–19, 2026.

Quick vocabulary. Input is what you send. Output is what Claude writes back, and it’s usually four to five times the input price. Cache write is the one-time fee to store a prompt prefix. Cache hit is the cheap reread of that prefix. Batch is half price if your job can wait.

Disclosure: no affiliate links here, and nobody paid to be listed. Here’s how we work. For the cross-vendor view, compare OpenAI, Claude and Gemini API prices. For the chat/Code subscriptions, see Claude Code pricing and the Claude plans tracker.

TL;DR by situation

  • Cheapest (small prompts): Claude Haiku 5.5 at $0.10 in / $0.50 out, as long as each prompt stays at or under 100,000 tokens.
  • Balanced (everyday agents): Claude Sonnet 5.5 or Sonnet 5 at $2 / $10. Same list price; Sonnet 5.5 is the newer line.
  • Top model: Claude Opus 5.5 at $4 / $20, with the full 1M context at that rate. Fable 5.1 is $10 / $50 if you need the top of the stack.
  • Heavy caching: pay once to write the prefix (1.25× or 2× input), then hit it at 0.025×–0.1× input. On Sonnet 5.5 a cache hit is $0.10 per 1M tokens.
  • Jobs that can wait: Batch is 50% off input and output, and it stacks with caching. Fast mode cannot combine with Batch.
  • Already on Max: Max 5x includes $100/mo in API credits; Ultra (formerly Max 20x; claude.com/pricing, verified Nov 1, 2026) includes $200/mo. No rollover. They cover the API, not interactive Claude Code.

Claude API price table

Standard tier, paid. Prices per 1M tokens. Short labels so the table survives a phone screen.

ModelInputOutput5m write1h writeCache hit
Fable 5.1$10$50$12.50$20$0.25
Opus 5.5$4$20$5$8$0.20
Sonnet 5.5$2$10$2.50$4$0.10
Sonnet 5$2$10$2.50$4$0.20
Haiku 5.5 (≤100K)$0.10$0.50$0.125$0.20$0.01
Haiku 5.5 (>100K)$0.50$2.50$0.625$1$0.05

Source: Anthropic Claude pricing, checked Oct 18, 2026. Sonnet 5’s $2 / $10 intro rate is now permanent (notes checked Oct 17, 2026).

Haiku 5.5’s two price bands

Most Claude models from 4.6 onward include the full 1M-token context at the standard rate. No long-prompt surcharge. Haiku 5.5 is the exception that actually hurts.

  • Prompts up to 100,000 tokens: $0.10 / $0.50
  • Prompts over 100,000 tokens: the whole request jumps to $0.50 / $2.50 — not just the tokens past the line

That is a 5× bump on both sides. Cache writes and hits step with it: $0.625 / $1 write and $0.05 hit once you’re over 100K.

Worked example: 1M input + 200K output on Haiku 5.5

  • Five clean 200K prompts? Every one is over 100K, so you pay the high band: 1 × $0.50 + 0.2 × $2.50 = $1.00.
  • Ten prompts of 100K each? Same tokens, low band: 1 × $0.10 + 0.2 × $0.50 = $0.20.
  • Same job on Sonnet 5.5: 1 × $2 + 0.2 × $10 = $4.00, with no 100K trap.

If your pipeline can chunk under 100K, Haiku stays the bargain. If it can’t, Sonnet or Opus may be rational before you even talk quality.

Prompt caching: when it pays off

Claude’s cache math looks like Dwight putting the stapler in Jello: one weird upfront cost, then everything gets easier if you stick with the system.

Multipliers from Anthropic’s pricing docs (checked Oct 17–19, 2026):

  • 5-minute cache write: 1.25× input
  • 1-hour cache write: 2× input
  • Cache hit / refresh: usually 0.1× input; Fable 5.1 is 0.025×; Opus 5.5 and Sonnet 5.5 are 0.05×

Worked example: Sonnet 5.5, 50K-token system prompt reused 100 times in an hour

Assume one 1-hour write, then 99 hits on that 50K prefix. Output ignored so we can see cache alone.

  • Write once: 0.05 × $4.00 = $0.20
  • 99 hits: 99 × 0.05 × $0.10 = $0.495
  • Cache path total for the prefix: about $0.70
  • Without caching, those 100 sends of the same 50K would be 100 × 0.05 × $2.00 = $10.00

Break-even is early. Pay the write when the same long prefix comes back more than a couple of times in the TTL window. Skip it for one-off prompts — you’re buying Jello for a stapler you’ll never open again.

Batch and long-context rates

Batch: 50% off input and output. Confirmed on Anthropic’s pricing page Oct 19, 2026. It stacks with prompt-caching multipliers. Fast mode is not available with Batch.

Batch list examples (half of Standard):

ModelBatch inBatch out
Fable 5.1$5$25
Opus 5.5$2$10
Sonnet 5.5$1$5
Haiku 5.5 ≤100K$0.05$0.25
Haiku 5.5 >100K$0.25$1.25

Long context: Claude 4.6+ (except Haiku 5.5) and Mythos Preview keep the full 1M context at standard rates — no step. Haiku’s >100K band is the only Claude long-prompt trap on the current table.

Web search tool: $10 per 1,000 searches, plus standard token costs for search-generated content. Errors are not billed. Web fetch has no extra fee beyond tokens. (Checked Oct 19, 2026.)

Fast mode (for context, not Batch): Opus 5.5 lists $8 / $40 on Fast. Not combinable with Batch.

Max and Team API credits vs paying per token

If you already pay for Claude Max, you are not starting from zero on the API. Max is the rare subscription that also funds Console spend.

PlanMonthly API creditsCoversDoes not cover
Max 5x ($100/mo)$100Claude API, Console Playground, Managed Agents, Agent SDKInteractive Claude Code; extra usage in Claude apps
Ultra ($200/mo)$200SameSame

Team plans also include monthly API credits (Anthropic’s help article covers Max and Team together). Unused credits do not roll over. Sources: Monthly API credits for Max and Team plans and API credits for subscribers, checked Oct 18, 2026.

When Max credits beat pure API

  • You already want Max for chat / Claude Code limits (5× or 20× Pro’s five-hour pool — see Claude Code pricing).
  • Your API spend sits near or under $100 or $200/mo. The credits eat that bill; overage is still list rates.
  • You use Console Playground, Managed Agents or the Agent SDK — those draw from the same credit pot.

When to stay on pure pay-as-you-go

  • You don’t need the Max chat pool. Paying $100/mo just to unlock $100 of credits is a round trip with fees.
  • You burn past $200/mo regularly. Credits are a discount on the first slice, not a volume deal.
  • Your main spend is interactive Claude Code. Credits explicitly don’t cover that; Code still rides the subscription usage pool (or a separate Console path).

Pro ($20/mo) does not include these API credits. Free has no API free tier either — Anthropic gives new users “a small amount of free credits to test the API,” then you’re on paid rates.

For how Max sits next to OpenAI’s API, see OpenAI API pricing. For chat product comparison, ChatGPT vs Claude.

FAQ

Is the Claude API free? No. There’s no free tier; new accounts get a small test credit, then paid rates.

Is Claude API cheaper than OpenAI? Depends on the model. Haiku 5.5 and gpt-6-luna both list $0.10 / $0.50; Opus 5.5 ($4 / $20) undercuts GPT-6 Astra ($10 / $50). Full side-by-side: AI API pricing.

How does Claude prompt caching work? You pay 1.25× or 2× input to write a prefix, then read it at a fraction of input (as low as $0.10/1M on Sonnet 5.5).

Do Max plan API credits roll over? No. Max 5x’s $100 and Ultra’s $200 reset monthly with no rollover.

What’s the cheapest Claude API model? Haiku 5.5 at $0.10 / $0.50 for prompts up to 100K tokens; over 100K it becomes $0.50 / $2.50.

Does Batch stack with caching? Yes. Batch is 50% off input and output and stacks with cache multipliers; Fast mode does not combine with Batch.

Sources

Related reading