Claude API Pricing 2026: Per Million Tokens, Caching, Batch and Cost Examples

Official API prices for Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5, how caching and batch processing save money, how long context is billed, plus two worked cost examples.

Short answer: the Claude API is billed per million tokens and output costs five times input; cache prompts you reuse and send non-urgent jobs through batch processing to cut the bill sharply. Prices come from Anthropic's official docs, checked in November 2026, all in US dollars per million tokens.

Current model prices

ModelInputOutput5-min cache write1-hour cache writeCache hit

Claude Fable 5.2$8$40$10$16$0.20

Claude Opus 5.5$4$20$5$8$0.20

Claude Sonnet 5.5$2$10$2.50$4$0.10

Claude Haiku 5.5 · prompts up to 100K tokens$0.10$0.50$0.125$0.20$0.01

Claude Haiku 5.5 · prompts over 100K tokens$0.50$2.50$0.625$1$0.05

Fable 5.2 is the most capable, Opus 5.5 is the recommended default, Sonnet 5.5 suits well-scoped daily work and Haiku 5.5 is the cheapest.

Cache hits on Opus 5.5 and Sonnet 5.5 cost 5% of the base input price, and 2.5% on Fable 5.2.

Batch processing at half price

Jobs that do not need an instant answer can use the Batch API at 50% off both input and output; most batches finish within an hour.

For example, Opus 5.5 in batch is $2 input and $10 output; Sonnet 5.5 is $1 and $5.

A batch holds up to 100,000 requests or 256 MB.

How long context is billed

Models from 4.6 onward support a 1M-token context at standard rates throughout: a 900K-token request costs the same per token as a 9K one.

Haiku 5.5 is the exception: a request whose prompt exceeds 100K tokens is billed at the higher tier, and cache hits count toward that length.

Requests pinned to US-only inference cost 1.1 times the standard price.

How to work out the cost

Cost = input tokens × input price + output tokens × output price, with cached input billed at the cache price. Two examples:

Sonnet 5.5, no caching: 5M input at $2 plus 1M output at $10 is $20.

Same volume with 80% of input cached: 1M uncached at $2, 4M cached at $0.10 and 1M output at $10 comes to about $12.40, almost 40% less.

Coding assistants read far more than they write, so a higher cache hit rate saves more.

Common mistakes

Anthropic notes that Claude 4.7 and later use a newer tokenizer that produces about 30% more tokens for the same text, so leave headroom.

Switching models means rewriting the cache, so the first turn is billed at cache-write prices.

Fast mode gives quicker replies at a higher price and is only offered on Opus 5.5 and Fable 5.2.

Pick credit that fits your usage

Our Claude API comes as dollar credit or token packs: dollar credit is spent at the prices above, token packs are counted in tokens.

See the related guide to choose; official prices are on the Claude pricing docs.

Related product: Claude API. Related guides: Claude API credit vs token packs, Claude model guide.

Further reading: Claude API Credit vs Token Packs: Do the Math by Usage and Model, Fable, Opus, Sonnet or Haiku: Key Takeaways from Claude's Model Guide, Claude Code vs Codex vs Cursor: How They Work, What They Cost, Who They Suit.