Claude API pricing in 2026

Anthropic’s Claude line-up in 2026 spans four tiers, each priced per million tokens on the API.

Claude API prices

• Claude Fable 5.1 (frontier): $10 per 1M input tokens, $50 per 1M output tokens
• Claude Opus 5 (frontier): $5 per 1M input, $25 per 1M output
• Claude Sonnet 5 (balanced): $2 per 1M input, $10 per 1M output
• Claude Haiku 4.5 (fast): $1 per 1M input, $5 per 1M output

Two things worth knowing about this table. Fable 5.1 is the expensive one — it is Anthropic’s premium adaptive-reasoning model and the highest-indexed model we track, which is why it costs twice Opus 5. And Sonnet 5’s price is now permanent: the $2 / $10 rate was announced at launch as introductory pricing through 31 August 2026, but Anthropic has since confirmed it as the standard price and cancelled the scheduled increase to $3 / $15.

Prompt caching is where Claude prices get cheaper in practice: cache hits are billed at 10% of the base input price, and at just 2.5% ($0.25 per 1M) on Fable 5.1. The Batch API takes 50% off both directions.

Claude subscriptions

• Claude Free: $0 — includes Sonnet 5 and Haiku 4.5
• Claude Pro: $20 / month — adds Fable 5.1 usage credits, Opus 5 and Claude Code
• Claude Max: $100 / month (5x) or $200 (20x) — Fable 5.1 at 50% of your weekly limits

Fable 5.1 is never included outright on Pro: it draws on usage credits there, and on Max it is capped at half your weekly allowance. The cheaper models are the ones you can use without thinking about it.

Is Claude worth the premium?

Claude is often the go-to choice for coding and long-context work. It is rarely the cheapest option on the market, but for many developers the output quality justifies the price. The real question is how many tokens you get for your dollar — and that’s not a feature comparison, it’s arithmetic.

At $10 / $50, Fable 5.1 is one of the most expensive endpoints on the market, and its advantage is quality per task rather than tokens per dollar. Opus 5 is the better-balanced Claude buy at $5 / $25, and Sonnet 5 at $2 / $10 is the one that competes on price — though its tokenizer is famously wasteful, so its effective cost per task runs higher than the sticker suggests.

Pricing changes fast

Claude’s lineup and prices shift with each new release. The figures above are current as of this post; always check the last-verified date on the data you rely on.

Compare Claude’s tokens per dollar against every other provider in the AI intelligence calculator.