Anthropic API pricing
Anthropic API pricing: current per-token rates
Anthropic's current API rates per 1M tokens, input / output / cached-read, verified 2026-07-01. Every rate links to its official source. Introductory $2 / $10 per 1M through 31 Aug 2026; $3 / $15 standard from 1 Sep 2026.
Current Anthropic model prices
| Model | Input ($/1M) | Output ($/1M) | Cached in ($/1M) | Context |
|---|---|---|---|---|
| Claude Haiku 4.5 | $1 | $5 | $0.10 | — |
| Claude Opus 4.8 | $5 | $25 | $0.50 | — |
| Claude Sonnet 5 | $3 | $15 | $0.30 | 1M |
Subscription vs API: where the plan stops being cheaper
Break-even is the plan's monthly price divided by your blended per-token rate. At Sonnet 5 ($3/$15 per 1M), a 3:1 input:output mix blends to $6 per 1M tokens. Below your plan's break-even the subscription wins; above it, metered API does — and caching pushes the crossover higher.
| Plan | Price | Break-even vs API (Sonnet, uncached, 3:1) |
|---|---|---|
| Pro | $20/mo | ~3.3M tokens/mo ($20 ÷ $6) |
| Max 5x | $100/mo | ~17M tokens/mo ($100 ÷ $6) |
| Max 20x | ~$200/mo | ~33M tokens/mo ($200 ÷ $6) |
Common questions
Can I use the Anthropic API for free?
No — API access is metered per token at the rates above. Subscriptions (where offered) bundle usage into a flat monthly price; the API itself is pay-as-you-go.
Which model is cheapest?
The lowest input/output rate in the table above for light work; match the model to the task rather than defaulting to the cheapest — routing the hard steps up and the cheap steps down usually beats a single choice.
Prices change. Verified 2026-07-01 — re-check the source before relying on them. Corrections: hello@aiarch.dev.
Learn to cost-model agents like an architect.
What a Claude agent actually costs, or compare Professional Membership pricing.