How Claude API pricing works
Anthropic bills base input tokens and output tokens per million. On the official pricing table as of this snapshot (platform.claude.com/docs/en/about-claude/pricing), Claude Fable 5.1 is $10 input / $50 output with cache hits at $0.25 / MTok. Claude Fable 5 is the same $10 / $50 with the older $1 cache-hit rate. Claude Opus 5 is $5 / $25, Claude Sonnet 5 is $2 / $10, and Claude Haiku 4.5 is $1 / $5. Mythos 5.1 (invite-only / Project Glasswing) is omitted from this packed table.
Sonnet 5 launched with introductory $2 / $10 pricing that was scheduled to rise to $3 / $15 on 1 September 2026. Anthropic canceled that raise; $2 / $10 is now the standard list rate.
Fable 5.1 is the current top public SKU. Fable 5 is kept beside it because both remain on the official card. Opus 5 is the everyday flagship. Sonnet 5 is the default production workhorse. Haiku 4.5 is the cheap, fast lane.
Caching, batch, and tokenizer caveats
Prompt caching is a first-class Anthropic feature: cache writes cost more than base input; cache hits and refreshes cost much less. A calculator that only uses base input will overstate cost for apps that pin a large system prompt or document prefix. Batch is typically 50% of standard. Neither effect is applied in the numbers below.
Newer Claude tokenizers can emit more tokens for the same text than older families. Token counts you measured on Sonnet 4.x will not transfer 1:1. When you move a workload onto Sonnet 5 / Opus 5 / Fable 5, re-measure tokens; do not only multiply the old count by the new rate.
US-only inference routing and some marketplace deployments apply a multiplier. Regional Bedrock / Vertex endpoints can add about 10% over global. This tool assumes global first-party list rates.
When to pick Haiku, Sonnet, Opus, or Fable
Haiku 4.5 is the cost and latency default for classification, routing, extraction, and high-QPS assistants. Sonnet 5 is the production generalist. Opus 5 is the expensive everyday flagship. Fable 5.1 is the top public row. Because output is 5× input, a verbose Fable agent can outspend a quiet Sonnet job even at similar request counts.
Image, PDF, and other multimodal tokens are not modeled in v1. Extended thinking, if billed through output tokens, should be included in the output field you type. Grounding, web search, and code execution may have separate add-on fees on the official page.
Use the calculator to sanity-check an order of magnitude, then confirm on platform.claude.com/docs/en/about-claude/pricing. Update src/data/pricing.ts when Anthropic ships the next SKU rename — the UI date will follow lastUpdated.
Frequently asked questions
What is the Claude API price per million tokens?
In this snapshot: Haiku 4.5 is $1 / $5, Sonnet 5 is $2 / $10, Opus 5 is $5 / $25, and Fable 5.1 / Fable 5 are $10 / $50. Fable 5.1 cache hits are $0.25 / MTok. Mythos 5.1 is invite-only and not in the packed table.
Where is Opus 4.8 / Sonnet 4?
They are not in the default compare set. Current public SKUs on this calculator are Fable 5, Opus 5, Sonnet 5, and Haiku 4.5.
Does this include Anthropic prompt cache discounts?
No. Cache writes and hits have their own columns on Anthropic’s table. Entering only base input/output is a conservative estimate for cache-heavy apps.
Does this match claude.ai Pro or Team?
No. This is the Messages API (pay per token). Claude.ai subscriptions are a different product with a monthly seat price.