ANTHROPIC API PRICING

Anthropic Claude token cost calculator.

Calculate Claude API costs from official input, cache-read, cache-write, output, context-window, and routing details.

Calculate a workload Review rate cards →

Models: 5
Rate cards: 5
Input range: $1–$10
Output range: $5–$50
Cache-priced cards: 5
Verified: 2026-08-16

CACHE AMORTIZATION

Price writes and reads as an episode.

Model a stable reusable prefix, each cache creation, and the reads that actually land inside its lifetime. The counterfactual weights every billing category at its published rate.

Model tierCache episodes / monthCache reads after each writeTotal input tokens / requestReusable prefix tokensCache lifetime

CACHE EPISODE ECONOMICS

Claude Fable 5$10 ordinary input · $12.50 cache write · $1 cache read / 1M5m TTL · output tokens excluded because they are identical in both arms

Without caching$120.00

With caching$56.25

Estimated reduction: $63.75 (53.1%)

BREAK-EVEN1 cache read per write

600 modeled requests across 100 cache episodes.

This is an input-cost counterfactual, not realized savings. It assumes the prefix stays byte-stable and every planned read hits within the selected TTL. Verify provider usage fields and accepted-answer quality.

Claude API pricing ↗

ANTHROPIC WORKLOAD

Model the calls you actually make.

Enter monthly workload, tokens, and quality pass rates. The calculator resolves eligible price bands, then shows cost per accepted answer and the candidate break-even pass rate.

Provider and model tierRequests / monthInput tokens / requestOutput tokens / requestInput reduction25%Output reduction20%Warm cache-read share30%Baseline quality pass rate90%Candidate quality pass rate85%

Estimated monthly API spend · Anthropic

Claude Fable 5Before · Standard: $10 input · $1 cache read · $50 output / 1MAfter · Standard: $10 input · $1 cache read · $50 output / 1MGlobal routing · price bands selected automatically from input size

Before$550.00

After$364.25

Potential saving: $185.75 (33.8%)

QUALITY-ADJUSTED COST

Before / accepted answer$0.061190% pass assumption

After / accepted answer$0.042985% pass assumption

Candidate break-even: 59.6% pass rate. Your scenario assumes 85%.

Warm cache-read scenario, not general ROI. It excludes cache writes/storage, tools, regional uplifts, retries, and quality failures. A dash means no published cache-read rate.

Claude API pricing ↗

NEXT STEPValidate the 29.9% accepted-answer advantage.

The estimate is not a saving until the same workload still passes its quality bar. Compare a supported recipe in the lab, or use the evidence library to design a provider-specific test.

Run a controlled test Attribute by workflow Audit an existing bill Browse evidence methods See one-time access

Prompt-cache reads and writes have different rates, and one-hour cache writes can cost more than five-minute writes. The scenario calculator models warm reads only and labels that limitation.

DATED RATE CARDS

Anthropic prices with their scope intact.

Rates are USD per one million tokens. Future and transitional cards remain labeled with their effective dates instead of silently replacing today’s tier.

Official provider API rate cards, USD per one million tokens
Provider / model	Price scope	Input USD / 1M	Cache read USD / 1M	Output USD / 1M	Context	Source
AnthropicClaude Fable 5`claude-fable-5`	StandardGlobal routing	$10	$1	$50	1M	Official ↗
AnthropicClaude Haiku 4.5`claude-haiku-4-5`	StandardGlobal routing	$1	$0.1	$5	200K	Official ↗
AnthropicClaude Opus 5`claude-opus-5`	StandardGlobal routing	$5	$0.5	$25	1M	Official ↗
AnthropicClaude Sonnet 4.6`claude-sonnet-4-6`	StandardGlobal routing	$3	$0.3	$15	1M	Official ↗
AnthropicClaude Sonnet 5`claude-sonnet-5`	StandardGlobal routing	$2	$0.2	$10	1M	Official ↗

Rates in USD per 1M tokens

Anthropic

Claude Fable 5

claude-fable-5

StandardGlobal routing

Input: $10
Cache read: $1
Output: $50
Context: 1M

Open official source

Anthropic

Claude Haiku 4.5

claude-haiku-4-5

StandardGlobal routing

Input: $1
Cache read: $0.1
Output: $5
Context: 200K

Open official source

Anthropic

Claude Opus 5

claude-opus-5

StandardGlobal routing

Input: $5
Cache read: $0.5
Output: $25
Context: 1M

Open official source

Anthropic

Claude Sonnet 4.6

claude-sonnet-4-6

StandardGlobal routing

Input: $3
Cache read: $0.3
Output: $15
Context: 1M

Open official source

Anthropic

Claude Sonnet 5

claude-sonnet-5

StandardGlobal routing

Input: $2
Cache read: $0.2
Output: $10
Context: 1M

Open official source

USD per 1M tokens. Snapshot verified 2026-08-16. Cache writes, cache storage, tools, regions, and provider-specific thresholds may be billed separately.

A missing cache rate is shown as “—”, never treated as free. Consumer chat-plan quotas are not API prices.

BILLING CAVEAT

What to verify before shipping

Prompt-cache reads and writes have different rates, and one-hour cache writes can cost more than five-minute writes. The scenario calculator models warm reads only and labels that limitation.

Official source 1 ↗

MEASUREMENT RULE

Price accepted answers, not token deltas

Count retries, fallbacks, tool calls, latency failures, and answers rejected by the quality rubric. A lower estimated token bill is useful only when the workload still succeeds.

Open the controlled A/B lab

QUESTIONS

Before you trust the estimate.

How current are the Anthropic API prices?

This page uses TokenGauge’s 2026-08-16 official-source snapshot. Follow the linked provider pages before making a production commitment because prices can change.

Does this calculator include every Anthropic charge?

No. The workload calculator models input, output, and a warm cache-read share. The separate cache-episode calculator includes published cache writes and reads, but storage, tools, retries, regional uplifts, taxes, batch modes, and quality failures can still change the invoice.

Are consumer chat subscriptions included?

No. ChatGPT, Claude, Gemini, Grok, Kimi, and other consumer-plan quotas are separate from provider API token billing.