Як розрахувати Claude API Cost

Що таке Claude API Cost?

The Claude API Cost Calculator estimates your monthly Anthropic API spend across Claude model variants including Claude 3.5 Sonnet, Claude 3 Opus, Claude 3 Haiku, and Claude 3.5 Haiku. It accounts for the distinct input, output, and prompt caching pricing tiers.

Формула

Monthly Cost = (Input Tokens × Input Rate) + (Output Tokens × Output Rate) + (Cached Tokens × Cache Rate)

T_in: Input Tokens (tokens/request) — Average prompt tokens per API call
T_out: Output Tokens (tokens/request) — Average completion tokens per API response
H: Cache Hit Rate (%) — Percentage of input tokens served from prompt cache
N: Monthly Calls (calls/month) — Total API requests per month
P_cache: Cache Price ($/1M tokens) — Reduced rate for cached prompt tokens (typically 90% off)

Покрокова інструкція

1Select the Claude model variant (Opus, Sonnet, Haiku)
2Enter average input tokens, output tokens, and prompt cache hit rate per request
3Specify total monthly API call volume
4View total cost with breakdown showing cache savings potential

Розв'язані приклади

Введення

Claude 3.5 Sonnet: 1,500 input tokens, 800 output, 50% cache hit rate, 20,000 calls/month

Результат

Non-cached input: 15M tokens × $3.00/1M = $45.00. Cached input: 15M × $0.30/1M = $4.50. Output: 16M × $15.00/1M = $240.00. Total: $289.50/month. Cache savings: $40.50.

Введення

Claude 3 Haiku: 500 input, 200 output, 100,000 calls/month

Результат

Input: 50M × $0.25/1M = $12.50. Output: 20M × $1.25/1M = $25.00. Total: $37.50/month.

Поширені помилки

✕Not leveraging prompt caching for repeated system prompts, which can reduce input costs by 90%
✕Comparing Claude pricing to GPT without normalizing for output quality and token efficiency differences
✕Overlooking that Claude has a 200K context window — very long inputs significantly increase per-request cost

Часті запитання

How does Claude pricing compare to GPT-4o?

Claude 3.5 Sonnet ($3/$15 per 1M input/output) is comparable to GPT-4o ($2.50/$10). Claude 3 Haiku ($0.25/$1.25) competes with GPT-4o-mini ($0.15/$0.60). Claude 3 Opus ($15/$75) is the premium option. With prompt caching, Claude can be significantly cheaper for workloads with repeated system prompts.

What is prompt caching and how much does it save?

Prompt caching stores frequently used prompt prefixes (like system prompts) so they do not need to be re-processed on each call. Cached tokens are charged at 90% less than the standard input rate. For applications with long system prompts sent repeatedly, this can reduce input costs by 50-80%.

Готові порахувати? Спробуйте безкоштовний калькулятор Claude API Cost

Спробуйте самі →