Claude API Cost
Per 1,000 Tokens — 2025
Anthropic publishes prices per million tokens. Here's the breakdown per 1,000 — plus a free calculator to measure your exact Claude Code session costs.
Calculate My Session Cost →Claude API — Cost Per 1,000 Tokens (2025)
| Model | Input / 1K | Output / 1K | Cache Write / 1K | Cache Read / 1K |
|---|---|---|---|---|
| Claude Haiku 3.5 Budget | $0.00080 | $0.00400 | $0.00100 | $0.00008 |
| Claude Sonnet 4.5 Default | $0.00300 | $0.01500 | $0.00375 | $0.00030 |
| Claude Opus (latest) | $0.01500 | $0.07500 | $0.01875 | $0.00150 |
Cost Per API Call — Real Examples
Here's what typical Claude API calls actually cost per request, before and after prompt caching kicks in:
Without cache: $0.000400 + $0.000800 = $0.0012/call
With 10K cached system prompt (1 cache read): $0.000400 + $0.000800 - $0.000720 (saved) = $0.00048/call
Without cache: $0.00600 + $0.01200 = $0.018/call
With 8K cached system prompt: ~$0.007/call after cache hit
Standard: $0.24 + $0.30 = $0.54 total
With prompt caching (60% hit rate): ~$0.30–$0.35 total
Haiku (no cache): $0.00080/doc × 50K = $40 input + $20 output = $60 total
Haiku with system-prompt caching: ~$32 total (47% savings)
Frequently Asked Questions
How much does Claude cost per 1,000 tokens?
Per 1,000 tokens (input / output): Claude Haiku 3.5 — $0.00080 / $0.00400. Claude Sonnet 4.5 — $0.00300 / $0.01500. Claude Opus — $0.01500 / $0.07500. Cache read prices are ~10× lower: Haiku $0.00008/1K, Sonnet $0.00030/1K, Opus $0.00150/1K. Anthropic's official pricing is quoted per million tokens — divide by 1,000 to get these per-thousand figures.
How many tokens is a typical Claude API call?
A short Claude Code tool call typically uses 300–800 input tokens and 100–400 output tokens. A full conversation turn with a large system prompt may use 10,000–50,000 input tokens (where caching makes a big difference). A typical 1-hour Claude Code session accumulates ~60,000–120,000 input tokens and 15,000–30,000 output tokens, depending on the task complexity.
How many tokens is 1,000 words?
Approximately 1,300–1,400 tokens for 1,000 words of natural English text. Code is slightly more token-dense. A 500-word email is ~650 tokens; a 2,000-word blog post is ~2,600 tokens. JSON and structured data use more tokens per character than prose. The Claude tokenizer (BPE) is similar to GPT's — common short words are often 1 token, longer or technical words may be 2–4 tokens.
How do I track exact token usage in my Claude Code sessions?
Claude Code logs every API call to ~/.claude/projects/<project>/*.jsonl. Each entry includes model name, input_tokens, output_tokens, cache_creation_input_tokens, and cache_read_input_tokens. Paste the file into the Claude Code Cost Calculator above — it applies per-model pricing and shows a full cost breakdown. You can also monitor usage in the Anthropic console at console.anthropic.com.