Claude API Cost Calculator

About This Tool

The Claude API Cost Calculator handles the three current Anthropic tiers — Opus 4.7, Sonnet 4.6, and Haiku 4.5 — with full support for prompt caching math. Enter input and output tokens per call, optional cached prefix size, the number of times that cache gets reused before refresh, and a monthly request volume. The calculator outputs per-call and monthly cost for each model.

Anthropic pricing (April 2026)

Opus 4.7: $15 input / $75 output per million tokens. Sonnet 4.6: $3 / $15. Haiku 4.5: $1 / $5. The 5x output multiplier is consistent across the lineup. Sonnet sits in a sweet spot — five times cheaper than Opus on input, but loses very little on most evaluation benchmarks for non-frontier tasks.

Prompt caching mechanics

Cache writes cost 1.25x the base input rate. Cache reads cost 0.1x — a flat 90% discount. If you have a 10,000-token system prompt that gets reused across 100 follow-up turns, you pay 1.25x once and 0.1x ninety-nine times. The effective input cost on that prefix collapses from $0.15 per call (Opus) to roughly $0.018 per call after the first.

Caching breaks even at two cache reads. Beyond that, every additional reuse compounds the savings. The default cache TTL is 5 minutes; an extended 1-hour cache is available for batch workloads. Cache hit requires identical prefix bytes — even a single-character change invalidates the entry.

When to skip caching

Skip if your prompt changes on every call. Skip for low-volume workloads where the 1.25x write premium isn't recouped. Skip when prompts are under ~1024 tokens — caching minimums often exceed the savings on small contexts.

Compare pricing against alternatives: GPT API cost calculator, Gemini API cost calculator, and the side-by-side LLM cost comparison. For deeper cache modeling, use the prompt caching savings calculator. To estimate raw token counts from text, use the token counter.