Claude Haiku 4.5 pricing
Claude Haiku 4.5 costs $1 per million input tokens and $5 per million output tokens , with cached input at $0.1. Its context window is 200K tokens. Batch requests are discounted 50%.
claude-haiku-4-5. See all shutdown dates → Price per million tokens
| Input | $1 |
| Cached input | $0.1 |
| Output | $5 |
| Context window | 200K tokens |
| Batch discount | −50% |
| Minimum prefix to cache | 4,096 tokens |
What it costs in practice
Monthly cost for three common workload shapes, with prompt caching applied.
| Workload | Shape | Per request | Per month |
|---|---|---|---|
| Support chatbot | 1.5k in / 400 out, 100k requests/month, 30% cached | $0.00309 | $309.50 |
| RAG / document Q&A | 12k in / 700 out, 50k requests/month, 60% cached | $0.00902 | $451.00 |
| Coding agent | 25k in / 2.5k out, 20k requests/month, 80% cached | $0.0195 | $390.00 |
Model your own workload with caching, batching and tokenizer correction →
Caching on Claude Haiku 4.5
A cached prefix on this model must be at least 4,096 tokens. Below that, Anthropic ignores cache_control silently — no error is returned and you pay full input price on
every request. A cache hit costs $0.1 per million tokens, roughly 10% of the
base input rate.
Comparable models from other providers
| Model | Provider | Input | Output |
|---|---|---|---|
| Grok Build 0.1 | xAI | $1 | $2 |
| o4-mini | OpenAI | $1.1 | $4.4 |
| GPT-5.1 | OpenAI | $1.25 | $10 |
| GPT-5 | OpenAI | $1.25 | $10 |
All figures verified 2026-08-18 against Anthropic's official pricing documentation . Fields the provider does not publish are shown as “—” rather than estimated.