Anthropic
Claude Haiku 4.5 pricing
Anthropic's fastest and cheapest model. 3x cheaper input than Sonnet, strong enough for routine chat, routing, and classification.
Input
$1.00/ 1M tok
Output
$5.00/ 1M tok
- Context window
- 200K
- Max output
- 64K
- Cached input
- -
- Verified
- 2026-04-06
Haiku 4.5 is Anthropic's entry-level tier: $1 per 1M input and $5 per 1M output, 200K context window, same tokenizer and API surface as the Sonnet and Opus tiers. If your workload is customer support chat, log enrichment, classification, or any task where you'd happily accept a slightly less careful answer for a third of the cost, start here.
The output-to-input ratio is the same 5:1 shape as the rest of the Claude family, so a 1K-token prompt with a 500-token reply costs about $0.0035. At 100K requests a month that's $350 - versus about $1,050 on Sonnet 4.6 for the same traffic. Most teams leave too much on the table by defaulting to Sonnet when Haiku would do.
Calcis uses Anthropic's own countTokens API to get an exact input token count before you send, and predicts response length from the prompt so you can see the full dollar number rather than guessing output size.
Estimate your cost on Claude Haiku 4.5
Paste your prompt into the estimator, pick Claude Haiku 4.5, and see the exact dollar cost - input tokens counted with the provider's own tokenizer, output tokens predicted by our regression model.
Frequently asked
- How much does Claude Haiku 4.5 cost per request?
- A 1,000-token prompt with a 500-token reply costs about $0.0035 ($0.001 input + $0.0025 output). At 100,000 requests a month, that's around $350 in API fees.
- Is Haiku 4.5 cheaper than GPT-4o mini?
- Haiku 4.5 costs $1/$5 per 1M tokens. GPT-4o mini costs $0.15/$0.60 - about 7x cheaper on both sides. Pick Haiku for Claude-family tooling or reasoning quality; pick 4o mini on pure price.
- Does Haiku 4.5 support the same features as Opus?
- Same API surface and tokenizer, but the 200K context window is smaller than Opus 4.6's 1M. Max output is 64K, and some advanced features (extended thinking) behave differently at the Haiku tier.
- What tokenizer does Claude Haiku 4.5 use?
- Anthropic doesn't publish the tokenizer, so Calcis uses Anthropic's countTokens API - the same boundary Anthropic bills against.
Pricing verified 2026-04-06 from the provider's rate card.