← Back to the A-Z glossary
Model · Anthropic · 2026

Claude Haiku 4.5

Claude Haiku 4.5 is Anthropic's small, fast, cheap model. It's less sharp than Sonnet, but 3-4x cheaper and a lot faster. You reach for it for repetitive, high-volume tasks: sorting 10,000 emails, pulling data out of hundreds of PDFs, running a support chatbot.

Model 4 min read Updated 2026-05-23
— What it is

Claude Haiku 4.5, in plain words

In plain words: it's the 'light' model in the Claude lineup. Not the smartest, but unbeatable when you've got a pile of small, well-scoped tasks to knock out for pennies.

In practice: for repetitive, high-volume tasks (sorting 10,000 emails, pulling data from 500 PDFs, answering in a support chatbot), it's the right pick.

200K token context window, like the other Claude models. Multimodal too (images, PDFs).

What I use it for

  • Classify in bulk: Haiku sorts 1,000 emails by category for pennies. Sonnet would do the same for 10x the cost.
  • Extract data: take 200 PDF invoices, pull out the pre-tax total, the vendor, the date. Haiku takes it without breaking a sweat.
  • Support chatbot: a bot that answers the basic questions on a site. Haiku is good enough, and the volume makes Sonnet a non-starter.
  • Pre-filter before Sonnet: Haiku sorts first (signal vs. noise), then Sonnet handles only the signal. Massive savings.
  • Automated newsletters: my AI watch runs on Haiku as a first pass to score the articles.

How it stacks up

vs Claude Sonnet 4.6: Haiku is 3x cheaper, ~2x faster, but loses some reasoning finesse. Which one wins depends on the task.

vs GPT-5 nano / mini: comparable on price and speed. Haiku is better at following long instructions, GPT-5 mini sometimes wins on reasoning.

vs Gemini 2.5 Flash: Flash is even cheaper and faster, but Haiku is more reliable on tasks that call for a bit of judgment.

What it costs

API pricing as of May 23, 2026 (source: anthropic.com/pricing):

  • Input: $1 / million tokens
  • Output: $5 / million tokens
  • Prompt caching: up to 90% savings

In practice: 1,000 emails classified with Haiku = ~$0.05-0.10. The same thing with Sonnet = ~$0.50. On a script that runs every day, that's the difference between €30/month and €300/month.

My take

I really like Haiku for high-volume automation. It's not the model that dazzles me in conversation, but it's the model that saves me the most money when I'm coding agents.

Simple rule at my desk: if the task is repetitive and well-scoped, Haiku. If it needs nuanced judgment, Sonnet. If it's genuinely hard, Opus.

Quick questions

Can Haiku replace Sonnet?

Claude Haiku 4.5 is Anthropic's light, fast model, 3-4x cheaper than Sonnet. It handles repetitive tasks at volume: classification, extraction, chatbots. 200K token context window, multimodal.

Is Haiku 4.5 multimodal?

Yes, it reads images and PDFs just like the other Claude models.

Cheapest in the Anthropic lineup?

Yes, the most affordable on the API side. To go even cheaper, you'd look at Gemini Flash or the small open-weight models.

Verified 2026-05-25 · next review 2026-11-25

Prices and specs checked on anthropic.com/pricing. Personal use: Haiku has been running in my watch and classification scripts since late 2025.

Want to keep going in order?

Head back to the full glossary, search a word, then open only the pages that deserve more than a short definition.

Open the AI glossary