You ask your AI how to get a coffee stain out of a shirt. Answer in three seconds. Do you need the priciest flagship model for that? No. That's like starting a lorry to walk to the mailbox.
That's what Haiku is for. On 7 October, Anthropic released Claude Haiku 5.5. It is the smallest model in the current Claude family. And the fastest.
What is Haiku?
Anthropic builds its models in three sizes. Opus is the heavyweight, Sonnet the all-rounder, Haiku the sprinter. Smaller means less computing per answer, so it is quicker and cheaper. For tricky logic or long, convoluted tasks, the bigger model usually wins.
According to Anthropic, Haiku 5.5 has a context window of one million tokens. A token is roughly a syllable. A million of them equals several thick books the model keeps in view at once. For comparison: its predecessor managed around 200,000.
What does it cost?
Through the API, you currently pay roughly 10 US cents per million input tokens and roughly 50 cents per million output tokens. For very long inputs above 100,000 tokens, the price goes up. Reports differ on some details, so check the provider's pricing page before relying on it.
For a beginner that means: a normal conversation costs fractions of a cent via the API. If you automate a lot, you save noticeably compared with the big models.
And in the chat app?
According to consistent reports, Haiku 5.5 can be selected in the Claude apps and on claude.ai, including the free plan. Free to try, but with a limit. That applies to every provider: if you write a lot every day, you eventually hit the message cap or get a weaker model. Inference (computing the answers) costs real money. Not a scandal, just physics and electricity.
Whether it shows up in your model menu, the app will tell you. Looking costs nothing.
Is it for me?
Yes, if you:
- want quick answers to everyday questions
- have texts summarised, reworded or translated
- sort emails and notes
Probably not if you're deep in a long, multi-step task, like a coding project with many files. The bigger model pays off there. Developers like Haiku as a cheap helper in the background. Beginners don't need a single line of code.
Try it right now
Take a task you already have. For example:
Summarise this text in five bullet points and name the one thing I must not overlook.
Ask the same question to the bigger model. Compare. Often you won't notice a difference. Sometimes you will, and then you know when the bigger engine is worth it.
The catch
Small models hallucinate (invent facts) just as readily as big ones, sometimes more brazenly. Trust it like an intern on day one. Check important numbers yourself.
Unpaid recommendation, own opinion: for everyday questions a fast small model is often the smarter choice. Try it before paying for more.