Claude Haiku 5.5: a much cheaper AI model for busy chats

On 7 October 2026, Anthropic released Claude Haiku 5.5. It's an AI model, the "brain" behind a chatbot. This one is the small, cheap model in the Claude family, meant for busy, repetitive work. Verdict: if a chatbot answers your customers, ask your provider about it. The price per use is far lower.

What changed

Anthropic says Haiku 5.5 is its cheapest, fastest and most capable small model so far. It lists live customer support and sorting messages into groups as good uses. It's out now for developers through Anthropic's own platform, and on Amazon, Google and Microsoft's cloud services. Source

Anthropic says it costs about 75% less on average than the last small model, Haiku 4.5. These prices apply when a developer's software uses it. They are per million tokens, which are small pieces of words:

Haiku 5.5: $0.10 for text going in and $0.50 for text coming out, for jobs up to 100,000 tokens.
Haiku 4.5: $1 in and $5 out.
Sonnet 5.5, the middle model: $2 in and $10 out.

Longer jobs, above 100,000 tokens, cost $0.50 in and $2.50 out. Source

It also has an effort setting, from Low to Max. A lower setting is cheaper and faster, and a higher one thinks harder.

Why it matters for a small business

A chatbot on your WhatsApp or website pays for every message it reads and writes. Most customer questions are simple, like opening hours or prices. A small, cheap model can handle those.

Anthropic's own tests show Haiku 5.5 scoring far above Haiku 4.5, and below Sonnet 5.5 on every test it published. So it is cheaper, but not as strong. It can still make mistakes with hard or unusual questions.

What to do now

If you only use Claude's app or a ready-made chatbot, nothing changes for you today. Anthropic's page doesn't say whether the Claude app or any free plan includes Haiku 5.5.
If a provider runs your chatbot, copy this message:

Hi, Anthropic released Claude Haiku 5.5 on 7 October. Could our chatbot use it for simple questions like opening hours and prices? Would that lower what we pay?

Test before switching. Send the five questions your customers ask most to the old and new versions. Check that prices, hours and policies come out right.
Keep a stronger model for hard cases. A common setup is a cheap model for the easy questions and a bigger one for complaints or complicated bookings. If you're building this yourself, the model name is claude-haiku-5-5.

The catch

Most numbers come from Anthropic. The scores and the "75% less" figure are its own. Customer results on its page, such as Asana, HubSpot and Box, are the companies' own claims too.
It uses a few more tokens per job than Haiku 4.5, because of a new tokenizer, the part that cuts text into pieces. So your real saving may be a bit smaller.
It's weaker on hard tasks. Anthropic says Sonnet 5.5 and Opus 5.5 remain better for complex work.

Want a chatbot that answers customers without a big monthly bill? See our business automation service, or book a free consultation.