You work alone, so you already hand the small jobs to an AI: sorting emails, summarising a client call, filing invoices. Do you really need to pay for the most expensive model to do that?
Claude Haiku 5.5 is the new small model from Anthropic, announced on 7 October 2026. Anthropic describes it as the cheapest, fastest and most capable small model it has released. It costs ten times less than its predecessor, Haiku 4.5.
So let's look at what it can do, what it costs, and whether it has a place in your work.
What is Claude Haiku 5.5?
Anthropic sorts its models by size. Fable and Opus are the most powerful, Sonnet is the all-rounder, and Haiku is the smallest of the family.
Small does not mean bad. You don't call an architect to put up a shelf: for a simple job, a small model answers faster and costs far less.
For a long time I let my biggest model do everything, including renaming three files (a bit like taking a taxi to cross the street 🥲).
According to Anthropic's announcement, Haiku 5.5 is built for short jobs that come back often:
- Summarising a document, a call or a long conversation;
- Classifying requests, for example sorting incoming emails by urgency;
- Looking something up in a database;
- Answering your clients live, where every second of waiting counts;
- Using a browser for you, to fill in a form or check a price.
It also works as a subagent. A large model keeps the hard work, and it hands the small lookups to Haiku.
Rogo, which builds an AI for finance professionals, gives an example. While a large model builds a presentation, Haiku 5.5 goes into an annual report to fetch the revenue figure the deck needs.
How much does Claude Haiku 5.5 cost?
At launch, Anthropic charges $0.10 per million input tokens, and $0.50 per million output tokens. Haiku 4.5 cost ten times more, at $1 and $5.
A token is a piece of a word. Input is what you send the model (your request, your documents). Output is what it writes back.
In practice? Imagine it drafts a reply to a client's email, so about 2,000 tokens to read and 1,000 to write. A thousand replies cost you $0.70.
Here is the same job with two other Anthropic models:
| Model | Input, per million tokens | Output, per million tokens | A thousand email replies |
|---|---|---|---|
| Claude Haiku 5.5 | $0.10 | $0.50 | $0.70 |
| Claude Haiku 4.5 | $1 | $5 | $7 |
| Claude Sonnet 5.5 | $2 | $10 | $14 |
Twenty times cheaper than Claude Sonnet 5.5! At that price, you can have it reread every email and every quote without watching the bill.
This rate applies to requests of up to 100,000 tokens. Above that, Haiku 5.5 moves to $0.50 and $2.50, five times more. Anthropic says nine in ten requests sent to Haiku 4.5 stayed under that threshold.
That threshold matters, because Haiku 5.5 can read up to 1 million tokens in one go, against 200,000 for Haiku 4.5. You can give it a whole folder, but you pay the higher rate for it.
Ten times cheaper on the price list, and yet Anthropic announces 75% savings on average. Why? Haiku 5.5 splits text differently, and the same text counts for about 30% more tokens. What cost $1 to read with Haiku 4.5 therefore costs about $0.13, not $0.10.
These are the prices of the API, the access a program or an automation uses to talk to the model. The Haiku 5.5 model page gives the current rate.
How can you try Claude Haiku 5.5?
It all depends on where you already use Claude:
- In the Claude app, on the web, iOS or Android. At launch, you pick Haiku 5.5 from the list of models, including on a free account;
- In Claude Code, where it can play the subagent;
- In your automations, with n8n or Make for example. You replace the model name with
claude-haiku-5-5at the step in question, then compare the result; - On the big cloud hosts, namely Amazon Web Services, Google Cloud and Microsoft Azure.
Our n8n course shows how to plug a different model into each step.
Haiku 5.5 is also the first Haiku to come with an effort setting. You decide whether it should think a little, to answer fast and cheap, or a lot, to answer better.
By default it is set to medium effort. Keep that setting to begin with, and only go up a notch if its answers disappoint you.
Does one of your automations already call Haiku 4.5? Changing the model name is not enough. Some settings Haiku 4.5 accepted, such as temperature, return an error with Haiku 5.5. Anthropic's migration guide lists them.
Is Haiku 5.5 as good as Sonnet 5.5?
No, and Anthropic does not claim it is. Here are three tests from its announcement, with each model's success rate:
| Test | Haiku 5.5 | Haiku 4.5 | GPT-6 Luna | Sonnet 5.5 |
|---|---|---|---|---|
| Using a computer (OSWorld 2.1) | 72.4% | 15.7% | 48.9% | 83.9% |
| Reading charts (Chartography) | 46.4% | 6.4% | 29.1% | 61.6% |
| Coding alone in a terminal (Terminal-Bench 4.0) | 39.2% | 0.0% | 16.4% | 70.6% |
The jump from Haiku 4.5 is huge. On the test where the model has to operate a computer by itself, Haiku 5.5 goes from 15.7% to 72.4%.
It also beats GPT-6 Luna, the small model from OpenAI, on these three tests. But Anthropic chose the tests and took the measurements.
Against Sonnet 5.5, the gap grows as soon as the job gets long. In coding, Haiku 5.5 completes 39.2% of the tasks, where Sonnet 5.5 completes 70.6%.
Anthropic itself writes that Sonnet 5.5 and Opus 5.5 are better choices for hard coding work. Haiku 5.5 is built for narrowly scoped jobs.
And speed? Asana, the project management tool, measured over 30% less waiting per task, compared with the model it was using before.
"Fastest" means at standard speed. Anthropic notes that its Opus models go faster still when their fast mode is switched on.
What else changes for Sonnet 5.5 and for subscribers?
Anthropic used this release to announce two other changes:
- Sonnet 5.5 costs less to reread. When a model rereads text it has already been sent, it bills it at a reduced price (that is the cache). That price drops from $0.20 to $0.10 per million tokens, and Anthropic puts the saving at 20% on the long jobs it runs alone;
- Max and Team subscribers get an API credit every month. It comes to $100 for Max 5x, $200 for Max 20x, and up to $500 for a Team plan.
This credit is for running your own tools and automations on the API. The free and Pro plans are not eligible, and whatever is not spent by the end of the billing period is lost.
With $100 of credit and Haiku 5.5, you have enough to draft more than a hundred thousand email replies every month. Enough to test an idea without getting your card out!
Anthropic's help page explains how to activate it from the billing settings. To pick a plan, we have detailed the differences between Claude Pro, Max and Team.
Should you switch to Haiku 5.5?
The answer changes with the way you use AI:
- You chat with Claude to write or think. Try Haiku 5.5 for quick questions, and keep Sonnet or Opus for a deeper subject;
- You have automations running all day. Move one repetitive step to Haiku 5.5, then compare the result and the bill;
- You are building an agent or an app, with code or without. Let the large model decide, and give the small jobs to Haiku 5.5;
- You were already using Haiku 4.5 through the API. Read the migration guide before you change, then recount your costs.
Nothing forces you to pick a single model. Keep the large one for whatever needs thinking, and leave the rest to Haiku 5.5.
Start with one job, the one that comes back most often for you, and see how it goes. 😉
Frequently asked questions about Claude Haiku 5.5
Is Claude Haiku 5.5 free?
At launch, you can pick Haiku 5.5 in the Claude app on a free account, on the web, iOS and Android.
Through the API, it is paid. Anthropic charges $0.10 per million input tokens and $0.50 per million output tokens, for requests of up to 100,000 tokens.
How much does Claude Haiku 5.5 cost through the API?
Announced on 7 October 2026 at $0.10 per million input tokens and $0.50 per million output tokens, it costs ten times less than Haiku 4.5.
This rate applies up to 100,000 tokens per request. Above that, it moves to $0.50 and $2.50. Drafting a thousand email replies comes to about $0.70.
What is the difference between Claude Haiku 5.5 and Claude Sonnet 5.5?
Haiku 5.5 is twenty times cheaper than Sonnet 5.5, but weaker on long jobs. In coding, it completes 39.2% of the Terminal-Bench 4.0 test, against 70.6% for Sonnet 5.5.
Anthropic recommends Haiku 5.5 for short, narrowly scoped jobs, such as summarising or classifying.
Is Claude Haiku 5.5 better than GPT-6 Luna?
According to the tests Anthropic published, yes. On OSWorld 2.1, where the model operates a computer alone, Haiku 5.5 scores 72.4% against 48.9% for GPT-6 Luna.
These measurements come from Anthropic, which chose the tests. Compare the two models on your own job before you decide.
Who gets Anthropic's monthly API credit?
Max and Team subscribers. The credit announced is $100 a month for Max 5x, $200 for Max 20x, and up to $500 for a Team plan.
The free and Pro plans are not eligible. Unspent credit is lost at the end of each billing period.