On 7 October 2026, Anthropic released Claude Haiku 5.5 and called it the cheapest, fastest and most capable small model the company has ever shipped. Small models rarely get headlines, because people tend to talk about the biggest and smartest systems. But for most businesses and developers, the small model is the one that does the bulk of the actual work. So it is worth understanding what Haiku 5.5 is built for, and where it fits.
What makes a small model useful
Anthropic offers Claude in three main sizes. Opus is the most capable and the most expensive. Sonnet sits in the middle and handles most everyday professional work. Haiku is the smallest, built for speed and low cost.
The trade-off has always been simple. A small model answers quickly and costs very little per request, but it struggles with long, complicated tasks. That makes it ideal for jobs you need to run thousands or millions of times: sorting support tickets, tagging content, pulling names and dates out of documents, or answering simple customer questions.
What changed with Haiku 5.5 is how much capability Anthropic has packed into that small, cheap package.
The pricing story
The headline number is cost. Anthropic says the average task on Haiku 5.5 costs about 75 percent less than the same task on Haiku 4.5. For requests under 100,000 tokens, the list price is $0.10 per million input tokens and $0.50 per million output tokens.
To put that in perspective, a million tokens is roughly 750,000 English words. A company processing a large volume of customer emails, product descriptions or form submissions every day could see its AI bill fall sharply simply by switching the routine work to this model.
Part of the saving comes from a new tokenizer, the component that breaks text into pieces for the model to read. A more efficient tokenizer means the same text uses fewer tokens, which directly lowers cost.
One thing to watch: the low price applies to requests up to 100,000 tokens. If you regularly send very long prompts, check Anthropic's pricing page for the rate above that threshold before you plan a budget.
Effort control comes to Haiku
Haiku 5.5 is the first Haiku model with adjustable effort levels. Developers can tell the model how hard to think. At low effort, it answers almost instantly. At higher effort, it spends more time reasoning before replying, which improves accuracy on trickier requests.
This matters more than it sounds. Many applications have a mix of easy and hard requests. With effort control, a single model can handle both, keeping simple answers fast and cheap while giving difficult questions a bit more care.
Bigger memory than you would expect
Haiku 5.5 has a context window of one million tokens and can produce up to 128,000 tokens of output. The context window is how much text the model can consider at once. A million tokens means it can read very long documents, large chunks of a codebase or long conversation histories in a single request.
For a small model, that is unusual. It opens up uses like summarising an entire policy manual or scanning a large set of contracts for specific clauses, at a fraction of what a larger model would cost.
Where Haiku 5.5 shines
Anthropic positions Haiku 5.5 for a specific set of jobs:
- Classification and tagging, such as sorting incoming messages by topic or urgency.
- Summarisation of emails, articles, meeting notes and reports.
- Query processing and database work, such as turning a plain-English question into a structured lookup.
- Live customer support, where speed matters as much as accuracy.
- Sub-agents inside larger AI systems, where a bigger model plans the work and hands smaller pieces to Haiku.
- Browser automation and computer use, such as filling forms or navigating websites.
That last point is where the biggest jump shows up. On OSWorld 2.1, a test of how well a model can operate a computer, Anthropic reports Haiku 5.5 scoring 72.4 percent, compared with 15.7 percent for Haiku 4.5. On Terminal-Bench 4.0, a test of agent-style programming in a command line, it went from 0 percent to 39.2 percent. These are Anthropic's own figures, so independent testing will tell us more over time, but the direction is clear: the small model is now genuinely usable for agent work.
When you should still use Sonnet or Opus
Haiku 5.5 is not a replacement for the larger models. Anthropic itself says Sonnet 5.5 and Opus 5.5 remain the better choice for complex, multi-step work. If you are asking for a detailed legal analysis, a tricky piece of software architecture, or a long report that needs careful reasoning from start to finish, the larger models will still produce better results.
A sensible approach many teams use is routing. Simple, high-volume requests go to Haiku. Harder requests go to Sonnet. The most demanding work goes to Opus. This keeps quality high where it matters and costs low everywhere else.
Where you can use it
Haiku 5.5 is available on the Claude Platform for developers, through Amazon Web Services, Google Cloud and Microsoft Azure, and inside Claude Code, Anthropic's coding tool. Alongside the launch, Anthropic also halved the cache read price for Sonnet 5.5 and added beta support for computer and browser control to its Python and TypeScript software development kits.
What this means for Indian businesses and developers
For startups and small companies in India, cost has always been one of the biggest barriers to building with AI. A customer support bot or a document processing tool that handles thousands of requests a day can become expensive quickly on a large model. Haiku 5.5 changes that calculation. Tasks that once needed careful budgeting can now run at a very low cost per request.
It also makes experimentation cheaper. A developer can test an idea, run it on real data and measure results without worrying much about the bill.
The bottom line
Claude Haiku 5.5 is not trying to be the smartest model in the room. It is trying to be the most useful model for the work that happens thousands of times a day. With much lower prices, effort control, a one million token context window and a big jump in computer-use ability, it is a strong option for anyone building high-volume AI features. Start with Haiku for routine tasks, keep the bigger models for the hard problems, and test both on your own data before deciding.
