Claude Haiku 5.5: built for high-volume work, with 1M context

The new Haiku gets a million token window, adaptive thinking, the effort parameter and 128k output tokens.

On 7 October 2026 Anthropic launched Claude Haiku 5.5, which it calls “our most capable model tuned for high-volume and latency-sensitive work”. It can read up to one million tokens at once, write up to 128k tokens and think adaptively.

Key facts

Claude Haiku 5.5 (`claude-haiku-5-5`) launched on 7 October 2026.
1M token context window and up to 128k output tokens.
Adaptive thinking on by default, plus the effort parameter.
Available on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry.
Some settings that worked with Claude Haiku 4.5 can return errors.

What happened

Anthropic describes Haiku 5.5 as built for high-volume, latency-sensitive work, the kind of job where many requests need quick answers. The new version has a very large context window of one million tokens, enough to read long reports, contracts or a big folder of notes in one go, and an output limit of 128k tokens.

The model thinks adaptively by default: it decides how much to reason based on the request. Developers can also use the effort parameter to ask for quicker or more careful answers. According to Anthropic, it is the first Haiku-class model with an adjustable effort setting. By default the thinking text is not shown unless developers ask for a summary.

Anthropic warned developers that code written for Claude Haiku 4.5 may break. Manual thinking budgets, temperature, top_p and top_k settings and assistant message prefill can return errors, as can an older computer use tool version. The same day, Anthropic also added monthly API credits to Claude Max and Team plans.

Why it matters

High-volume models do much of the work you never notice: summaries in apps, quick rewrites, classification, support replies. When a model built for that kind of work can read a million tokens, it can take on longer documents without slowing the whole workflow down.

For everyday users, that usually means faster, more capable assistants inside the tools they already use. The catch is that quick, repeated tasks leave no room for vague instructions. They work best with short, precise prompts: one task, a clear format, explicit limits. Give them that and they are often all you need.

How to try it

Rewriting is a perfect job for a fast model. Open our free prompt Rewrite text more clearly and paste a paragraph or email that feels too long or confusing. In the brackets, say who will read it and the tone you want, for example “Reader: my manager. Tone: direct and polite.”

The prompt asks the model to keep your meaning, make the text shorter and clearer, and list the changes it made. Run it and check three things: the meaning is the same, nothing new has been added, and the list of changes matches what you see.

If you use the API, try the same prompt with Haiku 5.5 and with a larger model. For a task like this, you may find that Haiku 5.5 is all you need.

Try it

Rewrite text more clearly

Frequently asked questions

What is new in Claude Haiku 5.5?

A 1M token context window, up to 128k output tokens, adaptive thinking on by default and the effort parameter, according to Anthropic's release notes of 7 October 2026.

Where can I use Claude Haiku 5.5?

On the Claude API, Claude in Amazon Bedrock, Claude Platform on AWS, Claude on Google Cloud and Claude in Microsoft Foundry.

Will my Haiku 4.5 code still work?

Not always. Anthropic says manual thinking budgets, temperature, top_p, top_k and assistant prefill can return errors with Haiku 5.5, so developers should check the migration notes.

Sources