Claude Haiku 5.5: What's New, Pricing, and Who Should Use It

Anthropic released Claude Haiku 5.5 on October 7, 2026, and its pricing page describes it as the fastest, most cost-efficient Claude model. Through the Claude APIAPIAn API (application programming interface) is a defined way for one piece of software to request data or actions from another. AI companies offer APIs so developers can build their models into other products.Read the full definition it costs $0.10 per million input tokens and $0.50 per million output tokens on prompts up to 100,000 tokens, according to that page, a tenth of Haiku 4.5's $1 and $5. The model is also available in the Claude app.

The rate fell 90%, but the bill falls less, and for a small business the saving can be a few dollars a month. What matters for an owner is three decisions: which model to pick in the Claude app, whether to change the model in a tool or automation, and how to use the monthly API credits that now come with Max and Team plans. Two things are not yet known: which plans show Haiku 5.5 by name, and how fast it is in independent tests. No hands-on test is included; the article works from Anthropic's pages and help center and from Artificial Analysis's published scores.

Claude Haiku 5.5 at a glance

• Released: October 7, 2026
• API price per million tokens: $0.10 input and $0.50 output on prompts up to 100,000 tokens; $0.50 and $2.50 above that; Haiku 4.5 costs $1 and $5
• Where: the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry; the Claude app on the web, iOS, and Android
• New for Haiku: an effort setting; text and images as input
• Shipped alongside: Sonnet 5.5 cache reads cut from $0.20 to $0.10 per million tokens, and monthly API credits on Max and Team plans

What is new in Claude Haiku 5.5?

1. A 90% lower rate and a smaller drop in the bill

Anthropic prices Haiku 5.5 90% below Haiku 4.5 on prompts up to 100,000 tokens and 50% below on longer ones. Two things narrow the gap on an actual bill. Haiku 5.5 uses a newer tokenizer, so the same text counts as about 30% more tokens, according to Anthropic's documentation. And thinking adds output tokens: in the Claude app it cannot be turned off for Haiku 5.5, and on the API it can be turned off only at effort levels high and below, according to the Claude Help Center. Anthropic's own estimate is that Haiku 5.5 costs around 75% less to run on average, based on Haiku 4.5 traffic in which 90% of requests were under 100,000 tokens. A different workload can land above or below that figure.

2. An effort setting on a small model

Effort trades thoroughness against tokens on each request. Artificial Analysis shows the range in independent tests: Haiku 5.5 scores 29 on its Intelligence Index at Low effort and 43 at Max, against 17 for Haiku 4.5 and 56 for Sonnet 5.5. Its cost per index task starts at $0.02 at Low and varies up to 8.7 times across levels. As of October 8, 2026, the firm had not published a speed measurement. Anthropic positions Haiku 5.5 for narrowly scoped, high-volume work such as summaries and classification, and keeps Sonnet 5.5 and Opus 5.5 for complex agentic coding.

3. Long prompts move to a higher price tier

Haiku 5.5 accepts up to 1 million tokens in one request, against 200,000 for Haiku 4.5. Prompts above 100,000 tokens cost five times more per token, $0.50 for input and $2.50 for output. A larger window means more material fits in one request; it does not mean every detail in that material gets picked up, so answers about a long document need the same checking as answers about a short one.

What Claude Haiku 5.5 changes for a small business

Each of the three decisions has a different answer, and only one of them depends on the 90% price cut.

In the Claude app: match the model to the task

The Help Center describes a model menu next to the send button: the current model's name appears there, clicking it switches models, and "More models" opens the rest. It lists Haiku 5.5 among the models with an effort selector and says Low and Medium effort work well for routine tasks and stretch usage further. The pricing page lists a Haiku model on the Free, Pro, and Max plans without naming the version, and Anthropic's documentation shows Haiku 5.5 running in claude.ai and the mobile apps. Which plans show Haiku 5.5 by name, and whether it replaced Haiku 4.5 in the menu, was not checked for this article; the menu itself shows which model is answering.

A workable rule for choosing: a fast model at low effort suits tasks where the result is quick to check and a mistake is cheap, such as rewording a reply or summarizing call notes. Where a mistake is expensive, such as a quote, contract terms, or a refund policy, a careful read by a person matters more than which model wrote the draft. The lesson on using AI for work in the productivity course uses the same test: useful material in, and a quick way to review the result.

In your own automation: when the lower rate matters

A rough calculation from the published prices shows where the cut stops mattering. A support reply that reads 3,000 tokens and writes 300 costs about $0.0045 on Haiku 4.5. On Haiku 5.5 the same text counts as more tokens and costs about $0.0006, before thinking tokens. At 1,000 replies a month that is roughly $4.50 against $0.60, a saving of about $3.90, which a developer's time to switch and retest can easily exceed. At 100,000 replies a month the same arithmetic gives about $450 against $60, and the switch starts to pay. The same logic applies to tools you pay for: those that switch from Haiku 4.5 to Haiku 5.5 may reduce their model costs, depending on the workload and settings, and whether that changes your subscription price is the vendor's decision. There is no deadline forcing a move: Anthropic's model deprecations page lists Haiku 4.5 as active, with retirement not sooner than October 15, 2026, and no deprecation notice as of October 8, 2026.

Monthly API credits on Max and Team plans: how to use them

According to the Help Center, Max 5x subscribers get $100 a month, Max 20x subscribers $200, and Team plans $20 per Standard seat and $100 per Premium seat, pooled and capped at $500; a team of three Standard seats gets $60. The steps:

• Who can claim: the Max subscriber, or a Team Owner or Primary Owner, after seven days on the plan. The person also needs the Owner, Admin, or Billing role in a Claude Console organization, or can create one during the claim.
• Where: on claude.ai, Settings > Billing for Max or Organization settings > Billing for Team, then "Link organization" in the API credits section. Only one Console organization can be linked, and changing it later goes through support.
• How to check: the Console shows the credits and their expiry date under Settings > Billing.
• When it runs out: API requests stop until the next month's credits, unless the organization has bought credits or turned on auto-reload. Nothing is charged to the Claude plan.

The credits do not raise usage limits in the Claude app, and unused credits expire at the end of each billing cycle. One concrete use for an owner without a developer: before paying for a tool to be switched to Haiku 5.5, run ten real customer questions through Haiku 5.5 in the Console Playground, which the credits cover, and compare the answers with what the tool gives now.

Whichever model writes a draft, a person still reads it before a client does. Empacer is a voice-first AI business assistant for small service business owners. You say what needs doing, review the finished work, and decide whether it goes out.

Who should try it now and who can wait

• Fewer than a few thousand AI requests a month: no reason to request a migration; the saving is a few dollars.
• Tens of thousands of requests, or long documents through the API: ask for a test on your own tasks, with thinking tokens counted.
• A Max or Team subscription: claim the credits if someone will use them this cycle, since they do not roll over.
• The Claude app for quick tasks: Haiku 5.5 at low effort suits checkable, low-stakes drafts and stretches usage further.

What is published and what is not