Claude Haiku 5.5 cuts the cost of sorting support tickets

Anthropic released Claude Haiku 5.5 on October 7, 2026. It is the company's new small model, priced for high-volume, narrow work such as sorting, tagging and summarizing. The same day, Anthropic added monthly API credits to its Max and Team plans. A small team that already pays for Claude Team can now try the model on a real job at no extra cost.

This piece covers what shipped, who it matters to, and one job to try this week: support ticket triage. Emeri did not test Haiku 5.5 for this article. Everything below comes from Anthropic’s published pages, which are listed in the sources.

What shipped

Claude Haiku 5.5 is designed for high-volume, cost-sensitive tasks.

Anthropic names summaries, database queries and classification as typical work for the model. It also names live customer support and browser use, because Haiku 5.5 is its fastest model at standard speed. The model id on the Claude API is claude-haiku-5-5.

Haiku 5.5 is priced by prompt length. The pricing page lists these rates per million tokens:

• Prompts up to 100,000 tokens: $0.10 input, $0.50 output.
• Prompts over 100,000 tokens: $0.50 input, $2.50 output.
• Haiku 4.5, for comparison: $1 input, $5 output.
• Batch API, for prompts up to 100,000 tokens: $0.05 input, $0.25 output.

Anthropic says prompts up to 100,000 tokens made up about 90% of requests to Haiku 4.5. It estimates that Haiku 5.5 costs about 75% less to run on average. That estimate allows for a new tokenizer, which counts more tokens for the same text.

Haiku 5.5 is the first Haiku model with an effort setting. A lower setting makes the model think less and cost less. A higher setting buys more reasoning at a higher cost. The launch post also cut the price of Sonnet 5.5 cache reads by half, to $0.10 per million tokens.

The new API credits

The release notes dated October 7 say Max and Team plans now include monthly credits for the Claude Platform. The credits pay for your own apps and agents that call the API. The help article sets the amounts:

• Max 5x: $100 a month. Max 20x: $200 a month.
• Team: $20 per Standard seat and $100 per Premium seat, pooled into one balance and capped at $500 a month.
• Free, Pro and Enterprise plans are not eligible.
• A plan must be active for seven days before its credits can be claimed.

The help article gives an example: a team with three Standard seats and two Premium seats receives $260 a month. Unused credits expire at the end of each billing cycle. If the balance runs out and the organization holds no other credits, API requests stop until the next deposit.

The credits cover the Messages API, the Message Batches API, the Console Playground, Claude Managed Agents and the Claude Agent SDK. They do not cover interactive Claude Code, extra usage in the Claude apps, or Claude on Amazon Bedrock, Google Cloud Vertex AI or Microsoft Foundry.

Who this matters to

It matters most to a team with a repetitive job that has a fixed set of answers. Support ticket triage fits that shape. So do tagging inbound sales enquiries, pulling fields from invoices, and summarizing call notes into a set format.

Here is Emeri’s arithmetic for triage. It is an estimate from the published prices, not a measurement. Suppose each ticket and its instructions come to 1,500 input tokens and the answer is 100 output tokens. At the rates for short prompts, one ticket costs $0.0002. Ten thousand tickets cost about $2. Thinking tokens count as output, so a higher effort setting raises that figure.

It matters less for complex work. Anthropic says Sonnet 5.5 and Opus 5.5 remain the better choices for complex agentic coding. It describes Haiku 5.5 as best for narrowly scoped tasks.

What to do this week

The steps below set up a triage trial without touching live customers. They use the Console Playground or a short script, both of which the credits cover.

1. Claim the credits. On a Team plan, a Primary Owner or Owner opens Organization settings, then Billing, and selects “Link organization” under API credits. You can link only one Console organization, so choose the one you will build in.
2. Set a spend limit. The help article says everyone with an API key in the linked organization draws from the same balance. Set workspace spend limits in the Console before you share a key.
3. Build a test set. Export 50 recent tickets. Remove names, email addresses, order numbers and anything else that identifies a customer. Have the person who triages today label each ticket with the right category and urgency.
4. Write the categories down. Use the categories your team already routes by, with one line on what belongs in each. Add “unclear” as an allowed answer.
5. Run the prompt below on each ticket with model claude-haiku-5-5. Start at a low effort setting.
6. Compare every answer with the human label. Count matches, and read every miss.
7. Raise the effort setting once and run the misses again. Keep the lowest setting that meets your bar.

The prompt asks for the deciding words from the ticket. A reviewer can then check each answer against the ticket in seconds. The “Unclear” option gives the model a way to say it does not know, instead of forcing a guess into a category.

If you already use Haiku 4.5

Anthropic’s migration guide lists changes that can break existing code. These are the ones most likely to affect a small team:

• Token counts rise. The guide says the same text produces about 30% more tokens on Haiku 5.5. Recheck max_tokens limits and cost estimates.
• Sampling settings go. Remove temperature, top_p and top_k from requests.
• Thinking works differently. A fixed thinking budget returns an error. Use the effort setting instead.
• Refusals need handling. Haiku 5.5 runs safety classifiers that can decline a request, with no server-side fallback. Your code should catch a stop_reason of “refusal”.
• Priority Tier is not supported on Haiku 5.5. Plan capacity separately if you rely on it.

The guide says users of Claude Managed Agents need to change only the model name.

What we did not test

Emeri did not run Haiku 5.5 for this article. We have not measured its triage accuracy, its speed or its real cost per ticket. The benchmark scores and customer results in Anthropic’s launch post come from Anthropic and its early-access customers. Your own 50-ticket test is the number to trust for your queue.

Emeri’s certification grading runs on Claude models.