Claude Haiku 5.5: what's new and how to use it for cheap workflows

Anthropic released Claude Haiku 5.5 on October 7, 2026. The company calls it its cheapest, fastest and most capable small model. It is built for high-volume jobs where each task is simple and repeated thousands of times, such as sorting messages or pulling details out of documents.

What it costs

For prompts up to 100,000 tokens, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens. A token is a small chunk of text, roughly a word or part of one. By comparison, the pricing listed for Sonnet 5.5 is $2.00 and $10.00, and for Opus 5.5 it is $4.00 and $20.00.

The 90% cut in list prices is not the whole story. Haiku 5.5 uses a newer tokenizer, so the same text counts as roughly 30% more tokens. About 10% of requests also run over 100,000 tokens. Anthropic puts the average saving at around 75%.

What it does

Anthropic points the model at summaries, compaction, database queries and classification. It can also act as a helper, called a subagent, working under Opus 5.5 or Sonnet 5.5. It adds adaptive thinking, where the model decides how long to reason, and a 1 million token context window, meaning how much text it can read at once.

The benchmark scores come from Anthropic itself. The launch page shows Sonnet 5.5 ahead on all four, for example 70.6% on Terminal-Bench 4.0. The bigger models remain the better pick for harder work.

What can break

Teams moving over from Haiku 4.5 should check their code first. Anthropic lists three breaking changes:

• Sending temperature, top_p or top_k now returns an error.
• Ending a conversation with an assistant prefill returns an error.
• Manual thinking with budget_tokens returns an error.

Behavior changes too. A reply can start with a thinking block, and thinking tokens count toward the length limit. Safety filters can also decline a request, so software must handle a refusal. Anthropic says the cybersecurity safeguards are stricter than in Haiku 4.5 and block penetration testing.

Who it affects

Mostly developers and the businesses that hire them. The model is available on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure. Anthropic's deprecation page lists the earliest retirement date as October 7, 2027.

What to do

A business running chatbots or document sorting should ask its developer or vendor whether the savings hold on real workloads. Test on a small batch first, and compare cost and accuracy against the current setup.