Claude Haiku 5.5: Up to 90% Cheaper, With a 100K-Token Catch

Anthropic released Claude Haiku 5.5 on October 7, 2026, calling it the cheapest, fastest and most capable small model it has shipped. On the API it costs $0.10 per million input tokens and $0.50 per million output tokens, up to 90% less than Haiku 4.5. It matters to any business that pays for repetitive AI tasks.

What we know

What it is: a model for high-volume, cost-sensitive work such as summaries, classification and database queries, according to Anthropic's announcement of October 7, 2026.
Price for prompts up to 100,000 tokens: $0.10 per million input tokens and $0.50 per million output tokens, per the announcement and the pricing page in the documentation. Haiku 4.5 costs $1 and $5.
Price above 100,000 tokens: $0.50 and $2.50, per the same page.
The cut, in Anthropic's words: 90% on requests up to 100,000 tokens and 50% above that. On average, around 75% less, after counting that the model uses more tokens per task.
Batch: half price: $0.05 and $0.25 up to 100,000 tokens.
More tokens for the same text: Anthropic's migration guide says the same text produces roughly 30% more tokens than on Haiku 4.5, and that the exact increase depends on the content.
Capacity: a 1-million-token context window and up to 128,000 output tokens, per the models table. It is the first Haiku with an effort setting, to trade cost against reasoning.
Where it runs: on the Claude Platform, under the model ID "claude-haiku-5-5", and on Amazon Web Services, Google Cloud and Microsoft Azure.
Benchmarks: they are Anthropic's own. On OSWorld 2.1, a test of operating a computer, it reports 72.4% for Haiku 5.5, against 15.7% for Haiku 4.5 and 48.9% for OpenAI's GPT-6 Luna.
Also in the announcement: cache reads on Claude Sonnet 5.5 drop from $0.20 to $0.10 per million tokens, and Max and Team plans get a monthly API credit this week: $100 on Max 5x, $200 on Max 20x and up to $500 on Team.

What changes and what doesn't

The math on repetitive work changes. Sorting thousands of emails or search queries, summarizing documents or pulling the fields out of an invoice now costs a fraction of what it did on the previous Haiku.

The real saving is not a round 90%. The price per token falls 90%, but the same text takes more tokens. Using the 30% in the migration guide, the saving on identical work comes to about 87%. That figure is an estimate from this article.

Nothing changes for hard problems. Anthropic says Sonnet 5.5 and Opus 5.5 remain the better choice for complex agentic coding.

The switch is not automatic. Haiku 4.5 is still listed as active on Anthropic's deprecations page, which only says it will not be retired before October 15, 2026. To get the new price, you have to migrate.

The list price now matches the model Anthropic benchmarks against. OpenAI's pricing page showed $0.10 and $0.50 for GPT-6 Luna on October 3, 2026.

How to tell if it affects you, and what to do today

Find out whether you run a Haiku model. If your site has a chatbot, a form classifier or automatic summaries, ask whoever built it which model sits behind it.
Run the numbers with your own volume. One thousand tasks of 2,000 input tokens and 500 output tokens cost $4.50 on Haiku 4.5. On Haiku 5.5, with 30% more tokens, they come to about $0.59. That is an estimate from this article, and your volume will differ.
Watch the 100,000-token step. If you send long documents, count tokens with the new model. A prompt that took 80,000 tokens on Haiku 4.5 can cross 100,000 and pay the higher rate. With that 30%, the line sits near 77,000 Haiku 4.5 tokens, and above it the estimated saving shrinks to about 35%.
Don't just swap the model name. The migration guide asks you to recount tokens, revisit the output limit ("max_tokens"), change how thinking is configured and remove the "temperature", "top_p" and "top_k" parameters. Test in a separate environment first.
Test on your own tasks. Take twenty real cases, in the language your customers write in, with the right answer for each, and compare your current model against Haiku 5.5. The benchmarks in the announcement are Anthropic's.
On Max or Team, claim the credit. You have to link a Console organization from your billing settings, after seven days on the plan. Unused credit expires at the end of each billing cycle, per Anthropic's help page. Free, Pro and Enterprise plans don't get it.
If you use Sonnet 5.5 with caching, check your bill. The announcement puts cache reads at $0.10 from October 7. That day, the main table on the pricing page still showed $0.20.

Sources

Updates: if Anthropic changes the prices, corrects the pricing table or sets a retirement date for Haiku 4.5, it will be added here with a link.