Haiku 5.5 prices the common call at a tenth of the last small Claude model

Anthropic put Claude Haiku 5.5 on sale on 7 October and priced it, for the common case, at one tenth of the previous small model. For prompts up to 100,000 tokens, input is 10 cents per million tokens and output is 50 cents. Haiku 4.5 charged 1 dollar for input and 5 dollars for output. Cache reads on the new model are 1 cent per million tokens below that 100,000-token line, against 10 cents on Haiku 4.5. Cache writes are 12.5 cents, against 1.25 dollars. Above 100,000 tokens the new prices rise to 50 cents of input, 2.50 dollars of output, 5 cents of cache reads and 62.5 cents of cache writes. Anthropic says the average bill to run the model is about 75 percent lower.

The 100,000-token split is the part that decides whether the headline cut is the cut a buyer actually gets. Anthropic says about 90 percent of requests to the previous Haiku fell under that line. A team whose prompts are short summaries, classifications, database lookups or compactions will see something close to the tenfold drop on the sticker. A team that stuffs long transcripts into every call will pay the higher band, which is still half the old Haiku input price and half the old output price, but no longer a tenth.

Where it sits in the 5.5 family

Haiku 5.5 is the third model in the Claude 5.5 series, after Opus 5.5 and Sonnet 5.5. Anthropic describes it as the cheapest and fastest small model it has released, built for high-volume work and as a subagent beside the larger models on coding jobs. It is the first Haiku with an adjustable effort setting. The levels are Low, Medium, High, Xhigh and Max. A caller can spend less and accept a thinner answer, or spend more and ask for a longer pass, without switching model names.

The same announcement cuts the price of Sonnet 5.5 cache reads in half. Anthropic says that makes Sonnet about 20 percent cheaper on most agentic work, the kind of job in which a model reads a long cached prompt many times and writes a shorter result. The Haiku cut and the Sonnet cache cut point the same way. Anthropic is making the repeated read, not the first write, the cheap part of the bill.

Who can call it, and under what name

Haiku 5.5 is available on claude.ai for Free, Pro, Max, Team and Enterprise accounts, on the web and on iOS and Android. Developers call it as claude-haiku-5-5 on the Claude Platform. It is also on Amazon Web Services, Google Cloud and Microsoft Azure, and inside Claude Code. Biology safeguards match those on Sonnet 5, Sonnet 5.5 and Opus 5, according to Anthropic's launch note. The company says the model is a step up on Haiku 4.5 in coding, tool use, computer use and agent tasks. It has not published a single headline score that a buyer can check against a rival's table without reading the full evaluation chart.

The practical comparison for a finance or support team is the price at the volume they already run. A million short input tokens that cost a dollar last month cost 10 cents on the new sticker, if they stay under 100,000 tokens a call and if cache reads are doing the work Anthropic says they do. A customer-support desk that needs a reply inside a second is the use Anthropic names first. A coding agent that farms out file summaries to a small model and keeps the hard edit on Sonnet is the use it names second.

What the price does not say

A cheaper small model does not, by itself, move Anthropic closer to or further from a public listing. The launch sits in a month when the company has also put out Opus 5.5 and Sonnet 5.5, and reporters have described the pace as preparation for a possible offering. The filing, if it comes, will be a document with revenue and margin. Thursday's document is a price card. The card is specific enough to audit. A buyer who was paying Haiku 4.5 rates can multiply last month's token counts by the new rates and see the difference before the next invoice. That arithmetic, not the adjective in the launch headline, is the fact the release adds.