Anthropic launches Claude Haiku 5.5, its cheapest and fastest small model yet
IA en un minuto newsroom · Editor: Jon Elgezabal

In 30 seconds
Haiku 5.5 is Anthropic's new small model: on average it comes in around 75% cheaper than Haiku 4.5, answers faster than any other model from the company at standard speed and lets users choose the effort level. Sonnet 5.5 is also getting cheaper, because reading from its cache now costs half as much. This week, Max and Team plans start receiving a monthly balance for the API, up to $500 on Team.
Anthropic has released Claude Haiku 5.5, which it presents as the most capable small model it has ever launched, and also the fastest and the cheapest.
The model targets work that comes in large volumes and where cost matters, such as summarizing or classifying. Anthropic says it costs around 75% less than Haiku 4.5 on average and is the quickest model it has made so far at standard speed (the Opus models are faster in Fast Mode), which suits it particularly to live customer support and to operating a browser. It is the first Haiku, too, to offer an effort setting for trading off cost against intelligence.
On requests of up to 100,000 tokens, the price is $0.10 per million input tokens and $0.50 per million output tokens, against $1 and $5 for Haiku 4.5; beyond that size it is $0.50 and $2.50. Anthropic cautions that the new tokenizer uses somewhat more tokens per task and says the 75% average already allows for that. The model is available from today on every platform, Microsoft Azure, Google Cloud and Amazon Web Services included.
Anthropic is also cutting in half what Sonnet 5.5 charges for cache reads (reusing text that has already been processed), from $0.20 to $0.10 per million tokens. As those reads account for a good part of what models consume, Sonnet 5.5 works out around 20% cheaper for most agent work.
Starting this week, Claude Max and Team subscribers will also receive a monthly credit for building tools, apps and agents with the API: $100 or $200 a month on Max, depending on the tier (5x or 20x), and as much as $500 on Team, shared among its users.
Why it matters · analysis and opinion
More than a new model, the announcement is a price cut on three fronts at once: the small model, the Sonnet cache and a balance for the API on paid plans. For anyone already automating repetitive work, such as classifying requests, summarizing documents or handling live customer support, it can make uses that used to be too expensive pay off. The effort setting adds a practical choice: not every task needs the same capability, and now it can also be tuned on the cheap model. What is still missing is checking the real quality outside Anthropic's own tests, and keeping in mind that the new tokenizer uses a little more per task. The sensible move is to try it on a sample of your own work before switching models.
Official source: Anthropic · Written with the help of AI: how we make the news


