On October 7, Anthropic introduced Claude Haiku 5.5. It is the third model in the Claude 5.5 family released in the past month, expanding the company's artificial intelligence (AI) product lineup ahead of a planned IPO. The model is aimed mainly at tasks such as classification, text summarization, and data extraction, with live customer support, voice agents, and in-app assistants cited as the primary use cases. According to Reuters, the release is the company's next step in its compact-model lineup.
Release Details
Haiku 5.5 joins Anthropic's lineup of smallest and cheapest models. The company sees the price as its main advantage: the model runs 75% cheaper than the previous Haiku 4.5. Input tokens are priced at $0.10 per million, output tokens at $0.50 per million. This rate applies to prompts shorter than 100,000 tokens; for longer prompts the price rises to $0.50 and $2.50 respectively.
According to VentureBeat's calculations, factoring in the new tokenizer and request volumes, workload costs drop by roughly 75% compared to Haiku 4.5. A tokenizer is the system that splits text into the units a model processes; its update means fewer tokens are spent on the same text.
The pricing policy is not limited to Haiku. Along with the release, the cache-read fee in Claude Sonnet 5.5 was cut from $0.20 to $0.10 per million tokens. Max and Team subscribers also receive monthly API credits — from $100 to $500. Cache read is a discounted rate for reusing text the model has already processed; halving it lowers the cost of long conversations and repeated queries.
In addition, the monthly API credits ($100–500) are intended to partially offset testing and development costs for Max and Team subscribers — lowering the financial barrier that kept small teams from trying the new model.
Price Competition: Matching GPT-6 Luna
Haiku 5.5's listed price — $0.10/$0.50 — exactly matches the price of OpenAI's GPT-6 Luna model, according to VentureBeat. This shows intensifying price competition in the compact, fast large language model (LLM) segment: both companies, standing at the same price point, are competing for developers' attention. Price wars are nothing new in this segment: the previous Haiku 4.5 also stood out for its affordability, but the new model simultaneously cut prices sharply and raised benchmark scores. For developers, this means more requests for the same budget or lower costs for the same request volume. The Reuters report notes that Haiku 5.5 is the third major release in this segment, continuing the 5.5 generation that began a month ago.
The low price does not mean limited capabilities. According to Anthropic's published benchmark results, Haiku 5.5 scored 72.4% on the offline computer-use subset of the OSWorld 2.1 test — compared with 15.7% for Haiku 4.5. The company also added a configurable effort parameter to the model, with the default set to medium. This parameter lets users control how much compute the model spends on a single request.
Security and Availability
The release paid special attention to cybersecurity. According to Reuters:
"According to Anthropic, this is the first Haiku model with internal safeguards in place for a narrow set of high-risk cybersecurity requests; most everyday tasks are unaffected."
In other words, the safeguards target a narrow range of dangerous requests, while ordinary use cases remain unchanged. This is seen as a sign that compact models are catching up with larger ones on safety.
Haiku 5.5 is distributed under the API identifier claude-haiku-5-5. The model is available both through Anthropic's own platform and via major cloud providers — Amazon Web Services, Google Cloud, and Microsoft Azure. This multi-cloud availability simplifies integration into existing infrastructure for enterprise customers — clients can access the model through the cloud provider they already use.
Context: Expanding Before the IPO
The Haiku 5.5 launch is seen as a move to fill out Anthropic's product lineup ahead of its planned IPO. The Claude 5.5 family has grown by a third model in a month, and the company now offers new-generation models in all three segments — large, mid-size, and small. The sharp price cuts could change the criteria for choosing a model for startups and companies building API-based products: some tasks that previously required large models now appear achievable with cheaper compact models.
