Anthropic launches Claude Haiku 5.5, third model in the family in a month, with 75% lower price

The most affordable model in the Claude 5.5 series arrives with pricing tiers that change based on prompt size and with Anthropic's first blocks for high-risk cybersecurity requests.

Anthropic launched Claude Haiku 5.5 this Wednesday (October 7), the third model in the Claude 5.5 family to arrive in less than a month, according to a Reuters report published by the Economic Times. The launch comes as the company expands its AI model lineup ahead of a planned IPO, which helps explain the accelerated pace of releases.

Haiku has historically been Anthropic's entry-level line: smaller, faster, and cheaper than the Sonnet and Opus models of the same generation. Haiku 5.5 follows that logic, with a 75% price cut compared to its predecessor, Haiku 4.5, according to the company.

What Haiku 5.5 Is For

According to Anthropic, Haiku 5.5 was built for the following use cases:

• Text classification
• Summarization
• Structured information extraction
• Live support
• Voice agents
• In-app assistants

This scope matters for architecture decisions: Haiku 5.5 is not a candidate to replace a Sonnet or Opus in pipelines that require multi-step reasoning, but it is the natural model for routing simple, repetitive calls that today consume unnecessary budget on larger models.

Price: 75% Cheaper, But in Two Tiers

Anthropic states that Haiku 5.5 costs 75% less than its predecessor, Haiku 4.5. The published pricing table shows that the price changes depending on prompt size:

The price jump in the tier above 100,000 tokens is relevant in practice: a pipeline that operates close to the limit but occasionally exceeds 100,000 tokens on larger prompts pays up to 5 times more for that specific call. Teams designing the API integration need to monitor average size and peak tokens per request to avoid being surprised by the bill.

Anthropic did not disclose, in the same announcement, the exact price of Haiku 4.5 before the cut. Doing the simple math from the reported percentage (a 75% reduction equals a quarter of the previous price), the up-to-100,000-token tier of Haiku 4.5 would have been around US$0.40 for input and US$2.00 for output. This is an arithmetic inference from the published figure, not a number confirmed by the company, and it may not hold equally for both tiers.

What Changes for Teams Running AI in Production in Brazil

For Brazilian teams that treat token cost as a meaningful budget line, Haiku 5.5's price cut lowers the barrier to automating high-volume tasks, such as support ticket triage, lead classification, or document data extraction. Since pricing is charged in dollars, exchange rate fluctuation remains a factor that product and finance teams need to track alongside the per-token price, especially in applications with a high volume of daily calls.

The split into two pricing tiers also changes how it's worth designing the prompt. Keeping context lean, cutting redundant few-shot examples, and summarizing conversation history before sending it to the model stop being just good prompt engineering practices and start having a direct, measurable impact on the bill, since exceeding 100,000 tokens multiplies the cost of the call.

First Haiku With Security Safeguards

Anthropic described Haiku 5.5 as the first model in the Haiku line with built-in safeguards for a restricted set of high-risk cybersecurity requests. The company states that most everyday tasks will not be affected by these restrictions.

The announcement does not detail which specific request categories trigger the block, nor how the restriction is technically implemented (a pre-generation filter, a downstream classifier, or a tweak to the model itself). For teams building offensive security tools, AI-assisted pentesting, or red team automation, it's worth testing early whether Haiku 5.5 fits the use case before migrating production to the model, since there is no public guarantee that legitimate security tasks fall outside the block's scope.

The Context of the Launch

Haiku 5.5 is the third model released under the "Claude 5.5" umbrella in less than a month, according to Reuters, which suggests Anthropic had already launched other variants of the 5.5 generation (presumably in the Sonnet and Opus lines) before this announcement. The report does not detail the other versions but frames the move as part of the company's portfolio expansion ahead of a planned initial public offering.

This accelerated release pace is itself a data point: companies preparing for an IPO tend to want to show product traction and revenue expansion, and launching three models in a month is a direct way to signal that to the market. For those following the race among LLM providers, Anthropic's move reinforces the pricing pressure that was already coming from competitors, with successive cuts at the entry layer of model families.

For now, it remains unclear what the exact price of Haiku 4.5 was before the cut, the precise scope of the new cybersecurity safeguards, and whether Anthropic will extend the same two-tier pricing scheme to upcoming updates of the Sonnet and Opus lines.

Fonte: ET Tech (Índia), via Reuters (https://economictimes.indiatimes.com/tech/artificial-intelligence/anthropic-launches-third-claude-5-5-model-expanding-ai-lineup-before-planned-ipo/articleshow/134773835.cms)