Anthropic has launched Claude Haiku 5.5, completing a rapid three-model expansion of its Claude 5.5 family and giving the AI company a cheaper, faster model aimed at the high-volume workloads that increasingly define commercial AI.
The launch comes as Anthropic prepares for a planned initial public offering, according to Reuters. Haiku 5.5 follows Claude Opus 5.5 and Claude Sonnet 5.5, both introduced in September, giving Anthropic a three-tier Claude 5.5 lineup in roughly a month.
But the significance of Haiku 5.5 is not simply that Anthropic has added another model. The company is trying to make advanced AI cheaper enough and fast enough for workloads where millions of relatively small requests matter more than maximum reasoning capability.
Anthropic’s smallest Claude 5.5 model is built for volume
Anthropic describes Haiku 5.5 as its cheapest and fastest model to date, designed for repetitive and latency-sensitive work such as summarization, classification, extraction, database queries, live customer support and browser-based tasks.
The company says Haiku 5.5 costs about 75% less to run than Haiku 4.5. For prompts up to 100,000 tokens, pricing is $0.10 per million input tokens and $0.50 per million output tokens. For longer prompts, those prices rise to $0.50 and $2.50 respectively.
That pricing is important because the economics of enterprise AI increasingly depend on inference costs. A model that is slightly less capable than a frontier system can be far more commercially useful when it can handle a large share of routine work at a fraction of the cost.
Anthropic says Haiku 5.5 is available immediately through the Claude platform as well as Amazon Web Services, Google Cloud and Microsoft Azure. Developers can access it through the claude-haiku-5-5 model identifier.
Three models, three different jobs
The Claude 5.5 rollout gives Anthropic a clearer division of labor across its model family.
Claude Opus 5.5, introduced on September 22, is positioned as Anthropic’s leading model for complex work. Anthropic says it delivers major gains on difficult tasks while costing 40% less to run than the previous Opus 5.
Claude Sonnet 5.5, introduced on September 28, is the faster and lower-cost option for well-scoped everyday work, including coding, documents, slides and spreadsheets.
Claude Haiku 5.5 sits at the high-volume end of the range. Its role is less about doing the hardest possible reasoning and more about handling huge numbers of relatively straightforward tasks quickly and cheaply.
That architecture reflects a broader change in how companies deploy AI. Instead of sending every request to the most capable model available, developers increasingly route different tasks to different models based on complexity, latency and cost.
The performance gains are aimed at everyday AI operations
Anthropic’s own evaluations show Haiku 5.5 making substantial gains over Haiku 4.5 in areas such as computer use, knowledge work and reasoning. The company reports a 72.4% result on the offline subset of OSWorld 2.1, compared with 15.7% for Haiku 4.5, and a 45.9% score on Humanity’s Last Exam without tools.
Those numbers should be read as Anthropic’s benchmark results rather than independent proof that the model is better for every workload. More useful for businesses may be the early customer feedback Anthropic published alongside the launch.
Asana said early testing produced more than a 30% reduction in latency for task completions and up to 2.5 times faster inference per agent turn. HubSpot reported a 92.8% average score on its CRM evaluation suite, while Box said Haiku 5.5 scored 11 points higher than Haiku 4.5 on its testing at about half the latency.
The pattern is consistent: Anthropic is selling efficiency as much as raw model intelligence.
Cybersecurity safeguards are becoming part of the product strategy
Haiku 5.5 also introduces a notable safety change. Anthropic says it is the first Haiku model with built-in safeguards for a narrow set of high-risk cybersecurity requests.
The company says those protections are more restrictive than Haiku 4.5’s cybersecurity safeguards but less restrictive than those applied to some of its more capable models. Defensive security tasks remain available, while penetration testing and other techniques that are more directly useful to attackers are restricted.
Anthropic has been expanding its cybersecurity verification program at the same time. On October 6, the company announced a broader version of that program that gives qualifying security professionals access to advanced cyber capabilities with reduced blocking classifiers.
That creates an increasingly explicit distinction between ordinary commercial users and verified organizations that need advanced cybersecurity capabilities. It is also part of a larger effort by AI companies to manage the tension between useful cyber assistance and models becoming easier to misuse.
The IPO backdrop makes the timing more interesting
The three-model rollout is happening while Anthropic is preparing for a potential public-market future. Reuters described the Haiku release as part of the company’s effort to expand its AI lineup ahead of a planned IPO.
For investors, the model family tells a broader story than a single product launch. Anthropic is trying to demonstrate that its technology can support multiple layers of enterprise demand, from difficult coding and reasoning to inexpensive, high-frequency AI calls.
That matters because inference economics can become just as important as benchmark leadership once an AI company operates at large scale. A model that cuts the cost of millions of customer interactions can have a direct effect on margins, cloud spending and the economics of AI agents.
Anthropic is also making changes around its subscription products. With the Haiku 5.5 launch, the company said it was cutting the price of Sonnet 5.5 cache reads in half, making Sonnet roughly 20% cheaper for most agentic work, and introducing a monthly API credit for Claude Max and Team subscribers.
The combined strategy is clear: put a more capable model at each price and performance point, then make it easier for customers to move workloads between them.
What Haiku 5.5 says about the next phase of AI competition
The AI model race is often described as a contest over which company has the smartest system. Anthropic’s latest release points to a different competition: who can make useful intelligence inexpensive enough to run continuously.
That is particularly important for AI agents. An agent may make dozens or hundreds of model calls while completing a larger task. Sending every step to a flagship model can make the economics unattractive. A fast small model can instead handle routine subtasks while a more capable model is reserved for the parts that require deeper reasoning.
Anthropic explicitly positions Haiku 5.5 as a companion to Opus 5.5 and Sonnet 5.5 for agentic workloads. That suggests the company increasingly sees its model family as a coordinated system rather than three independent products.
The commercial challenge will be proving that the performance gains survive outside controlled benchmarks and early customer tests. Enterprise buyers care about accuracy, latency and price, but they also care about reliability, security, data controls and whether a model can be integrated into existing workflows without creating new operational costs.
Haiku 5.5 gives Anthropic a stronger answer on the cost and speed side of that equation. The planned IPO gives the company an additional reason to show that its model portfolio can support a broad and durable commercial business.
Sources: Reuters, Anthropic, Anthropic Newsroom.
