Anthropic on Wednesday launched Claude Haiku 5.5, rounding out its flagship Claude 5.5 model family with a low-cost, high-speed release for high-volume enterprise tasks.
The new model targets lightweight, repetitive applications such as document summarization, data extraction, live customer support, and browser automation in which businesses process thousands of queries daily. According to Anthropic, Haiku 5.5 delivers average cost savings of 75% compared to its predecessor, Haiku 4.5.
For requests under 100,000 tokens, which account for roughly 90% of Haiku’s historical workload, the price reduction reaches 90%. Developers using Anthropic’s API will pay $0.10 per million input tokens and $0.50 per million output tokens for shorter prompts, aligning directly with rates set by rival OpenAI for its small model, GPT-6 Luna, released late last month. Requests exceeding 100,000 tokens receive a 50% discount off Haiku 4.5 rates.
Haiku 5.5 also introduces an adjustable effort setting, giving developers direct control over compute usage. Users can dial down reasoning levels to lower latency and expenses for simple chores or increase compute intensity when tackling more complex logic.
The launch – which comes as the company reportedly prepares for a landmark initial public offering as early as next month — completes Anthropic’s rollout of its 5.5 generation, following the September releases of Claude Opus 5.5 and Claude Sonnet 5.5.
While Opus is engineered for demanding agentic reasoning and Sonnet serves as a mid-tier model for general workplace operations, Haiku 5.5 is designed to handle high-frequency tasks or act as a budget-friendly sub-agent executing specific sub-tasks for larger models.
Beyond Haiku, Anthropic announced a 50% price reduction for prompt caching on Sonnet 5.5, lowering cache reads to $0.10 per million tokens, a move aimed at slashing overhead for agentic systems that continuously reference identical background contexts. Additionally, the company is rolling out monthly API credits ranging from $100 to $500 for subscribers on its upper-tier and Team plans.
Despite its compact size, benchmark performance indicates notable gains over previous generations and competing small models. On OSWorld 2.1, an evaluation measuring multi-step computer interaction, Haiku 5.5 scored a 72.4% success rate, topping GPT-6 Luna’s 48.9%. On Terminal-Bench 4.0, which tests autonomous command-line execution, Haiku 5.5 registered 39.2%, compared to Luna’s 16.4% and Haiku 4.5’s 0%. Early enterprise testers, including Box Inc. and HubSpot, reported faster processing speeds and improved performance on proprietary evaluation datasets.
Anthropic also highlighted new safety protocols embedded in Haiku 5.5, including default restrictions on penetration testing and high-risk cybersecurity functions, following industry-wide scrutiny over autonomous AI agent behavior.
The aggressive pricing strategy arrives as Anthropic builds out its product ecosystem ahead of a prospective Wall Street debut. Market discussions indicate the company could begin actively marketing an IPO during the week of Nov. 9, setting up a potential stock market debut before Thanksgiving on Nov. 26.
Reports indicate Anthropic is targeting a valuation of approximately $2 trillion. Financial disclosures revealed the company achieved $4.6 billion in sales in 2025 — a twelvefold year-over-year increase — while recording $42 billion in net losses over the same period.
By sharply reducing inference costs, Anthropic aims to drive developer adoption and expand its customer base, even as Wall Street closely evaluates the long-term profitability of high-volume AI deployments.
Haiku 5.5 is available immediately via the Claude web platform, as well as through Amazon Web Services, Google Cloud, and Microsoft Azure.
