Lower-Cost Model Targets High-Volume Tasks

Anthropic launched Claude Haiku 5.5 on October 7, adding the final piece to its Claude 5.5 family that now includes Sonnet 5.5 and Opus 5.5. The new model is built for repetitive, cost-sensitive workloads such as document summarization, data classification, database queries, live customer support, and AI agent operations. It arrives as enterprises seek ways to manage rising AI expenses while maintaining performance across daily processes.

Pricing starts at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Beyond that threshold, rates increase fivefold. For the first time, a Haiku version includes adjustable effort settings that let users balance speed and intelligence against cost.

These rates apply primarily to the types of prompts that made up about 90% of requests to the prior Haiku model. Developers integrating the model into their own applications pay these developer rates directly, with the lower pricing intended to encourage wider adoption among businesses focused on scale.

Performance Benchmarks Show Gains

Benchmarks indicate Haiku 5.5 delivers meaningful improvements over its predecessor. On the Artificial Analysis Intelligence Index, it scored 43, up 26 points from Haiku 4.5, trailing only Sonnet 5.5 by 13 points at maximum effort. It surpassed OpenAI’s GPT-6 Luna at 38 while using roughly three times the output tokens per task at peak settings.

On OSWorld 2.1, a test of multi-step computer operations, Haiku 5.5 achieved 72.4% success. On Terminal-Bench 4.0, which measures agentic task completion, it reached 39.2%, compared with 16.4% for GPT-6 Luna. These results highlight its suitability for practical, non-creative workloads rather than frontier-level reasoning.

Safety Enhancements and Availability

Anthropic introduced new cybersecurity safeguards for Haiku 5.5, the first in the Haiku lineup. Controls target a narrow set of high-risk requests that could enable harmful cyber activity, with most routine business and consumer tasks unaffected. This addition addresses growing concerns as smaller models proliferate in automated systems.

The model is now available on the Claude website, Amazon Web Services, Google Cloud, and Microsoft Azure. Developers can access it via the claude-haiku-5-5 identifier. Anthropic is also rolling out monthly API credits this week: $100 for Max 5x subscribers, $200 for Max 20x, and pooled credits up to $500 for Team plans.

Strategic Context Ahead of IPO

The release fits Anthropic’s broader strategy of completing its 5.5 model lineup nine days after Sonnet 5.5 and 15 days after Opus 5.5. Earlier plans for these releases had been outlined by the company. The timing coincides with preparations for an initial public offering, with marketing potentially starting around November 9 and a possible listing before Thanksgiving.

Rapid revenue growth supports the valuation trajectory. In 2025, revenue reached nearly $4.6 billion after a twelvefold increase from the prior year, though operating losses stood at $8.06 billion alongside $518 billion in projected future infrastructure commitments. Analysts note the emphasis on balancing expanded usage with sustainable margins as the company continues heavy investment in computing resources.

Market observers point to broader AI spending trends as context. Worldwide platforms and models are projected to grow 63% in 2026 to $64 billion from $39 billion in 2025. Smaller models like Haiku 5.5 are gaining traction because they match specific tasks more effectively than larger systems, according to industry analyses.

Company Performance and Market Position

Anthropic has gained ground in enterprise adoption, surpassing OpenAI in the United States based on recent benchmarks covering more than 70,000 businesses. However, 52% of customers using either provider utilize both, indicating mixed loyalty. Business switching rates reached 8% in September, reflecting growing flexibility among buyers.

The company’s heavy infrastructure outlays and rapid scaling efforts continue to shape its path to public markets. As smaller models expand access, Anthropic aims to increase overall usage while covering the costs of continued development and deployment. The Haiku 5.5 release provides another tool for developers navigating these dynamics.