Anthropic has released its fastest and most affordable small AI model, Claude Haiku 5.5, alongside price cuts for its Sonnet 5.5 model.
Key facts
- •Claude Haiku 5.5 is available on Amazon Web Services, Google Cloud, and Microsoft Azure.
- •The model scored 45.9 percent on the Humanity's Last Exam benchmark without tools and 57.4 percent with tools.
- •Anthropic is introducing new monthly API credits for Claude Max and Team subscribers.
- •Haiku 5.5 is designed to function as a subagent alongside larger models like Sonnet 5.5 and Opus 5.5.
- •The model is the first Haiku-class release to feature adjustable reasoning levels.
Anthropic announced the release of Claude Haiku 5.5 on Wednesday, October 7, 2026, positioning it as the company's most capable and cost-effective small model to date. Designed for high-volume, repetitive tasks, the model is available through platforms including Amazon Web Services, Google Cloud, and Microsoft Azure. The launch also includes price reductions for the Sonnet 5.5 model and new API credits for subscribers.
By the numbers
Performance and Capabilities
Claude Haiku 5.5 demonstrates significant performance improvements over its predecessor, achieving a score of 1,620 on the GDPval-AA v2.1 benchmark compared to 735 previously. The model shows substantial gains in computer-use tasks, scoring 72.4 percent on OSWorld-2.1, and reaches 39.2 percent on the Terminal-Bench 4.0 coding benchmark. Anthropic notes that the model is particularly suited for sub-agent work, summarization, and live customer support.
Pricing and Token Usage
The model offers average cost savings of approximately 75 percent compared to Haiku 4.5. For prompts under 100,000 tokens, which Anthropic states represent 90 percent of previous usage, costs are reduced by up to 90 percent, though prompts exceeding this limit are five times more expensive. Anthropic also cautioned that an updated tokenizer may increase token consumption per task, potentially narrowing the real-world savings. Additionally, the company is halving the price of cache reads for Sonnet 5.5, resulting in a roughly 20 percent cost reduction for agentic workflows.
Safety and Accessibility
Anthropic reports that Haiku 5.5 features tighter cybersecurity safeguards than previous versions, with fewer instances of misaligned behavior. While the model is restricted from penetration testing, organizations can apply for verification programs related to life sciences and cybersecurity. The model is the first in the Haiku class to offer adjustable reasoning levels, allowing users to balance cost and quality.
Advertisement
This article was independently rewritten by ManyPress editorial AI from reporting originally published by The Decoder, The Hindu Technology.
