Anthropic unveils Haiku 5.5, its third AI model in a month ahead of planned IPO

On October 8, Anthropic introduced Haiku 5.5, its third AI model in a month, in preparation for a potential IPO. The model is part of the Claude 5.5 family and is positioned as the cheapest, fastest, and most capable small model in the Haiku class. Designed to handle high-volume and cost-sensitive tasks, Haiku 5.5 can efficiently process tasks such as summaries, compactions, database queries, and classification requests.

When paired with more powerful models like Opus 5.5 and Sonnet 5.5, it demonstrates strong performance in coding and speed-sensitive tasks like live customer support and browser use. Pricing for Haiku 5.5 is significantly lower than its predecessor, Haiku 4.5, costing around 75% less to run. The model scored notably higher than GPT-6 Luna on GDPval-AA v2.1 in knowledge work tasks and outperformed it in computer use tasks.

Haiku 5.5 secured above-average scores in advanced reasoning benchmarks, including Humanity’s Last Exam. Additionally, it boasts an adjustable effort setting, allowing users to prioritize cost or intelligence based on their needs. Anthropic claims that Haiku 5.5 has undergone significant improvements in alignment evaluations, with fewer instances of misaligned behavior and a lower willingness to engage in misuse.

Furthermore, its cybersecurity measures are more restrictive compared to previous models, although not as strict as those applied to other recent AI models.