Anthropic on Wednesday released its cheapest and fastest Claude Haiku 5.5 model. The model runs at approximately 75% lower cost than its predecessor and is designed for high-throughput tasks such as document summarization and real-time customer support. Its input token pricing is $0.10 per million, and output tokens are $0.50 per million (for prompts up to 100,000 tokens). Benchmarks show that Haiku 5.5 outperformed OpenAI's GPT-6 Luna model in tests such as OSWorld 2.1, Terminal-Bench 4.0, and GDPval-AA v2.1. Additionally, Anthropic also halved the cached read price for Claude Sonnet 5.5 to $0.10 per million tokens.