Anthropic - Claude Haiku 4.5

A highly optimized, state-of-the-art small model engineered for ultra-low-latency enterprise and large-scale agentic workloads. Supporting a 200,000-token context window, it delivers a massive processing speed of roughly 21,000 tokens per second for near-instant data ingestion. Remarkably, it achieves near-frontier performance, scoring 73.3% on SWE-bench Verified to match the coding, computer use, and agent proficiency of legacy Sonnet 4. It is specially tuned for multi-agent orchestration, running complex subtasks in parallel under a central planner model. This makes Haiku 4.5 the ultimate economic backbone for real-time customer service agents, rapid code refactoring, and high-volume multi-file data synthesis.