The Claude 3.5 Haiku combines speed with state-of-the-art performance, scoring 40.6% on SWE-bench Verified—outperforming GPT-4o and the original Claude 3.5 Sonnet. Designed for latency-sensitive applications, it improves instruction following and tool use accuracy, ideal for user-facing products or processing large datasets (e.g., inventory records). While initially text-only, image support will follow. Its cost revision (0.80input/0.80 input/0.80input/4 output per MTok) makes it a budget-friendly alternative to larger models without sacrificing capability.
