Claude Haiku 4.5 (latest)

Claude Haiku 4.5 sits in Anthropic's claude-haiku family and is positioned by third-party catalogs as the fast lane of the Claude lineup, aimed at lightweight agents, everyday office tasks, and responsive chat where latency matters more than maximum depth. It accepts text, image, and PDF inputs and produces text output, which gives it enough flexibility to handle document-heavy requests such as summarizing PDFs, reading images, or extracting content from mixed attachments. The catalog pricing snapshot puts input at roughly one dollar per million tokens and output at about five per million, with cached input at ten cents per million, which keeps it in the budget tier compared with larger Claude models like Opus 5.

In practical terms, Haiku 4.5 is a sensible pick when throughput, cost per call, and short-cycle interactions matter more than the heaviest reasoning. The 200,000-token context window with a 64,000-token output ceiling lets it work through moderately long documents and hold multi-turn agent loops without losing the thread, while reasoning, tool calling, and temperature control are all confirmed by aggregator capability flags. On opencode.ai's normalized scoring it scores perfectly on reasoning but lower on coding than flagship Claude models, reinforcing its fit for general chat, classification, summarization, retrieval-augmented workflows, and lightweight tool-using agents rather than the deepest code generation tasks.