Claude 3 Haiku

Claude 3 Haiku is the lightweight tier of Anthropic's Claude 3 family, released on March 13, 2024, nine days after Opus and Sonnet. It was positioned from launch as the fastest and most affordable model in its intelligence class, aimed at use cases where latency and cost matter more than frontier reasoning depth: customer support, content moderation, large-scale document processing, and agentic sub-tasks.

Like the rest of the Claude 3 family it ships with a 200K-token context window and native vision input, and it later gained tool use and prompt caching through the Anthropic API. Haiku's defining trait is throughput: Anthropic reported it handling roughly 21,000 tokens per second on short prompts, letting enterprises process thousands of images or hundreds of legal documents for about a dollar. It is a closed, API-only model with no public weights.