# Claude Haiku [[Anthropic]]'s fastest and most cost-effective model tier, designed for high-throughput tasks where speed and price matter more than maximum capability. Haiku runs 4–5x faster than [[Claude Sonnet]] at a fraction of the cost, making it the default choice for classification, summarization, customer support automation, and any application processing millions of requests. The name follows the poetic convention: a haiku is short, dense, and economical — three lines, seventeen syllables, nothing wasted. Haiku sits at the base of Anthropic's model hierarchy, below [[Claude Sonnet]] (balanced) and [[Claude Opus]] (maximum capability). ## Current version: Haiku 4.5 Released October 15, 2025. Key specifications: 200K-token context window, pricing at $1/$5 per million input/output tokens (one-fifth the cost of Sonnet, one-twenty-fifth the cost of Opus). Haiku 4.5 is the only current Claude model with a context window below 1M tokens — a tradeoff for its speed advantage. ## Release timeline | Version | Date | Notes | |---------|------|-------| | Claude 3 Haiku | March 2024 | Launched with the three-tier naming system | | Claude 3.5 Haiku | October 2024 | Released alongside upgraded Sonnet 3.5 | | Claude Haiku 4.5 | October 2025 | Current version; no generation-5 Haiku announced | ## KB relevance Haiku's significance for the vault is primarily economic and structural. It is the model that makes [[Anthropic]]'s technology accessible at scale to the government agencies, defense contractors, and enterprises the KB tracks — the organizations processing massive document sets, screening communications, or running automated analysis don't use Opus at $25/million output tokens; they use Haiku at $5. This means the [[Anthropic Government Ban]] hit Haiku deployments hardest by volume. When federal agencies were directed to cease using Anthropic products, the bulk of the operational disruption was in Haiku-powered automation — the classification, triage, and summarization pipelines that agencies had integrated into daily workflows. Replacing these with competitor models (or xAI's Grok, as the Pentagon did for [[Project Maven]]) required rewriting integration code across thousands of applications. Haiku also illustrates the [[Dual-Use AI]] access problem: a model cheap enough for mass deployment is a model that reaches the widest range of users, including those the government might prefer didn't have access. The [[Anthropic Model Export Directive]] did not cover Haiku — only [[Claude Fable 5]] and [[Claude Mythos 5]] — but the precedent it set applies to any model tier if capabilities cross the national security threshold in a future generation. ## See Also - [[Claude Sonnet]] — the balanced mid-tier model - [[Claude Opus]] — the premium high-capability tier - [[Claude Fable 5]] — the Mythos-class tier above Opus - [[Anthropic]] — the company behind the model family - [[Dual-Use AI]] — the framework for AI serving both civilian and military purposes