Introducing Claude Haiku 4.5 (www.anthropic.com)

Anthropic today launched Claude Haiku 4.5, a new small/efficient model positioned as a near‑frontier option for latency‑sensitive applications. Haiku 4.5 is available everywhere (Claude API, Claude Code, Amazon Bedrock, Google Vertex AI) and is priced at $1 per million input tokens and $5 per million output tokens. Anthropic says Haiku 4.5 matches much of the coding quality of its frontier Sonnet 4.5 while running up to 4–5× faster and costing a fraction of larger models, making it aimed at real‑time chat assistants, customer service agents, pair programming/Copilot, and multi‑agent orchestration where responsiveness matters. On benchmarks Anthropic reports Haiku 4.5 achieves ~90% of Sonnet 4.5’s performance on Augment’s agentic coding evaluation, 73.3% on SWE‑bench Verified (50 trials, 128K thinking budget, tool use), and strong gains on tasks like computer use and slide text generation (65% vs 44% for a premium model). Testing used extended thinking budgets, tool scaffolds (bash, file edits), and agent frameworks (Terminus 2). Safety evaluations claim Haiku 4.5 has lower rates of misaligned behavior than Sonnet 4.5 and Opus 4.1 and is released at AI Safety Level 2 (ASL‑2). The result: a model that prioritizes speed and cost‑efficiency while preserving deep reasoning and multi‑agent orchestration capabilities for developers building responsive AI systems.