Claude Haiku 4.5, Anthropic's newest small model, has been made available to all users starting today.
Performance that previously existed at the cutting edge is now available with improved speed and lower costs. Just five months ago, Claude Sonnet 4 represented state-of-the-art capabilities. Now, Claude Haiku 4.5 delivers comparable coding performance at just one-third the price and over double the speed.
Claude Haiku 4.5 actually exceeds Claude Sonnet 4's performance in specific areas, including computer use capabilities. These improvements enhance applications such as Claude for Chrome, making them more responsive and practical than previously possible.
Applications requiring real-time, low-latency responses-including chat assistants, customer service systems, and pair programming tools-will benefit from Haiku 4.5's balance of advanced intelligence and exceptional speed. Developers using Claude Code will experience significant improvements in responsiveness for multi-agent projects and rapid prototyping workflows.
While Claude Sonnet 4.5 (introduced two weeks prior) continues as Anthropic's frontier model and leading coding solution worldwide, Claude Haiku 4.5 provides an alternative for scenarios demanding near-frontier capabilities with superior cost efficiency. This creates opportunities for combined model usage-Sonnet 4.5 can decompose complex challenges into multi-step strategies, then coordinate teams of Haiku 4.5 instances executing subtasks simultaneously.
Claude Haiku 4.5 can be accessed immediately. Developers should specify claude-haiku-4-5 through the Claude API. The pricing structure is $1 per million input tokens and $5 per million output tokens.
Performance Metrics
Claude Haiku 4.5 ranks among Anthropic's highest-performing models. On SWE-bench Verified (measuring real-world coding performance), it achieves results approaching frontier models while maintaining speed and cost advantages. Multiple industry partners report that Haiku 4.5 delivers 90% of Sonnet 4.5's performance in coding tasks while operating at 4-5 times faster speeds.
Customer testimonials highlight the model's ability to deliver what was considered state-of-the-art performance six months ago, now at dramatically reduced costs and latencies. Organizations report improvements in instruction-following accuracy, with one partner seeing 65% accuracy compared to 44% from their previous premium model.
Safety Assessments
Comprehensive safety and alignment testing was conducted on Claude Haiku 4.5. The model demonstrated minimal concerning behaviors and showed substantial alignment improvements over its predecessor, Claude Haiku 3.5. Automated alignment assessments revealed statistically significant reductions in misaligned behaviors compared to both Claude Sonnet 4.5 and Claude Opus 4.1-establishing Claude Haiku 4.5 as Anthropic's safest model according to these metrics.
Safety evaluations indicated limited risks regarding chemical, biological, radiological, and nuclear (CBRN) weapons production. Consequently, the model has been classified under AI Safety Level 2 (ASL-2) standards-less restrictive than the ASL-3 designation for Sonnet 4.5 and Opus 4.1. Complete safety test results and ASL-2 classification reasoning are documented in the Claude Haiku 4.5 system card.
Availability and Implementation
Claude Haiku 4.5 is currently accessible through Claude Code and Anthropic's applications. Its efficiency allows users to achieve more within usage limits while maintaining high-performance standards.
Developers can implement Claude Haiku 4.5 through Anthropic's API, Amazon Bedrock, and Google Cloud's Vertex AI. The model serves as a direct replacement for both Haiku 3.5 and Sonnet 4, offered at Anthropic's most economical pricing tier.
Technical specifications, evaluation results, and comprehensive documentation are available through the system card, model page, and technical documentation.