Claude Haiku 4.5 marks a turning point for Anthropic's budget-friendly model line, introducing capabilities that were previously exclusive to higher-tier models. This release brings extended thinking, computer use, and context awareness to the Haiku family for the first time, representing frontier-level functionality compressed into an efficient architecture. The model achieves 73.3% on SWE-bench Verified, a benchmark score that places it among the world's strongest coding models and demonstrates that the efficiency focus did not come at the cost of genuine capability. By bringing these advanced features to a lightweight model, Anthropic has made near-frontier intelligence practical for applications where running larger models would be prohibitive.
The rapid capability diffusion visible in Haiku 4.5 reflects a broader trend in Anthropic's recent model development. Following Opus 4.1 and Sonnet 4.5, Haiku 4.5 delivers performance roughly on par with Sonnet 4 while operating at a significantly lower price point, which the company describes as roughly one-third the cost. This positioning makes Haiku 4.5 particularly well-suited for multi-agent orchestration, scaled deployments, and workflows requiring high-volume inference across complex tasks. The combination of extended thinking with the model's efficiency creates opportunities for developers building automated pipelines that previously required more expensive infrastructure, opening near-frontier AI capabilities to a wider range of production use cases.
