Now loading...
Anthropic has announced the release of Claude Haiku 4.5, a new model designed to deliver high-quality performance to users, available now for all. In a striking shift from the past, the latest offering boasts similar capabilities to the previously elite Claude Sonnet 4 but at a significantly reduced cost—one-third of the price—and with over double the speed.
Beyond matching the performance of its predecessor, Claude Haiku 4.5 surpasses Sonnet 4 in specific computing tasks, thereby enhancing the efficiency of applications such as Claude for Chrome, making them faster and more effective.
This new model is particularly beneficial for tasks requiring real-time responses, such as operating chat assistants or customer service bots, as well as collaborative coding projects. Users of Claude Code are expected to experience a more agile coding journey, from managing joint projects to rapid testing and prototyping.
Earlier this month, Anthropic also unveiled Claude Sonnet 4.5, which remains the company’s top-tier model for coding tasks. However, Claude Haiku 4.5 presents a budget-friendly alternative without sacrificing performance. The model also facilitates innovative collaborative strategies, allowing Claude Sonnet 4.5 to decompose complex issues into step-by-step plans while coordinating multiple Claude Haiku 4.5 systems to handle subtasks concurrently.
Moreover, developers can access Claude Haiku 4.5 seamlessly through the Claude API, with pricing set at an economical $1 for every million input and output tokens.
The performance benchmarks of Claude Haiku 4.5 have indicated a significant advancement; it operates at speeds that are four to five times faster than Sonnet 4.5 while maintaining competitive output quality. This model not only achieves near-frontier coding success but also does so efficiently, as highlighted by users praising its exceptional balance of speed and responsiveness.
Extensive safety evaluations conducted on Claude Haiku 4.5 revealed that the model exhibits significantly fewer concerning behaviors compared to its predecessor, Claude Haiku 3.5. It also maintained a better alignment rate than both Claude Sonnet 4.5 and Claude Opus 4.1, which classifies it as the safest AI model produced to date. The model carries an AI Safety Level 2 (ASL-2) designation due to its limited risks associated with the production of hazardous materials, compared to the stricter ASL-3 for Sonnet 4.5 and Opus 4.1, detailed in the Claude Haiku 4.5 system card.
Available now on Claude Code and various applications, Claude Haiku 4.5 allows users to maximize their capabilities while adhering to usage restrictions, all while maintaining the standard of a premium model. The model can be accessed on platforms like Amazon Bedrock and Google Cloud’s Vertex AI, where it can effectively replace both Haiku 3.5 and Sonnet 4 at a more advantageous pricing structure.
For more technical specifics and to explore evaluation results, additional information can be found on the system card, the model page, and the documentation.
