Get ready for a game-changing AI experience! We're thrilled to introduce Claude Haiku 4.5, a small model that packs a powerful punch. This innovative release is a testament to the rapid advancements in AI technology, offering users a unique blend of intelligence, speed, and cost-efficiency.
Just five months ago, Claude Sonnet 4 was the state-of-the-art model, but now, Haiku 4.5 has taken the lead. It delivers similar coding performance at a fraction of the cost and with incredible speed. In fact, it surpasses its predecessor in certain tasks, especially when it comes to computer-related functions. This breakthrough has revolutionized applications like Claude for Chrome, making them faster and more efficient than ever.
Haiku 4.5 is a game-changer for real-time, low-latency tasks. Whether you're building chat assistants, customer service agents, or engaging in pair programming, Haiku 4.5's combination of high intelligence and remarkable speed will enhance your productivity. And for coding enthusiasts, Claude Code will feel more responsive and intuitive with Haiku 4.5.
While Claude Sonnet 4.5 remains our frontier model, Haiku 4.5 offers a new, cost-efficient option for near-frontier performance. It opens up exciting possibilities for collaborative model use. For instance, Sonnet 4.5 can break down complex tasks into multi-step plans, and then Haiku 4.5s can work together in parallel to complete these subtasks efficiently.
Haiku 4.5 is now available to everyone. Developers can access it via the Claude API, and the pricing is incredibly affordable at just $1/$5 per million input and output tokens.
Benchmarks reveal that Haiku 4.5 has achieved a sweet spot that was previously unimaginable. It offers near-frontier coding quality with lightning-fast speed and cost efficiency. In Augment's agentic coding evaluation, Haiku 4.5 achieved an impressive 90% of Sonnet 4.5's performance, matching larger models. This model is a leap forward for agentic coding, especially in sub-agent orchestration and computer use tasks. The responsiveness is so seamless that AI-assisted development feels instantaneous.
Historically, models have had to sacrifice speed and cost to achieve quality. But Haiku 4.5 challenges this trade-off. It's a fast frontier model that keeps costs efficient, and it hints at the future direction of this class of models. With Haiku 4.5, you get intelligence without compromising speed. It enables the development of AI applications that combine deep reasoning with real-time responsiveness.
The capabilities of Haiku 4.5 are truly remarkable. Just six months ago, this level of performance would have been considered state-of-the-art. Now, it runs up to 4-5 times faster than Sonnet 4.5 at a fraction of the cost, opening up a whole new world of use cases.
Speed is the new frontier for AI agents in feedback loops, and Haiku 4.5 proves that you can have both intelligence and rapid output. It handles complex workflows with ease, self-corrects in real-time, and maintains momentum without latency issues. For most development tasks, Haiku 4.5 strikes the perfect balance.
In terms of safety, Claude Haiku 4.5 has undergone a series of rigorous evaluations. It exhibits low rates of concerning behaviors and is significantly more aligned than Claude Haiku 3.5. In our automated alignment assessment, Haiku 4.5 showed a statistically significant reduction in misaligned behaviors compared to Sonnet 4.5 and Opus 4.1, making it our safest model to date.
Our safety testing also revealed that Haiku 4.5 poses limited risks in terms of CBRN (Chemical, Biological, Radiological, and Nuclear) weapons production. As a result, we've released it under the AI Safety Level 2 (ASL-2) standard, which is less restrictive than the ASL-3 applied to Sonnet 4.5 and Opus 4.1. You can find the full reasoning behind the ASL-2 classification and details of our safety tests in the Claude Haiku 4.5 system card.
Haiku 4.5 is now available on Claude Code and our apps, and its efficiency means you can do more while maintaining premium model performance. Developers can use it on our API, Amazon Bedrock, and Google Cloud's Vertex AI, where it serves as a cost-effective replacement for Haiku 3.5 and Sonnet 4.
For more technical details and evaluation results, check out our system card, model page, and documentation. We've also included the methodology used for benchmarking, so you can understand the process behind these impressive results.
Get your hands on Claude Haiku 4.5 today and experience the future of AI! And remember, with great power comes great responsibility. We encourage you to use this model ethically and responsibly. Join the conversation in the comments and share your thoughts on this exciting development!
