Anthropic introduced a new compact AI model on Wednesday—Claude Haiku 4.5—which had been rumored in recent leaks. According to the company, Haiku 4.5 delivers the same coding performance as Claude Sonnet 4, Anthropic’s state-of-the-art model launched in May, while being significantly faster and more affordable. The new model will roll out to Free plan subscribers starting this week.
Performance and Efficiency
In its announcement, Anthropic stated that Claude Haiku 4.5 is over twice as fast as Sonnet 4 while costing just one-third as much to operate. Internal benchmarks show that Haiku 4.5 not only exceeds Sonnet 4’s performance but also outperforms ChatGPT (GPT‑5) and Gemini 2.5 Pro in software engineering evaluations.
The model also scored highly in additional benchmark categories, including agentic tool use, computer interaction, and visual reasoning, reinforcing its versatility beyond pure coding tasks.
Safety Enhancements
Anthropic emphasized Haiku 4.5’s safety improvements, claiming it achieved a “statistically significantly lower overall rate of misaligned behaviors” than both Sonnet 4.5 and Opus 4.1. The company described it as its safest model to date.
Claude Haiku 4.5 will be globally available starting Wednesday, with usage priced at $1 per million input tokens and $5 per million output tokens.
Real‑World Applications
Anthropic positions Haiku 4.5 as an ideal AI for users who need real‑time, low‑latency interactions. According to the company, the model’s blend of speed and intelligence makes it well‑suited for chat assistants, customer service bots, and pair programming tools. Free‑tier users will likely notice faster responses and smoother performance immediately.
Collaborative Model Capabilities
Developers can now pair Haiku 4.5 with larger Claude models to streamline complex work. For example, Sonnet 4.5 might decompose a complex task into a multi‑step plan, while multiple Haiku 4.5 instances execute those subtasks in parallel—a powerful configuration for scaling workflows or automating intricate processes.
Early Feedback from Enterprise Users
Some enterprise testers have already praised Haiku 4.5’s responsiveness and stability. Brad Axen, Head of AI at Block Inc., described speed as “the new frontier for AI agents operating in feedback loops,” adding that “Haiku 4.5 proves you can have both intelligence and rapid output. It handles complex workflows reliably, self‑corrects in real time, and maintains momentum without latency overhead.”
Zach Lloyd, CEO of Warp AI, highlighted its impact on agentic coding: “Haiku 4.5 is a leap forward for sub‑agent orchestration and computer use tasks. The responsiveness makes AI‑assisted development in Warp feel instantaneous.”
