Anthropic's Claude Haiku 4.5 redefines AI economics

Anthropic released Claude Haiku 4.5 on 15 October 2025, significantly altering the cost and speed equation for large language model (LLM) deployment in the technology sector. The new small model achieves performance levels similar to the former state-of-the-art Claude Sonnet 4, but operates at one-third the cost and more than twice the speed.

This move positions Haiku 4.5 as a direct challenger for high-volume, low-latency enterprise applications, where speed and operational expenses matter most.

Performance and Unit Economics

Only five months ago, Claude Sonnet 4 defined the benchmark for coding tasks. Today, Haiku 4.5 effectively matches that level of capability. The model even surpasses Sonnet 4 on specific metrics, particularly tasks involving computer use. This capability push makes the model immediately useful for tools like browser extensions and rapid developer environments, such as Claude Code and GitHub Copilot.

For developers, the economic proposition is clear. Haiku 4.5 pricing starts at USD 1 per million input tokens and USD 5 per million output tokens. This cost structure creates the most economical option available for Anthropic's current generation of models, acting as a direct, economical replacement for the older Haiku 3.5 and Sonnet 4 models across major cloud services. The ability to access near-frontier performance at a lower price point changes unit economics for any organisation running AI at scale.

Orchestration and Agentic Workflows

The model’s core strength lies in its speed, which makes it suitable for real-time applications. Businesses running chat assistants, customer service agents, or pair programming tools benefit from Haiku 4.5's responsiveness.

Furthermore, Haiku 4.5 introduces extended thinking and expanded computer-use tools to the Haiku model family for the first time. This feature parity with larger models enables a powerful multi-model architecture. For example, the current frontier model, Claude Sonnet 4.5 (released two weeks prior), can map out a complex multi-step plan. Then, the system can use multiple, faster Haiku 4.5 models to execute the subtasks in parallel, maximising throughput while managing cost.

This architecture supports demanding use cases, including rapid prototyping, multi-agent coding projects, and large-scale financial analysis involving thousands of simultaneous data streams.

Safety and Enterprise Availability

Anthropic places Haiku 4.5 under the AI Safety Level 2 (ASL-2) standard, a less restrictive classification than the ASL-3 assigned to Sonnet 4.5 and Opus 4.1. This rating reflects the company’s internal safety testing, which found the model posed only limited risks related to chemical, biological, radiological, and nuclear (CBRN) weapon production.

In automated alignment testing, Haiku 4.5 recorded a statistically lower overall rate of misaligned behaviours than both Sonnet 4.5 and Opus 4.1. By this metric, the small model stands as Anthropic's safest model released to date.

For enterprise adoption, Haiku 4.5 is available immediately via the Claude API. It is also available to developers on cloud platforms, serving as a drop-in model replacement on both Amazon Bedrock and Google Cloud’s Vertex AI. This broad accessibility supports enterprises looking to deploy high-performance, cost-sensitive AI services globally.