Anthropic releases Claude Haiku 4.5

Background

Claude is Anthropic's family of AI models. Anthropic offers them to people through the Claude apps and to developers through an API (application programming interface), a standard way for one program to call another. Developers are charged per million input and output tokens, and a token is the unit of text a model reads and writes.^[1]

What happened

On 15 October 2025 Anthropic released Claude Haiku 4.5, and the model was available to all users.^[1] It cost $1 per million input tokens and $5 per million output tokens.^[1]

Key details

Anthropic said the model matched Claude Sonnet 4's coding performance at one-third of the cost and more than twice the speed.^[1]

It scored 73.3% on SWE-bench Verified, a coding benchmark, and surpassed Sonnet 4 at computer use, the ability of a model to operate a computer interface.^[1]

The model was released under AI Safety Level 2 (ASL-2) protections. In Anthropic's automated assessment, it showed lower rates of misaligned behaviour than Sonnet 4.5 and Opus 4.1.^[1]

It was available in the Claude apps, Claude Code, the Claude API, Amazon Bedrock and Google Cloud Vertex AI.^[1]

Anthropic's AI Safety Level (ASL) framework groups models by the protections they receive, with higher levels applied to models that carry greater potential risk. In Anthropic's automated assessment, the company runs tests that check how often a model behaves in ways its developers did not intend, and the misaligned-behaviour rate is the share of such behaviour found in those tests.

Impact

Because pricing is set per token, the cost of a task depends on how much text the model reads and writes, so a lower per-token price matters most for applications that process large volumes. Readers can weigh that price against the performance Anthropic reported for the model.^[1]

For users choosing between models, the launch offered a lower-cost option whose coding performance Anthropic said matched Claude Sonnet 4, at one-third of the cost.^[1]