Anthropic launches Claude Sonnet 4.5 to boost AI reasoning, coding and long-task handling

AI giant Anthropic launched Claude Sonnet 4.5, its latest large language model, along with a series of updates across its products on Monday. The company says the new model performs better in reasoning, coding, and computer-based tasks than earlier versions.

Anthropic unveils Claude Sonnet 4.5 and updates to Claude Code

One of the key additions is a set of updates to Claude Code. These include checkpoints that let users save progress and return to earlier states, a refreshed terminal interface, and a new extension for Visual Studio Code. The Claude API now has context editing and a memory tool to help agents manage longer and more complex tasks. Within the Claude apps, code execution and file creation features have been added, covering spreadsheets, presentations, and documents.

Claude Agent SDK released

Anthropic has also introduced the Claude Agent SDK, which provides developers with the same infrastructure used in its in-house tools. This framework is designed for building AI agents and includes systems for memory management, permissions, and coordination across sub-agents.

Benchmark performance

The company reports that Claude Sonnet 4.5 has improved results on benchmarks measuring real-world coding and computer use. For example, it leads the OSWorld test at 61.4 per cent, an increase from 42.2 per cent with the previous Sonnet 4 model.

According to Anthropic, experts in sectors such as finance, law, medicine, and science have found the model more accurate in domain-specific reasoning. The release is also described as the firm’s most aligned model to date, with additional safeguards against issues such as prompt injection attacks.

Availability and safeguards

Claude Sonnet 4.5 is available through the Claude API at the same price as the earlier Sonnet 4. The model is being released under Anthropic’s AI Safety Level 3 framework, which includes filters to limit harmful or high-risk content.