Anthropic has unveiled Claude Sonnet 4.5, which. It is said to be the strongest coding model in the world. The model brings improvements in reasoning, mathematics, and computer use, making it suited for building complex AI agents.
The company has also introduced some updates across its ecosystem. For example, Claude code now includes checkpoints, a refresh of the terminal interface, and a native VS Code extension. Additionally, the ClaudeAPI receives new context, editing, and memory tools, enabling longer and more complex agent runs.
For developers, Anthropic is releasing the Claude agent and SDK, the same infrastructure that powers Claude code. This will allow teams to build their own AI agents using the same foundation.
Claude Sonnet 4.5 has also made big improvements on benchmarks. On OSWorld, which tests real-world computer tasks, it scored 61.4%, up from 42.2% with Sonnet 4. The model also performs better in domain-specific reasoning, including finance, law, medicine, and STEM.
According to the company, it has improved the models, alignment, and safety, which reduces risks such as prompt injection attacks and unwanted behaviors.
The model is available today via the Claude API at the same pricing as Sonnet 4 ($3 or $15 per million tokens). Users of Claude apps and tools will get the upgrade automatically.
Claude Sonnet 4.5 is also now available on Geekflare Connect, a platform that lets users connect multiple AI API keys in one place. Geekflare Connect already supports models from OpenAI, Google, Grok, and previous Claude versions, now extending support to the latest Sonnet 4.5 as well.
