Claude Sonnet 4.5: Anthropic's best model for coding and long-running agents
In one sentence Anthropic releases Claude Sonnet 4.5: SOTA on SWE-bench Verified (77.2%), capable of 30+ hour agentic tasks. New Claude Agent SDK released alongside.
Anthropic updates Sonnet (the "middle" Claude 4 family model) with version 4.5. Sonnet 4.5 becomes the world's best coding model on serious benchmarks (SWE-bench Verified: 77.2%, above GPT-5 at 75%).
Most impressive capability: it can work autonomously on a task for 30+ hours straight without losing focus. For perspective: Sonnet 4 in May held up for 5-7 hours. Sonnet 4.5 enters "agent working a full workday" territory.
Alongside the model, Anthropic ships the Claude Agent SDK: an official toolkit for building agents, with primitives like memory, sandbox, sub-agents, tool use. It's the technical foundation for anyone building production-grade Claude agents.
Sonnet 4.5 keeps Sonnet 4's pricing ($3/$15 per million tokens), making it economically attractive for heavy use.
Companies
Anthropic
Tools
Claude Sonnet 4.5
Tags
Sources