Anthropic Releases Claude Sonnet 4.5 With Agent SDK

Anthropic has released Claude Sonnet 4.5, which it bills as its best coding model and strongest model for building complex agents. The model scored 77.2% on SWE-Bench Verified (up from Opus 4.1’s 74.5%) and leads the OSWorld real-world tasks benchmark at 61.4%, with Anthropic saying it outperforms several competitors across benchmarks. New product features include checkpoints in Claude Code, a context-editing feature and memory tool in the API, in-chat code execution and file creation, and a Claude Agent SDK that supports building agentic applications and coordinating sub-agents. Anthropic says Sonnet 4.5 can sustain focus on long-running multi-step tasks (30+ hours) and shows improved domain knowledge in areas like finance, law and medicine. The company reports improved safety and alignment metrics — reductions in sycophancy, deception, power-seeking and prompt-injection vulnerability — and is keeping API pricing unchanged at $3/$15 per million tokens. Sonnet 4.5 is available via the Claude API, Claude.ai and Claude Code, and Anthropic recommends upgrading to it for most uses.