Anthropic Launches Claude Sonnet 4.5, Elevating Coding and Agent Capabilities

San Francisco / September 29, 2025 — Anthropic has officially launched Claude Sonnet 4.5, its most powerful mid-tier model to date, introducing major upgrades across coding, tool use, and long-context memory – all without increasing pricing.

The release adds Deep Memory tools, version checkpoints, and context editing inside Claude Code, allowing users to iterate on complex tasks across multiple sessions. A native VS Code extension is now available for developers, enabling direct in-editor collaboration with Claude during builds, patches, and refactors.

Inside the Claude web and mobile apps, users can now generate and edit spreadsheets, slides, and documents, as well as execute code directly in conversation. The rollout also includes the launch of the Claude Agent SDK, giving external developers access to the same orchestration system powering Claude’s own agent workflows.

According to Anthropic’s benchmarks, Sonnet 4.5 now leads the industry on SWE-bench Verified (software reasoning) and OSWorld (real-world computer interactions), scoring 61.4% versus 42.2% from Sonnet 4. Anthropic says the model can stay focused for 30+ hours across multi-step tasks, signaling a clear move toward persistent AI workers rather than single-shot chatbots.

Pricing remains unchanged from Sonnet 4 – $3 input / $15 output per million tokens, making it one of the highest-performing models at its tier. Developers can access the model immediately via API using claude-sonnet-4-5.