Introducing Claude Sonnet 4.6

Claude Sonnet 4.6 is the most powerful model in the Sonnet line, taking performance to the next level across coding, computer use, long-context reasoning, agent planning, knowledge work, and design. A 1M token context window is available in beta.

On Free and Pro plans, the default model for Claude.ai and Claude Cowork has switched to Sonnet 4.6, and API pricing starts at the same $3/$15 per million tokens as Sonnet 4.5.

Coding performance in particular has improved significantly. Early access users preferred Sonnet 4.6 far more than the previous model, with some rating it even better than Claude Opus 4.5. When editing code, it reads context better and shows a stronger tendency to consolidate shared logic rather than duplicate implementations, making long sessions feel less frustrating.

Computer use capability has also noticeably improved. On real-world software-based benchmarks like OSWorld, the Sonnet line has steadily improved over the past 16 months, getting closer to tasks that used to require a human, such as navigating complex spreadsheets or filling out multi-step web forms. However, it still lags behind a skilled human, and the risk of prompt injection—where hidden instructions on a website trick the model—remains.

In safety evaluations, Sonnet 4.6 showed major improvement over the previous version, and demonstrated a similar level of defense as Opus 4.6. Anthropic described it as a model with a "broadly warm, honest, prosocial" character and strong safety behavior.

For practical use, these are the key changes:

• Claude Platform: supports adaptive thinking, extended thinking, and context compaction (beta)
• API: web search and fetch now include built-in code execution that automatically filters and processes search results
• General availability (GA): code execution, memory, programmatic tool calling, tool search, tool use examples
• Claude in Excel: supports MCP connectors, integrating with external tools such as S&P Global, LSEG, Daloopa, PitchBook, Moody's, and FactSet

Anthropic stated that Opus 4.6 remains best suited for tasks requiring the deepest reasoning, such as codebase refactoring or coordinating multiple agents. Sonnet 4.6, on the other hand, has established itself as an option that can handle a broader range of tasks at a more realistic cost.

The free tier has also been upgraded with Sonnet 4.6 as the default, including file creation, connectors, skills, and compaction. Developers can get started right away with the claude-sonnet-4-6 model.