Claude Sonnet 4.5 coding model improves agentic capabilities

Anthropic has released Claude Sonnet 4.5, an update of the company’s coding model that features extended autonomous operation, greater context awareness, and enhanced tool usage. Claude Sonnet 4.5 shows significant improvements on real-world computer tasks, agentic coding, agentic terminal coding, and agentic tools use, as well as reasoning and math, according to Anthropic.

Claude Sonnet 4.5 can be accessed by developers using claude-sonnet-4-5 via the Claude API. Pricing is the same as Claude Sonnet 4 at $3 per million tokens input and $15 per million tokens output.

Introduced September 29, Cloud Sonnet 4.5 offers what the company describes as major advances for agents. The coding model can work independently for hours while maintaining clarity and focus on incremental progress, Anthropic said. Fact-based progress updates are provided about what has been accomplished. With increased context awareness, Claude Sonnet now tracks token usage throughout conversations, receiving updates after each tool call. This helps prevent premature task abandonment while enabling more-effective execution on long-running tasks, according to Anthropic.

Also with Claude Sonnet 4.5, the model more effectively uses parallel tool calls, firing off multiple speculative searches simultaneously during research and reading several files at once to more quickly build context. Advanced context management in Claude Sonnet 4.5 maintains state tracking in external files, preserving goal-orientation across sessions. Combined with more-effective context window usage and new context management API features, the model optimally handles information across extended sessions to maintain coherence over time, the company said.

A new memory tool, now in beta, enables Claude Sonnet to store and retrieve information outside the context window. Also, Claude Sonnet 4.5 introduces a model_context_window_exceeded stop reason that explicitly indicates when generation stopped due to hitting the context window limit, rather than the requested max_tokens limit. This makes it easier to handle context window limits in application logic, Anthropic said. Claude Sonnet 4.5 also includes a bug fix to preserve intentional formatting in tool call string parameters.

Separately, Anthropic announced improvements in Claude Code including checkpoints, which save the user’s progress and allow them to roll back to a previous state.