Anthropic's Claude Sonnet 4.5 is Live in Cline

Claude Sonnet 4.5 by Anthropic is now available in Cline. With a 200k context window at the same $3/$15 pricing as Sonnet 4, this update represents a refinement in how AI models approach real software development.

But Sonnet 4.5 doesn't just work better than previous models - it works differently. Here's what that means for how you use Cline.

The model writes its own documentation

Without prompting, Sonnet 4.5 creates CHANGELOG.md, SUMMARY.md, and progress.txt files. It treats the filesystem as memory, writing architectural documentation as it works. This isn't just note-taking; it's how the model thinks.

The model is noticeably more terse than previous versions, jumping straight to implementation without play-by-play commentary. It delivers fact-based progress reports when needed, stays quiet when executing, and only speaks up when it genuinely needs clarification.

It works test-first

Sonnet 4.5 will create tests in structured formats (like tests.json) before touching implementation code. It sets up init.sh scripts to manage development environments, maintains test suites, and builds the scaffolding it needs to work efficiently across sessions.

This test-first approach leads to more reliable long-running tasks. The model can work on a complex feature for an entire context window, properly save its state, then pick up in a fresh session exactly where it left off. It reads its own progress notes, checks git logs, runs test suites, and continues implementation without missing a beat.

Why this pairs with Auto Compact

Sonnet 4.5's documentation skill is the same skill it uses when Cline's Auto Compact asks it to summarize conversations. When Auto Compact compresses conversation history to continue past context limits, the model excels at creating concise, accurate summaries.

The result is remarkably smooth continuity across context windows. The model's terseness reduces context pollution, while its habit of maintaining external state files provides an additional layer of memory beyond Cline's compression. These capabilities align naturally with how Cline handles long-running tasks.

Getting the Most from Sonnet 4.5

Be explicit about intent. Say "implement these changes" rather than "can you suggest changes." The model precisely follows instructions - if you ask for suggestions, that's what you'll get.

For complex tasks, let the model set up comprehensive tests first. It excels at creating test harnesses that guide implementation and ensure correctness throughout the work.

Trust its ability to maintain state. You don't need to remind it of context or re-explain requirements when continuing work across sessions. The model's state tracking is remarkably robust.

The model is available now in Cline as claude-sonnet-4.5. 200k context window, same pricing as Sonnet 4 at $3 per million input tokens and $15 per million output tokens.

Ready to experience a model that focuses on shipping code? Update Cline and select Sonnet 4.5 today. Share your experiences with complex, multi-session projects on Reddit or Discord.