Anthropic shipped Claude Sonnet 4.6 on February 17, 2026, and the headline number is striking: in side-by-side comparisons with Sonnet 4.5, developers preferred the new model's output in 70% of Claude Code tests.
What Changed in Sonnet 4.6
Multi-File Coherence
Maintains consistent intent across large refactors spanning multiple files and modules.
Bug Localisation
Improved reasoning about stack traces and runtime errors, with fewer false-positive diagnoses.
Test Generation
Higher quality unit and integration test output that actually catches real bugs.
Instruction Following
Substantially reduced rate of ignoring explicit constraints during code edits.
Job-Specific Enterprise Plugins
Alongside the model release, Anthropic launched job-specific plugins connecting Claude to enterprise tools:
Excel
Formula generation & data analysis
PowerPoint
Slide drafting & structure
Gmail
Email drafting & triage
Drive
Document search & summarisation
Claude can now complete multistep actions autonomously across these tools — read a spreadsheet, identify anomalies, draft a summary email, and schedule a follow-up.
The jump from Sonnet 4.5 to Sonnet 4.6 in code tasks is the clearest quality delta I've seen in a minor version bump. It's the difference between a junior who needs step-by-step instructions and one who reads the PR description and gets it right first time.
Our AI automation service helps clients design lightweight, high-ROI integrations with these tools.
Integrating Claude Into Your Workflow?
Our team has hands-on experience deploying Anthropic models in production CRM, analytics, and automation pipelines.
Talk to Our AI TeamTags:
AnthropicClaudeAI ModelsDeveloper ToolsLLM
