Claude 3.5 Sonnet on GitHub Copilot

Source: Anthropic Date: 2024-10-29 URL: https://www.anthropic.com/news/github-copilot

Summary

Anthropic announced Claude 3.5 Sonnet rolling out on GitHub Copilot, available in Visual Studio Code and GitHub.com via public preview. Claims: “outperforms all publicly available models” on SWE-bench Verified; 93.7% on HumanEval. Delivered via Amazon Bedrock cross-region inference infrastructure.

Implications

• Claude Code / developer tool distribution thread. GitHub Copilot has 1.8M+ paid subscribers — making Claude 3.5 Sonnet available in Copilot is the largest single developer tool distribution event in Anthropic’s history to that point. Every GitHub Copilot user becomes a potential Claude model user.
• Amazon Bedrock as the delivery layer. Copilot running Claude through Bedrock means Microsoft (GitHub’s owner) is delivering Anthropic’s model via Amazon’s infrastructure — an unusual three-party arrangement that foreshadows the later Microsoft Azure/Foundry distribution relationship.
• SWE-bench “outperforms all publicly available models.” The strongest capability claim Anthropic had made to that point (October 2024). This is the benchmark that drove Cursor, Replit, and other coding tools to adopt Claude 3.5 Sonnet as their primary model.
• Timing. October 2024 — Claude 3.5 Sonnet was already four months old (launched June 2024). GitHub Copilot integration taking four months to ship suggests non-trivial integration complexity or negotiation time.
• Watch: actual developer preference data (do Copilot users switch to Claude 3.5 Sonnet vs. the default GPT model); whether Copilot integration expands to Claude 3.7/4.x models; how the Bedrock delivery arrangement evolves with Microsoft’s Azure investment in Anthropic.