Why Microsoft Is Dropping Claude Code Despite Its Superior Performance

Microsoft announced that, by June 30, the Experiences + Devices team will discontinue most internal Claude Code licenses and require all engineers to adopt the GitHub Copilot CLI.

Benchmarking on SWE‑bench shows Claude Code achieving an 80.8% score, 8.3 points higher than the GPT‑4o‑based Copilot (72.5%). For tasks involving changes to more than five files, Claude Code’s success rate is 23% higher than Copilot’s, and on cross‑file refactoring/debugging scenarios Claude Code completes 89% of tasks versus Copilot’s 60%.

Since its internal release in December, Claude Code rapidly attracted thousands of engineers, diverting core users from Copilot CLI and becoming the preferred tool for many Microsoft developers.

The move occurs amid a broader Microsoft‑Anthropic partnership: Microsoft has invested heavily in Anthropic, integrated Claude models into Azure Foundry and Microsoft 365 Copilot, and even considered acquiring the AI‑coding startup Cursor before abandoning the plan for antitrust reasons.

According to an internal memo from Rajesh Jha, the decision serves two purposes: cutting the sizable subscription cost for thousands of Claude Code licenses at fiscal‑year‑end, and consolidating the internal tooling ecosystem to boost Copilot adoption and close the product loop.

Technical differences reinforce the strategic shift: Claude Code offers a million‑token context window and can handle roughly 3,000 files per session, whereas Copilot CLI is limited to a 128 KB token context, making large‑scale migrations challenging.

Engineers now face a six‑week migration window, during which they must adapt to a less mature Copilot CLI, effectively experiencing a forced downgrade in tooling capability.

While Microsoft will continue to use Anthropic models in Azure Foundry and Microsoft 365, the company’s stance remains clear: external tools may be valuable, but internal tooling priority prevails.