Microsoft steers VS Code to Claude 4, signaling an Anthropic tilt

Microsoft added auto model selection to VS Code that, for paid Copilot users, defaults to Anthropic’s Claude Sonnet 4 over OpenAI’s GPT-5. Internal benchmarks and guidance back the choice, and Microsoft plans to use Anthropic in Microsoft 365 where it tests better. Meanwhile, the company is scaling its own model training efforts and still deepening ties with OpenAI.

Key Points

• VS Code gains an auto model selector; paid Copilot users will primarily use Anthropic’s Claude Sonnet 4 over OpenAI’s GPT-5.
• Microsoft’s internal benchmarks recommend Claude Sonnet 4 for Copilot, and that guidance persists after GPT-5’s release.
• Microsoft 365 Copilot is expected to be partly powered by Anthropic models where they outperform OpenAI (notably in Excel and PowerPoint).
• Microsoft is ramping up in-house model training and compute, with plans to scale beyond the 15,000-H100 cluster used for MAI-1-preview.
• Microsoft maintains its partnership with OpenAI, recently striking a new deal that could aid OpenAI’s IPO.

Sentiment

The discussion is largely sympathetic to the article's premise that Claude is strong for coding and Microsoft's move makes strategic sense. Many commenters validate the choice from personal experience. However, there is a meaningful counter-current arguing GPT-5 and Codex are catching up or have surpassed Claude, and a general sense that model leadership is fleeting. The overall tone is moderately agreeing with the article but with significant caveats about durability.

In Agreement

• Claude is genuinely superior for coding tasks — it produces practical, actionable code while ChatGPT tends to describe what it would do rather than doing it
• Microsoft is making a pragmatic choice based on technical merit, treating AI models as commodities and positioning as a model-agnostic tool-maker
• It is notable that Microsoft would choose on merit over existing business relationships, suggesting a significant quality gap between Claude and OpenAI for coding
• Competitive pressure from Cursor likely forced Microsoft to offer the best available model in VS Code to protect its developer tool market share
• Claude also has an edge in creativity and style emulation beyond just code generation

Opposed

• GPT-5 and the new Codex models have matched or surpassed Claude for coding, making this preference potentially outdated
• The 'auto' model selector is really a mechanism to push users toward paid tiers rather than a genuine endorsement of Claude
• Anthropic has concerning practices including aggressive rate limiting, credit expiration, and data usage policies that make it a questionable partner
• Which model is 'best' changes every few months, so any default choice is temporary and benchmarks are not particularly helpful
• Claude Sonnet 4 has known weaknesses in math and formal proofs compared to GPT-5 Thinking