Microsoft's MAI Code 1.1 Flash Crushed by DeepSeek on Price and Performance

The Decoder · rss · 2026-08-12

Microsoft has released MAI Code 1.1 Flash, a new code model designed for GitHub Copilot. The company claims it is 25% more token-efficient than its predecessor at a quarter of the cost.

However, benchmarks reveal that it is crushed on both performance and price by the cheaper DeepSeek V4 Flash. The article notes this fits a broader pattern: Microsoft publicly champions open AI but bakes worse proprietary models into its apps to protect profit margins.

Original post →

More from Models

Models channel →