Editorial illustration for Microsoft's New Coding AI Falls Short Against DeepSeek on Price and Performance
Microsoft's MAI Code 1.1 Undercuts DeepSeek Price
Microsoft's New Coding AI Falls Short Against DeepSeek on Price and Performance
Microsoft rolled out MAI Code 1.1 Flash this week, the latest coding model feeding into GitHub Copilot, and the pitch sounds solid enough on its own terms. The company says it writes better code than its June predecessor, runs 25 percent more efficiently on tokens, and costs a quarter as much to use. Developers accepted 4 percent more of its suggested code, and Microsoft says the model trained on hundreds of thousands of reinforcement-learning environments built inside Copilot itself.
Stack it against its own family tree and MAI Code 1.1 Flash looks like real progress. Stack it against Deepseek-V4-Flash-0731, and the story changes. Microsoft's benchmarks show the new model edging out mini-models from Anthropic and OpenAI, but Deepseek's flash variant beats it on both price and raw performance, the two numbers that matter most to anyone actually choosing a model to build with.
That mismatch raises an obvious question: why release a coding model that a free alternative already beats on the metrics that count, especially from a company that has spent months talking up its enthusiasm for open, freely available AI.
MAI-Code-1.1-Flash looks like a budget model on paper, but it trails the more capable Deepseek on both price and performance. Cost per token doesn't tell the whole story without factoring in usage efficiency, but the gap in Deepseek's favor is likely significant either way.
Why this matters
For developers picking a coding model, the math still doesn't favor Microsoft. MAI-Code-1.1-Flash is cheaper than its June predecessor, but "cheaper than before" isn't the same as "cheapest option," and DeepSeek keeps winning on both price and output quality. That's the gap that matters to anyone budgeting API calls at scale, not the quarter-off discount Microsoft is touting in its own announcement.
We'd also flag the awkward positioning: Microsoft has spent months praising open-weight models like DeepSeek's, then turns around and ships a proprietary competitor that loses to them on the metrics developers actually check first. If you're building on GitHub Copilot, this is worth watching less for what MAI-Code-1.1-Flash does well and more for what it signals about Microsoft's roadmap, are they iterating toward parity, or is this a stopgap while they figure out whether in-house models are worth the investment at all. Until token efficiency claims translate into real cost advantages over open alternatives, treat Microsoft's benchmarks as a starting point for your own testing, not a reason to switch.
Related Reading
Browse all Microsoft news →
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
Related Topic Claude gains shared context in Excel, PowerPoint; Microsoft adds Copilot Cowork
AI Tools & Apps
Related Topic Microsoft adds ’vibe working’ to Word and Excel; Copilot Agent Mode now default
Business & Startups
From Archives Mistral's EU Data and Priority Access Have Stateful Feature Limits
From Archives Google’s AMIE AI conducts first real-time video medical consultations
From Archives Microsoft's Open Source Test Agent Hits 92.1% Completion, Outperforming Copilot
Open Source
From Archives Microsoft's USD 37B AI Business Relies 70% on OpenAI, Nadella Says
Open Source
Common Questions Answered
How does MAI Code 1.1 Flash's performance compare to Microsoft's June predecessor?
MAI Code 1.1 Flash writes better code than its June predecessor, runs 25 percent more efficiently on tokens, and costs a quarter as much to use. Developers also accepted 4 percent more of its suggested code compared to the earlier version.
Why does DeepSeek outperform Microsoft's MAI Code 1.1 Flash despite lower costs?
DeepSeek trails Microsoft on both price and performance metrics, meaning it offers superior capabilities at a lower cost per token. The gap in DeepSeek's favor is likely significant when factoring in usage efficiency, making it the more economical choice for developers budgeting API calls at scale.
How was MAI Code 1.1 Flash trained to improve code suggestions?
Microsoft trained the model on hundreds of thousands of reinforcement-learning environments built inside Copilot itself. This training approach contributed to the improved code quality and higher developer acceptance rates compared to its predecessor.
What is the key limitation of comparing MAI Code 1.1 Flash to competitors based on cost per token alone?
Cost per token doesn't tell the whole story without factoring in usage efficiency and overall performance quality. While MAI Code 1.1 Flash is cheaper than its June predecessor, being cheaper than before doesn't make it the cheapest option available, and developers must consider both price and output quality when selecting a coding model.
Further Reading
- MAI-Code-1.1-Flash: Better, faster, at a quarter of the cost - Microsoft AI
- Introducing MAI-Code-1-Flash - Microsoft AI
- MAI-Code-1-Flash model card - Microsoft AI
- MAI-Code-1-Flash: early results from real developer workflows - Visual Studio Code Blog
- MAI-Code-1-Flash for Copilot Business and Copilot Enterprise - GitHub Blog
