Microsoft released MAI Code 1.1 Flash on Aug 12, 2026, a proprietary code model for GitHub Copilot that the company says writes better code than its predecessor. Independent benchmarks tell a different story. The new model underperforms DeepSeek's open-weight alternative on key tests while costing more per token, a gap that raises fresh questions about Microsoft's public embrace of open AI.
What Microsoft Claims
Microsoft says MAI Code 1.1 Flash is 25% more token-efficient than its June predecessor and costs a quarter as much. The company also reports that developers accepted 4% more of its output compared with the earlier version, and that "code survival rose 4% and return visits increased 9%." Training involved "hundreds of thousands of reinforcement-learning environments in GitHub Copilot."
The official announcement touts these vague improvement metrics but skips direct comparisons with rivals. Microsoft's model card buries the benchmark results, according to The Decoder's analysis, leaving the headline numbers to carry the message.
The Benchmark Gap
On SWE-bench Verified, MAI Code 1.1 Flash scores 72.6%, edging past its predecessor's 71.6% and beating Anthropic's Claude Haiku 4.5 at 69.8% and OpenAI's GPT-5.4 mini at 69.2%. DeepSeek-V4-Flash-0731's SWE-bench Verified score is not published.
The Terminal Bench 2.1 results are more revealing. MAI Code 1.1 Flash scores 62.9%, up from 51.7% for MAI Code 1 Flash. Claude Haiku 4.5 trails at 49.4%, and GPT-5.4 mini reaches 60.7%. DeepSeek-V4-Flash-0731 crushes them all with an 82.7% score.
The Price Problem
Pricing widens the gap. DeepSeek-V4-Flash charges $0.14 per input token, $0.0028 per input token with cache, and $0.28 per output token. MAI Code 1.1 Flash costs $0.20 per input token, $0.02 per input token with cache, and $1.20 per output token. Claude Haiku 4.5 is pricier still at $1.00 per input token, $0.10 per input token with cache, and $5.00 per output token.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
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. MAI Code 1.1 Flash looks like a budget model on paper, yet it trails the more capable DeepSeek on both price and performance.
Open Talk, Closed Models
Microsoft has been positioning itself as an open AI champion. Its model strategy tells a different story. MAI Code 1.1 Flash is proprietary and likely won't get an open-weights release. The company recently shook up Copilot, swapping out OpenAI and Anthropic models for its own cheaper MAI alternatives to cut costs.
The trade-off was worse performance for better margins. Matthias Bastian, writing for The Decoder, argues that Microsoft is sinking resources into a weaker, pricier in-house model instead of tapping more capable, freely available alternatives like DeepSeek-V4-Flash. Microsoft's open AI talk doesn't match its model strategy.
What Comes Next
Customers in the Microsoft ecosystem can still pick from different models depending on the app and use case. That flexibility may not last. Microsoft will almost certainly make its own models the default eventually, according to the analysis. That would lock up a massive share of the market, since most users never actively choose a specific AI model anyway.
The pattern is clear. Microsoft's proprietary push prioritizes margins over capability, and its communication hides unfavorable comparisons. For developers who care about output quality and cost, the open-weight alternative from DeepSeek remains the stronger choice.