cd /news/artificial-intelligence/microsoft-built-its-own-coding-model… · home › topics › artificial-intelligence › article
[ARTICLE · art-171841] src=dev.to ↗ pub= topic=artificial-intelligence verified=true sentiment=↑ positive

Microsoft Built Its Own Coding Model, and the Price Is the Point

Microsoft unveiled MAI Code 2 and the faster MAI Code 2 Flash at its Ignite conference in San Francisco on November 17, making its own coding model available inside GitHub Copilot. Microsoft claims MAI Code 2 reaches about 98% of OpenAI's GPT-6.1 Sol on four self-selected coding benchmarks at roughly 40% of the cost, a figure based on a 50% promotional price. Microsoft AI chief Mustafa Suleyman told Reuters the strategy targets task-specific models rather than one frontier model, saying "This isn't just one model. It's not a flash in the pan.

key takeaway

Microsoft claims MAI Code 2 matches about 98% of GPT-6.1 Sol's coding benchmark performance at roughly 40% of the cost, though the figures come from four self-selected benchmarks and a 50% promotional price.

by read3 min views4 publishedNov 18, 2026

What happens when the company that sells you the coding assistant starts building the brain inside it? We got an answer this week, and it says a lot about where AI coding tools are heading.

On Tuesday, November 17, Microsoft unveiled MAI Code 2 at its Ignite conference in San Francisco. It is the company's own coding model, built by its Microsoft AI team, and it is available now inside GitHub Copilot. A smaller, faster sibling called MAI Code 2 Flash came with it, aimed at the quick jobs: explaining a chunk of code, making a targeted fix, clearing a routine bug.

Microsoft says MAI Code 2 delivers about 98% of the performance of OpenAI's GPT-6.1 Sol at about 40% of the cost. That is Microsoft's own claim, measured across four coding benchmarks it picked itself, and the cost figures use a 50% promotional price. So read the headline as a claim, not a verdict. Independent benchmarks will tell us how it holds up in real codebases.

But even if the real gap is wider, the direction is what matters. Most of the coding we do every day does not need the biggest, most expensive model. Debugging, refactoring, changes across a few files: that is the daily backlog, and a model that is nearly as good for a fraction of the price starts to look like the sensible default.

It is a bit like buying tools: you do not need the biggest wrench for every bolt.

For years, Microsoft's AI story has been its partnership with OpenAI. Copilot ran on someone else's frontier models. Now Microsoft is building its own, and the pitch is not that it built the smartest model in the world. The pitch is the price. Microsoft AI chief Mustafa Suleyman told Reuters the approach is about building models for specific jobs, not one giant model for everything. "This isn't just one model. It's not a flash in the pan," he said. The family announced at Ignite covers coding, cybersecurity, image generation, and voice, each tuned for its own task.

That is a real shift. Instead of renting the frontier from OpenAI and Anthropic for every task, Microsoft wants to own the everyday tier, running inside the tools developers already pay for. If the model lives inside Copilot, most developers will never choose it directly. They will just notice a new option in the dropdown.

Microsoft is also honest about the limits. The company says these models are not yet tuned for the hardest problems, like fundamental science research or very long-running agent work. This is a play for the middle, where most of us actually live.

Photo by Nubelson Fernandes on Unsplash

First, expect a price war on coding help. When one of the biggest sellers of AI coding tools starts undercutting the labs on price, everyone else has to answer. That is good news for anyone paying per token, whether you are a solo freelancer or a team watching the Copilot bill.

Second, choice is coming inside Copilot itself. Code 2 Flash is also in limited preview on Microsoft Foundry, the company's platform for building AI apps, so teams will soon be able to pick a cheaper model for routine work and save the expensive one for the hard problems. That kind of routing, simple work to the cheap model and hard work to the strong one, is how most of us will end up using AI models: not one model, but a shelf of them.

I find this genuinely exciting, and not because of the leaderboard numbers. For the last few years, the story of AI coding has been about who has the smartest model. This week, it became who can make a good model cheap enough that nobody has to think about the cost. That is when a technology stops being a demo and starts being infrastructure.

So here is my question for you: if the model writing most of your code costs half as much and does nearly the same job, how long before "which model" matters less than "what are you building with it"?

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @microsoft 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/microsoft-built-its-…] indexed:0 read:3min 2026-11-18 · —