Microsoft says MAI Code 2 matches 98% of GPT-6.1 Sol's coding at 40% of the price
The numbers are Microsoft's own, and it says the models aren't tuned for the hardest long-running work.
On November 17, 2026, Microsoft AI used its Ignite keynote to release what it calls a family of seven new models, and it priced them against OpenAI and Anthropic. The headline claim is for coding: Microsoft says MAI Code 2 gets 98% of GPT-6.1 Sol's score at 40% of the cost.
Against Claude Opus 5.5 the gap in price is far wider. Microsoft puts MAI Code 2 at 96% of Opus 5.5's performance for 5% of the cost.
- Against GPT-6.1 Sol
- 98% of the performance at 40% of the cost
- Against Claude Opus 5.5
- 96% of the performance at 5% of the cost
- MAI Cyber 2 Flash
- GPT-6 Astra quality at less than 2% of the cost per task
- Benchmarks
- DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1
How Microsoft measured it
The coding numbers are an average over four benchmarks Microsoft picked because they match the everyday work it sees in GitHub Copilot. It also ran a blind test of front-end apps, where reviewers preferred MAI Code 2's results over GPT-6 Sol's 64% of the time.

These are Microsoft's own tests. None of the three pages we saved points to an outside lab that checked them, and the pages don't list a price per token, so the 40% and 5% figures can't be checked against a price list yet.
Microsoft is upfront about what it left out.
They’re explicitly not yet tuned for harder problems, like fundamental science research and very long horizon agentic work.
The rest of the lineup
MAI Code 2 Flash is the cheaper sibling, for code explanations, small edits and routine fixes. Microsoft says it's comparable to GPT-6 Luna at 65% of the price and to Claude Haiku 5.5 at 20%.
The security model got the boldest number. MAI Cyber 2 Flash, Microsoft says, matches GPT-6 Astra's quality on the CyberGymE2E benchmark at less than 2% of the cost per task. There's also an image model, MAI Image 2.7, that Microsoft says ranks first for image editing on Artificial Analysis, and a voice model, MAI Realtime 1, starting at $0.01 per audio minute.
The other launch is about who holds the model. MAI Licensed Weights lets a company take an MAI model, tune it on its own data and run it inside its own Azure environment. AT&T's tuned version of Code 2 Flash, the keynote says, beat 6.1 Sol at a 5G network analysis task for a fiftieth of the cost.
Why the price is the pitch
The keynote is blunt about the audience. It says the complaint Microsoft hears from every CIO and CFO is that token spending keeps climbing, and it puts the problem in one line.
And at frontier prices, the math just isn’t sustainable.
That's the part that reaches past the AI industry. Microsoft's pitch is a price that lets a business run agents 24-7 on real engineering work, and every bug fix a model handles is billed by the token. If Microsoft's numbers hold up under outside testing, a firm could run far more of those jobs on the same budget, giving up a little quality and the hardest long tasks.
The keynote turns that into an argument against paying for frontier science. A model built to push that frontier, it says, charges you for the capability on every task, routine fixes included. Microsoft's answer is to build for the everyday work instead, and the next thing to watch is a per-token price that lets anyone check the 40% and 5% figures.
More on Microsoft
- Microsoft AI's Mustafa Suleyman calls Delaware's plan for AI-run companies deeply concerningNovember 7, 2026
- Microsoft says its MAI-DxO diagnoses hard cases more than twice as accurately as doctorsOctober 22, 2026
- Microsoft says a China-aligned group used the ARTEX AI agent to breach ConnectWise and hundreds moreOctober 22, 2026
- Satya Nadella says he treats every AI agent as an insider riskOctober 18, 2026
