studio raoulsoftware innovation, architecture & development
← Blog
AI in Practice · November 18, 2026

Microsoft's new AI models cost a fraction of rivals — what that means for your budget

Microsoft launched seven new AI models at Ignite 2026 that match the performance of OpenAI's and Anthropic's best, but at dramatically lower prices — which should change how any team budgeting for AI thinks.

Key facts
7
new AI models launched at Ignite
~40%
price of MAI Code 2 vs. GPT-6.1 Sol
<2%
cost of MAI Cyber 2 Flash vs. GPT-6 Astra
Maia 200
Microsoft's own chip powering all seven models

Microsoft launched seven new AI models at its Ignite conference on November 17. The headline numbers are striking: MAI Code 2, its flagship coding model, matches 98% of OpenAI's GPT-6.1 Sol coding quality at about 40% of the price. MAI Cyber 2 Flash, built for security work, matches a comparable frontier model at under 2% of the cost per task. These are Microsoft's own benchmark claims, but all seven models are already running in production on Microsoft's Maia 200 AI chips — this is not a preview.

Until recently, Microsoft was best known for investing in OpenAI, not competing with it. That is now changing. The seven models cover coding, cybersecurity, image generation, voice and transcription — the same ground the big AI labs are fighting over. Microsoft is now building its own frontier AI, not just reselling others'. The models are already live in GitHub Copilot and Azure, which means millions of developers have them available today.

The pricing claims come from Microsoft's own benchmarks, so treat the exact percentages with appropriate skepticism. But the direction is clear: frontier-quality AI is getting dramatically cheaper. When one large player prices at 40% of a competitor's cost, the market tends to follow. We saw this with cloud storage and compute — AWS cut prices, competitors followed. AI model pricing is likely following the same curve.

For anyone building a product or automating work with AI, now is a good time to review your model choices. The cheapest option that meets your quality bar is probably much closer to the best option than it was six months ago. Microsoft also announced Licensed Weights — meaning companies can run MAI models inside their own Azure environment instead of sending data to a shared API. If AI automation has felt too expensive to justify, take another look at the numbers today.

Sources
← When AI Gets Cheap Enough to Run All Day← Anyone can now hail a driverless car on Uber — here's why that matters
Microsoft's new AI models cost a fraction of rivals — what that means for your budget