Microsoft Ditches OpenAI Models for Cheaper In-House Alternatives

Microsoft is reducing its reliance on external partners by rolling out seven specialized in-house AI models across code, security, images, speech, and routing. According to The Decoder, the move aims to slash operating costs for companies running always-on enterprise agents.
How the new MAI models stack up on cost
The centerpiece of the release, MAI Code 2, reaches 98 percent of GPT-6.1 Sol's performance across four benchmarks while costing only 40 percent of the price. Microsoft priced the coding model at a promotional rate of $0.50 per million input tokens and $2.50 per million output tokens through May 2027.
Meanwhile, the security model MAI Cyber 2 Flash handles tasks at less than 2 percent of the cost of GPT-6 Astra. On the CyberGym-E2E benchmark, it solves 92.3 percent of tasks, reducing the per-task cost from $34.38 down to $0.45.
How to access and run the new models
Enterprise users can test and integrate these models directly through existing Microsoft platforms:
- GitHub Copilot: MAI Code 2 and Code 2 Flash are already available for everyday development tasks.
- Microsoft Foundry: The Model Router now automatically selects from the expanded pool of in-house models.
- Azure Cloud: Enterprise customers can fine-tune selected model weights using their own data via MAI Licensed Weights.
Are Microsoft's cheaper models worth using?
For organizations running continuous background agents, the token savings are substantial. While foundational research still requires heavier models, routine coding, security scans, and document retrieval run faster and significantly cheaper with Microsoft's new hardware-optimized alternatives.
Related reading
Sources
Frequently asked questions
- Which models did Microsoft release at Ignite?
- Microsoft unveiled seven specialized models covering code, IT security, image editing, speech, and workflow routing to reduce agent operating costs.
- How much does MAI Code 2 cost?
- MAI Code 2 costs $0.50 per million input tokens and $2.50 per million output tokens during its promotional period running through May 2027.
AI Briefing in 3 Minutes
Get what actually matters in AI & tech delivered free to your inbox. Zero spam, 1-click unsubscribe.
Comments
0 commentsDeixe seu comentário
Be the first to comment.


