Microsoft launches affordable AI models for coding and cybersecurity
The tech giant is releasing specialized models that aim to match the performance of top competitors like OpenAI and Anthropic at significantly lower costs.

The short version
- The MAI Code 2 model reportedly delivers 98% of GPT-6.1 Sol's performance at only 40% of the price.
- Code 2 Flash is designed for high-volume tasks like routine code fixes and explanations.
- Cyber 2 Flash is a security-focused model that Microsoft says outperforms frontier models at a lower cost per task.
- Microsoft is introducing MAI Licensed Weights to give trusted customers access to model weights for deployment on Azure.
Why it matters
For businesses, these models offer a way to use high-quality AI for specific workflows without the high overhead of 'frontier' models. It signals a shift toward specialized, cost-effective AI tools rather than just bigger, more expensive general models.
A new focus on specialization
Microsoft officially unveiled a new suite of AI models during its Ignite conference in San Francisco. These models are specifically designed for two major enterprise needs: writing software code and managing cybersecurity. CEO Satya Nadella announced the launch, highlighting that these tools are built to run at a fraction of the cost of the industry's largest frontier models.
The company is moving toward a strategy of domain-specific AI. Instead of trying to make one model do everything perfectly, Microsoft is creating models that excel at narrow tasks. This approach allows the company to maintain high performance while significantly reducing the expenses associated with running large-scale AI systems.
By focusing on these niches, Microsoft aims to compete directly with leaders like OpenAI and Anthropic. The goal is to provide tools that are efficient enough for high-volume work but powerful enough to handle complex professional requirements. These models are already being integrated into existing platforms like GitHub Copilot and Foundry.
Comparing performance and price
Microsoft provided specific benchmarks to show how its new models stack up against the competition. The MAI Code 2 model is reported to deliver 98% of the performance of OpenAI’s GPT-6.1 Sol while costing only 40% as much. When compared to Anthropic's Claude Opus 5.5, the model reaches 96% of the performance at just 5% of the cost.
For faster, high-volume tasks, the company introduced Code 2 Flash. This version is intended for routine work such as explaining code, making targeted edits, and performing simple fixes. Microsoft claims this model matches OpenAI's GPT-6 Luna while costing 65% as much, and it matches Anthropic's Claude Haiku 5.5 at one-fifth of the price.
The security-focused Cyber 2 Flash model follows a similar logic. Microsoft states that it outperforms frontier models at a much lower cost per task. On a specific security benchmark, Cyber 2 Flash reportedly outscores OpenAI's GPT-6 Astra while costing under 2% as much for each individual task.
How specialized models work
The technical logic behind these models relies on the idea of co-optimization. Microsoft AI chief Mustafa Suleyman explained that tailoring technology to specific domains helps keep expenses down. By narrowing the scope of what a model needs to know, the company can fit more relevant knowledge into the same number of parameters.
Parameters are the internal variables that determine how an AI processes information. A model that is only trained on routine coding work will naturally perform worse at complex tasks like scientific research or advanced mathematics. However, this trade-off is what allows the models to be faster and cheaper for the specific jobs they were built to do.
Suleyman described the approach as a system rather than a single product. He called it a "hill-climbing machine" that is designed to balance performance and price. This suggests that Microsoft is building a repeatable framework for creating specialized AI rather than just releasing one-off models.
Access and self-sufficiency
Microsoft is also changing how customers can interact with its underlying technology. The company announced MAI Licensed Weights, which will allow trusted customers to access the actual weights of its models. To use these licensed weights, customers must deploy them within the Azure cloud environment.
This move follows a broader strategy to increase self-sufficiency. In June, Microsoft introduced seven in-house MAI models to reduce its reliance on outside providers like OpenAI. While the Copilot app still uses models from both OpenAI and Anthropic, the company is increasingly leaning on its own internal technology.
For security specifically, Microsoft is taking a more controlled approach. The company intends to restrict access to the new Cyber 2 Flash model to a small set of customers. This model will be integrated with Perception, which is Microsoft's agentic security system that also became generally available on Tuesday.
What comes next for the market
The release of these models signals a shift toward the commoditization of specialized AI. As companies look to integrate AI into their daily workflows, the high cost of frontier models may become a barrier to entry. Microsoft's move suggests that the next phase of the AI boom will be defined by cost-efficiency and task-specific accuracy.
For developers and security teams, this means they may soon have access to tools that are
What to watch
- How many customers will eventually opt for licensed weights over standard API access?
- Whether other providers will respond by releasing their own low-cost, specialized models.
Try it
Check the GitHub Copilot settings to see if the new Code 2 or Code 2 Flash models are available for your workspace.
Sources and links
- Reportingqz.com: the article we sourced
First reported by qz.com. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.