Microsoft is attempting to bridge the divide between proprietary and open-weight artificial intelligence models with a new initiative called Licensed Weights. The announcement was made during the keynote address at the company’s Ignite conference on Tuesday, where CEO Satya Nadella introduced several new offerings from Microsoft AI (MAI), the internal research lab within the company.
The Licensed Weights program represents a selective approach to open-weight AI. It offers more accessibility than the fully proprietary models from competitors like OpenAI and Anthropic, yet it does not allow for the complete public download of weights seen with models like Mistral’s Le Chonk. Instead, interested companies must work through their Microsoft account representatives to gain access to the technology.
Customization with Data Control
The core promise of Licensed Weights is the ability for customers to customize frontier models using their own proprietary data. These customized models can be run securely within the Azure cloud infrastructure. Mustafa Suleyman, CEO of MAI, emphasized that the company would only use customer information to train its models if explicit agreement was provided.
“You control who can access it. Your data and adaptations stay entirely yours,” Suleyman stated during the keynote. He further explained that Microsoft would collaborate with clients through an entity named Frontier Company to deliver the custom model and embed it into the client’s own products and agents. This structure is designed to address the concerns of data-conscious businesses that are wary of pouring their intellectual property into external systems.
Suleyman referenced a sentiment often echoed by industry leaders, including Palantir CEO Alex Karp, noting that “no one wants to pour their IQ into someone else’s model.” This approach aims to provide a middle ground for organizations seeking affordable AI solutions that allow them to retain full control over their intellectual property.
Performance and New Model Releases
Alongside the Licensed Weights announcement, Microsoft revealed new capabilities from its AI lab. The company introduced MAI Code 2, a coding model that Suleyman claimed delivers 98% of the performance of GPT 6.1 Sol at just 40% of the cost. When compared to Opus 5.5, the company stated the model operates at an incredible 5% of that cost. These figures were based on an average of four widely used benchmarks targeting everyday enterprise coding tasks commonly seen in GitHub Copilot.
Microsoft also released MAI Image 2.7, a new image editing model. The company stated that this model now leads the category on Artificial Analysis, outperforming Nano Banana Pro and several GPT image models in terms of both price and quality.
While Microsoft’s AI models may not yet directly compete with the peak capabilities of OpenAI and Anthropic, the company leverages its legacy sway with countless businesses. By offering a customizable AI option, Microsoft aims to strike a balance in a market where Western labs like Thinking Machines and Mistral are releasing more competitive open-weight models. However, Anthropic and OpenAI still hold the overwhelming majority of the enterprise AI market.
The Licensed Weights initiative diversifies the ongoing debate between open and closed AI systems. It positions Microsoft as a provider for organizations seeking a reliable blend of return on investment tools, customization options, security controls, and capability. The company has positioned itself to capture the segment of the market that requires both high performance and strict data governance.
Source: ZDNet

