Microsoft is evaluating the integration of a self-hosted, fine-tuned version of Deepseek V4 as a cost-effective alternative for its Copilot Cowork platform. Concurrently, the company is transitioning Cowork from a flat-rate structure to usage-based billing. This shift addresses the high token consumption inherent in the product, which adapts Anthropic‘s Claude technology to focus on agentic reasoning. Charles Lamanna, executive vice president of Copilot, stated that the previous pricing model was unsustainable due to power users executing hundreds of tasks weekly. Microsoft previously applied a similar strategy to GitHub Copilot, moving away from subscription tiers to reflect actual consumption levels.
This strategic pivot aligns with CEO Satya Nadella’s recent advocacy for an ecosystem where enterprises select and tune specific models to match their unique cost and performance requirements. By potentially incorporating Deepseek, a Chinese-developed model, Microsoft aims to offer flexibility while maintaining data sovereignty through Azure hosting. The company emphasises that the option remains optional and includes safeguards against bias, though the move may attract scrutiny in the United States. Ultimately, this reflects Nadella’s vision of AI as a consumption business driven by intense usage. The decision on whether to adopt the Chinese model is expected within the coming weeks.
- Copilot Cowork is shifting to usage-based pricing to manage costs driven by heavy token consumption in agentic workflows.
- Microsoft is considering a self-hosted, fine-tuned Deepseek V4 model as a cheaper option for enterprise customers.
- The strategy supports Nadella’s goal of an open AI ecosystem where users select models based on specific needs and costs.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
