According to Axios, Microsoft is contemplating the integration of DeepSeek's V4 large-scale model—developed by a prominent Chinese AI firm—as a cost-effective substitute for its Copilot collaborative office intelligent assistant. Presently, Microsoft is conducting tests on locally deployed versions of DeepSeek's V4 and other open-source large-scale models within Microsoft Cloud. The goal is to replace the existing Anthropic and OpenAI models that currently power Copilot, thereby aiming to cut down on service expenses. Microsoft is set to roll out this budget-friendly large-scale model service in the forthcoming weeks.

Charles Lamanna, the Corporate Vice President of Microsoft Copilot Business, highlighted that certain users engage in hundreds of tasks on a weekly basis. While this boosts efficiency, it has also led to a surge in costs. To illustrate, Anthropic's newly launched Fable 5 model is priced at $50 per million tokens. In contrast, DeepSeek's V4 Pro model, after discounts, is available for a mere $0.87 per million tokens, marking a staggering cost difference of roughly 57 times. For standard office tasks, open-source models can already deliver functionality on par with that of closed-source models, with overall usage costs slashed by over 90%.

Microsoft has clarified that should DeepSeek's V4 model be the final choice, it will be entirely hosted on the Microsoft Azure cloud platform. This ensures that customer data remains exclusively within Microsoft's cloud ecosystem, safeguarded by Azure's enterprise-level security, compliance, and data residency protections. Furthermore, Microsoft has tailored and fine-tuned the model, incorporating safety enhancements such as bias reduction and content compliance screening.

DeepSeek recently secured its inaugural round of external financing, amassing over RMB 50 billion, with the company now valued at approximately RMB 338 billion. This positions it as one of the most highly valued AI companies in China.