Microsoft considers DeepSeek V4 for Copilot Cowork as OpenAI and Anthropic costs bite

Microsoft DeepSeek: a Chinese model for Copilot Cowork?

Agentic AI is so expensive that Microsoft is metering Copilot Cowork and shopping for a cheaper engine. The frontrunner is a Chinese model, just as Washington cracks down on foreign AI. Microsoft is considering putting a Chinese AI model inside its enterprise Copilot, and the reason is money. The company told Axios it is exploring a self-hosted, fine-tuned version of DeepSeek V4, or another open-source model, as a cheaper option to power Copilot Cowork, the agentic assistant in its Microsoft 365 suite. It expects to offer a lower-cost model within weeks. At the same time, Microsoft is moving Copilot Cowork to usage-based pricing, charging companies for the compute they actually burn rather than a flat fee. Why even Microsoft can't eat the bill The shift is a window into the economics of agentic AI. Tools like Copilot Cowork, Anthropic's Claude Code and OpenAI's Codex call a model over and over as they work through a task, which is powerful and, it turns out, expensive. "We have users who do hundreds of tasks a week, which is great, they're way productive, but the consequence is the costs can go very high," said Charles Lamanna, Microsoft's executive vice-president for Copilot, agents and platform. Copilot Cowork currently runs on Anthropic and OpenAI models, both of which have raised prices and pulled back from all-you-can-eat plans. Microsoft already metered GitHub Copilot for the same reason. A cheaper open-source engine underneath is the obvious next lever. The cheapest option happens to be Chinese That is where the calculation gets awkward. DeepSeek V4, released in April, is open-source, popular with developers and far cheaper to run, which is precisely why it is on Microsoft's shortlist. It is also Chinese, and the timing could hardly be worse politically. Washington has floated banning DeepSeek, threatened Chinese AI firms, and just forced Anthropic to cut off its top models for non-US users, a dispute that escalated into crisis talks with the Commerce Department. Microsoft is clearly aware of how this looks. It says any DeepSeek option would be optional for customers and fully hosted on Azure, keeping data inside Microsoft's cloud under its security, compliance and data-residency controls, and that it has fine-tuned the model and added safeguards, including changes meant to reduce bias. A hedge against its own suppliers The bigger picture is that Microsoft no longer wants to depend on any single lab. Freed from its tight, often tense exclusivity with OpenAI, it is pushing a multi-model approach, mixing and matching engines under its own roof. For now this is an evaluation, not a decision, and Microsoft says it will confirm its choice when the cheaper tier ships. But that it is willing to even name DeepSeek as a candidate, in this climate, says a lot about how hard the cost of running agents has started to bite.

Microsoft eyes DeepSeek for enterprise AI

Why it matters: Microsoft's move to add a model from a Chinese AI company could draw criticism. The big picture: Agentic tools like Copilot Cowork, Anthropic's Claude Code and OpenAI's Codex can keep calling AI models as they work through tasks -- boosting productivity but also creating bonkers AI bills. Driving the news: Microsoft says companies using Copilot Cowork will pay based on how much compute they use. * The company tells Axios it is exploring a fine-tuned version of DeepSeek V4, or another open-source model, as a lower-cost alternative to the Anthropic and OpenAI models now powering Copilot Cowork. * Microsoft says it expects to make a lower-cost model available in the coming weeks and will confirm its choice then. Zoom out: The testing reflects Microsoft's broader push toward a multi-model approach, rather than relying only on models from OpenAI and Anthropic. Between the lines: If Microsoft goes forward with DeepSeek, the company says the model would be optional for customers and fully hosted on Azure, keeping customer data within Microsoft's cloud and covered by Azure's enterprise security, compliance and data-residency controls. * Microsoft says it has also fine-tuned the model and added safeguards, including changes aimed at reducing bias. What they're saying: Charles Lamanna, Microsoft's executive vice president for Copilot, agents and platform, told Axios that testing showed Copilot Cowork could not be offered on an unlimited-use basis.

Microsoft Risks Trump's Ire By Abandoning The Costly OpenAI And Anthropic Models For China-Based DeepSeek's V4 Model For Enterprise Workloads

In a possible development that is almost guaranteed to raise a lot of hackles in Washington, Microsoft is considering the use of a self-hosted version of DeepSeek's V4 model for Copilot Cowork, as OpenAI and Anthropic appear determined to price themselves out of the market. Soaring token costs are driving enterprise customers away from OpenAI and Anthropic and right into the lap of China's DeepSeek For the benefit of those who might not be aware, a token is the smallest unit of data that an AI model processes, and is typically around 4-character long. A model's context window or working memory is entirely measured in tokens, and spans tokens consumed for understanding the input prompt as well as those used to generate the relevant output. As model complexity and agentic workloads increase, however, token costs are becoming a critical inhibitor, especially for coding-related tasks where the token consumption might soar unexpectedly. After all, who can forget Uber blowing through its entire AI budget for 2026 in just 4 months after incentivizing employees to use more AI? In fact, tech-savvy employees are now adopting tokenmaxxing as a way of topping internal AI-use leaderboards by using AI models for menial tasks, replete with long prompts and agentic loops. To add insult to the proverbial injury, OpenAI and Anthropic are not only increasing pricing for their enterprise plans, but are also using creative ways to limit the total number of tokens that can be consumed. This brings us to the core of today's topic. Axios is out with a scoop today, detailing that Microsoft might opt for DeeSeek's V4 model, or another similar one, albeit hosted on its own infrastructure, to power Copliot Cowork, which combines Copilot's enterprise-related functions with advanced AI models from OpenAI and Anthropic to make agentic AI-powered workflows a relative breeze. The change has been spurred by Microsoft's decision to transition Copilot Cowork towards a metered architecture where you pay for the total tokens consumed instead of a flat rate. Even though DeepSeek's models are based on an open-source architecture, such a move is unlikely to sit well in Washington, which just compelled Anthropic to pull its Mythos-class Fable 5 model from all users who are not U.S. citizens after Amazon allegedly disclosed a way to jailbreak the model's advanced cyber capabilities. Some have speculated, however, that the move starves China-based AI players who've made distillation of American models into an artform. Finally, the latest tidbit comes at a time when DeepSeek has just raised $7.4 billion at a $50 billion valuation, and is racing ahead to expand its compute footprint. Follow Wccftech on Google to get more of our news coverage in your feeds.