Microsoft Copilot Cowork considers using DeepSeek V4: 57 times price difference forces giants to switch

Microsoft Copilot Cowork moved to pay-as-you-go billing and considered a tweaked version of DeepSeek V4 as a low-cost alternative. The output pricing difference between Fable 5 and DeepSeek V4 Pro is 57 times, and the model will be fully hosted on Azure to ensure data security.

According to axios, Microsoft is moving the Copilot Cowork agent to a usage-based pricing model and is also exploring the use of a fine-tuned version of DeepSeek V4 as a low-cost alternative to Anthropic and OpenAI models. This is the first time that an international technology giant has publicly stated that it will include DeepSeek in its supplier options.

The driving force behind this is the huge price gap. Agent tools like Copilot Cowork, Claude Code, and Codex require continuous and repeated calls to the underlying AI model when completing complex tasks—this greatly improves productivity, but also brings "crazy AI bills." Anthropic’s latest flagship Fable 5 output is priced at $50/million tokens, while the output of DeepSeek V4 Pro is only $0.87/million tokens after a permanent 25% discount—a difference of about 57 times. For very large-scale enterprise deployments, this price difference means a cost difference of millions or even tens of millions of dollars.

Microsoft Copilot considers using DeepSeek

Charles Lamanna, executive vice president of Copilot, Agents and Platforms at Microsoft, said companies using Copilot Cowork will pay based on the amount of computation they perform. In the meantime, Microsoft is evaluating a tweaked version of DeepSeek V4 or another open source model, and expects to confirm the selection and launch a lower-cost model option in the coming weeks.

On the security and compliance level, Microsoft promises that models will be fully hosted on Azure, with customer data remaining in the Microsoft cloud and protected by Azure's enterprise security, compliance and data residency controls. This essentially solves the previous biggest concern of enterprise customers about DeepSeek data security - the model is trained in China, but the inference is completed on the US cloud, and the data does not leave the country.

From an industry perspective, Microsoft's move has multiple strategic implications. First, this is a clear price reduction signal for Anthropic and OpenAI - the introduction of DeepSeek can both be used as a bargaining chip to lower supplier quotes, and can also truly reduce Copilot Cowork's operating costs. Second, Microsoft deployed DeepSeek on Azure instead of directly calling Chinese servers, cleverly avoiding data sovereignty disputes and providing a replicable compliance path for other overseas companies to use Chinese AI models. Third, Copilot Cowork shifts to a combination of pay-as-you-go billing + low-price model, marking the shift in the pricing model of enterprise AI tools from a "subscription system" to a "volume-based system + cost optimization" dual-wheel drive.

The follow-up is worth paying attention to: the exact model and timeline that Microsoft finally chooses; whether Anthropic/OpenAI will adjust its enterprise-level pricing strategy under the 57-fold price difference; whether DeepSeek's overseas enterprise revenue through the Azure channel can become a new growth pole; and whether other cloud service providers (AWS, GCP) will follow up with similar Chinese model hosting solutions.

Copyright: Content sourced from IT Home / axios . This platform has compiled and organized this content for informational purposes and learning exchange only. If there are any copyright concerns, please contact us for resolution.

Reviews

  • Loading reviews...