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

Microsoft Explores Chinese Open-Source AI Model for Enterprise Tool

Microsoft is considering DeepSeek V4, a Chinese open-source AI model, as a cheaper alternative to OpenAI and Anthropic models currently powering Copilot Cowork, its enterprise AI tool for agentic workflows1

1

. The company told Axios it is exploring a self-hosted, fine-tuned version of DeepSeek V4 or another open-source model to reduce compute costs2

2

. Microsoft expects to make a lower-cost model available within weeks and will confirm its choice then2

2

. The timing is politically fraught, coming as Washington has floated banning DeepSeek and forced Anthropic to cut off its top models for non-US users1

1

.

Agentic AI Tools Drive Unsustainable Cost Pressures

The shift reveals the challenging economics of agentic AI tools like Copilot Cowork, Anthropic's Claude Code, and OpenAI's Codex, which call models repeatedly as they work through tasks1

1

. Charles Lamanna, Microsoft's executive vice president for Copilot, agents and platform, explained that some users perform hundreds of tasks weekly, driving productivity but also sending costs very high1

1

. Testing showed Copilot Cowork could not be offered on an unlimited-use basis2

2

. Token consumption for agentic workloads can soar unexpectedly, especially for coding-related tasks, with companies like Uber reportedly burning through entire AI budgets in just four months3

3

.

Microsoft Shifts to Token-Based Pricing Model

Microsoft is moving Copilot Cowork to usage-based pricing, charging companies for the compute they actually burn rather than a flat fee1

1

. This transition toward a token-based pricing model mirrors changes Microsoft already made to GitHub Copilot for the same reason1

1

. OpenAI and Anthropic are not only increasing pricing for their enterprise plans but also using creative ways to limit total token consumption3

3

. A cheaper open-source engine underneath is the obvious next lever to pull1

1

.

Multi-Model Approach Reduces Dependence on Single Providers

The testing reflects Microsoft's broader push toward a multi-model approach rather than relying only on models from OpenAI and Anthropic2

2

. Freed from its tight, often tense exclusivity with OpenAI, Microsoft is mixing and matching engines under its own roof1

1

. This strategy serves as a hedge against its own suppliers and gives the company more flexibility as model providers adjust pricing1

1

. Microsoft no longer wants to depend on any single lab1

1

.

Geopolitical Challenges Complicate DeepSeek Integration

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 controls2

2

. Microsoft says it has fine-tuned the model and added safeguards, including changes aimed at bias reduction2

2

. The 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. citizens3

3

. That Microsoft is willing to name DeepSeek as a candidate in this climate says much about how hard the cost of running agents has started to bite1

1

. DeepSeek recently raised $7.4 billion at a $50 billion valuation and is racing to expand its compute footprint3

3

.