Microsoft is shifting Copilot Cowork to usage-based pricing and is exploring a lower-cost model option, according to reporting by Axios and follow-ups. Axios reports Microsoft told the outlet it is testing a fine-tuned, self-hosted version of DeepSeek V4 or another open-source model as a cheaper alternative to the Anthropic and OpenAI models currently used. Axios also reports Microsoft expects to make a lower-cost model available in the coming weeks. Reporting by Axios and The Next Web says any DeepSeek option would be optional for customers and would be hosted on Azure, keeping data inside Microsoft's cloud. Reporting by IT之家 (IT Home) provides a token-price comparison, saying Anthropic's latest flagship model costs about $50 per million tokens while DeepSeek V4 Pro costs about $0.87 per million tokens, a roughly 57x price gap.
What happened
Microsoft is moving Copilot Cowork to usage-based pricing, reporting that customers will pay based on compute consumption rather than a flat fee, per reporting by Axios and follow-up coverage in major outlets. Per Axios, Microsoft told the outlet it is exploring a fine-tuned, self-hosted version of DeepSeek V4 or another open-source model as a lower-cost alternative to the Anthropic and OpenAI models that currently power Copilot Cowork. Axios reports Microsoft expects to make a lower-cost model available in the coming weeks and will confirm its choice then. The Next Web and other outlets report Microsoft says any DeepSeek option would be optional for customers and would be hosted on Azure so customer data remains inside Microsoft's cloud. IT之家 (IT Home) reports a published comparison that places Anthropic's latest flagship pricing near $50 per million tokens versus $0.87 per million tokens for DeepSeek V4 Pro, a roughly 57x difference.
Editorial analysis - technical context
The reporting frames this move as a response to the economics of agentic workflows. Agentic tools like Copilot Cowork, Anthropic Claude Code, and OpenAI Codex repeatedly invoke models while executing multi-step tasks, which increases token usage and compute. Industry-pattern observations: teams deploying agentic assistants commonly see token and API-call volume multiply because an agent iterates through planning, tool use, and verification steps, so the marginal cost per session can exceed expectations. Open-source, self-hosted models like DeepSeek V4 reduce per-token spend by removing commercial API margins, but they shift work to model hosting, optimization, and safety engineering.
Industry context
Reporting highlights a political and governance dimension: several outlets note that DeepSeek is Chinese-developed and has attracted regulatory scrutiny in Washington, making its inclusion politically sensitive, per The Next Web. Industry-pattern observations: when large cloud providers consider integrating foreign open-source models, reviewers typically examine data residency, export controls, and customer risk profiles. Microsoft's reported decision to host any alternative on Azure and to apply fine-tuning and safety filters is the common mitigation path companies use to combine cost savings with enterprise compliance requirements.
What to watch
- •Confirmation timing: Axios reports Microsoft expects to name the lower-cost model within weeks; observers will watch that announcement for model choice and deployment details. - •Pricing and performance trade-offs: practitioners should compare throughput, latency, instruction-following, hallucination rates, and safety mitigations between the incumbent Anthropic/OpenAI models and any chosen open-source alternative. Industry-pattern observations: total cost of ownership often depends on inference efficiency, required guardrails, and monitoring overhead, not only per-token list prices.
- •Regulatory signals: reporting notes political scrutiny of Chinese models, so governments or enterprise compliance teams may set new constraints that affect adoption windows.
Reported quotes and sourcing
Charles Lamanna, Microsoft's executive vice-president for Copilot, agents and platform, is quoted in coverage describing heavy per-user task volumes that drive costs, as reported by The Next Web. Pricing comparisons come from IT之家 (IT Home), which reports the $50 per million token figure for Anthropic and $0.87 per million token for DeepSeek V4 Pro, yielding the approximately 57x price gap cited above.
Bottom line
Reporting across Axios, The Next Web, Yahoo/Ina Fried, Gizmochina, IT Home, and others frames Microsoft's exploration of DeepSeek V4 as a cost-management step for agentic enterprise features, with attendant technical, compliance, and political trade-offs to monitor.
Scoring Rationale #
This story affects enterprise practitioners and platform teams because a major vendor is testing lower-cost model options and shifting pricing for agentic features. The operational and compliance implications are notable but not yet decisive because Microsoft has not confirmed the final model choice.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.