Axios reports that Microsoft is moving Copilot Cowork to usage-based pricing and is exploring a fine-tuned version of DeepSeek V4, or another open-source model, as a lower-cost option for the enterprise agent, Axios reports. Axios also reports Microsoft expects to make a lower-cost model available in the coming weeks and says the alternative would be optional for customers and fully hosted on Azure, keeping customer data in Microsoft's cloud and under Azure's enterprise security and compliance controls. Axios quotes Charles Lamanna, Microsoft's executive vice president for Copilot, agents and platform, saying heavy agent usage can produce very high costs for customers.
What happened
Axios reports Microsoft is moving Copilot Cowork to usage-based pricing and 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 currently used for the tool. Axios reports Microsoft expects to make a lower-cost model available in the coming weeks and says the optional model would be fully hosted on Azure, keeping customer data within Microsoft's cloud and covered by Azure's enterprise security and compliance controls. Axios also reports Microsoft has fine-tuned the model and added safeguards, including changes aimed at reducing bias. Charles Lamanna, Microsoft's executive vice president for Copilot, agents and platform, is quoted by Axios: "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."
Editorial analysis - technical context
Agentic tools such as Copilot Cowork, Anthropic's Claude Code, and OpenAI's Codex repeatedly call base models while chaining actions, which materially increases compute consumption and end-customer bills. Cheaper model options typically trade off raw capability, latency, or guardrail maturity for lower inference cost. Hosting an alternative model on Azure preserves a single-cloud control plane for telemetry, compliance, and data residency, but it does not eliminate the need for independent safety testing and bespoke fine-tuning to match enterprise expectations.
Industry context
Reporting frames this as part of a broader multi-model approach where large platform vendors mix higher-cost frontier models with lower-cost alternatives to manage total cost of ownership for customers. Axios notes that using a model from a Chinese AI company could draw criticism; observers and enterprise buyers routinely weigh geopolitics, supply-chain provenance, and compliance risk alongside cost and performance when evaluating third-party models.
What to watch
Indicators to follow include which model Microsoft confirms, published benchmark comparisons against Anthropic/OpenAI models, independent evaluations of the announced safeguards and bias mitigations, the detailed pricing metric(s) for usage-based Copilot Cowork, and any customer or regulator reaction to hosting choices or third-party provenance. For practitioners assessing vendor exhibits, attention should focus on inference cost per typical agent workflow, failure modes under chained calls, and the availability of enterprise-grade logging and isolation controls.
Scoring Rationale #
Notable product-level change from a major cloud vendor that affects cost and procurement decisions for enterprise AI. Important for practitioners assessing agent workloads and model selection, but not a frontier-model or regulatory landmark.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.