Microsoft Considers DeepSeek for Lower-cost Copilot
The framing has firmed up since this story first ran: Microsoft's own spokesperson told Computerworld that DeepSeek is 'one of several models under consideration' for a lower-cost Copilot Cowork tier - not a settled decision, despite headlines suggesting otherwise. Copilot Cowork became generally available June 16 with usage-based billing on top of its existing $30/user/month (enterprise) or $20/user/month (business) Microsoft 365 Copilot license: customers pay in Copilot Credits (1 cent each, pay-as-you-go, or a discounted committed-volume tier) calculated from model use, context retrieval, tool calls, and runtime. Today's confirmed model lineup is Anthropic's Opus 4.8 and Sonnet 4.6 generally available, with OpenAI's GPT 5.5 and Microsoft's own Cowork 1 model reserved for the Frontier preview program - DeepSeek is not yet in that lineup. Microsoft said it will share specifics on any additional lower-cost model closer to release. For practitioners, the more durable signal is the credit-metered pricing itself: as agentic workflows chain multiple model calls per task, usage-based billing - not the choice of a specific cheaper model - is what actually exposes runaway AI compute costs to finance teams and forces governance decisions.
The billing mechanics matter more than the DeepSeek speculation: usage-based Copilot Credits are what actually expose the cost of agentic AI to enterprise budgets, and that shift happened on schedule regardless of which model Microsoft eventually adds as a discount tier.
What happened
Axios first reported June 12 that Microsoft was exploring a fine-tuned, Azure-hosted version of DeepSeek V4, or another open-weight model, as a cheaper alternative to the Anthropic and OpenAI models powering Copilot Cowork. Copilot Cowork then went generally available June 16 with usage-based billing layered on top of the existing Microsoft 365 Copilot license ($30/user/month for large enterprises before discounts, $20/user/month for Microsoft 365 Copilot for Business), per Computerworld. Usage is billed in Copilot Credits (1 cent each on a pay-as-you-go basis, or at a discount under a committed-volume P3 tier), calculated from four factors: model use, context retrieval, tool calls, and runtime - meaning tasks that pull from more sources, use deeper reasoning, or generate multiple outputs cost more. Cowork is off by default; IT admins control rollout and can set spending caps at the tenant, group, and user level.
The DeepSeek status, precisely
Computerworld obtained a direct statement from a Microsoft spokesperson: "We are actively exploring a range of options that meet those requirements, and DeepSeek is one of several models under consideration. We'll share more specifics on the underlying model closer to release." That is a materially more cautious framing than 'Microsoft picks DeepSeek' - as of general availability, the confirmed model lineup is Anthropic's Opus 4.8 and Sonnet 4.6 (generally available), plus OpenAI's GPT 5.5 and Microsoft's own Cowork 1 model for customers enrolled in the Frontier preview program. No DeepSeek or other open-weight option has shipped.
For practitioners
Two separate decisions are getting conflated in coverage of this story: which model Microsoft eventually adds as a lower-cost tier, and the shift to usage-based billing itself. The second is already live and is the more consequential one for IT budgeting - it converts agentic AI from a flat per-seat cost into a metered compute expense that scales with how aggressively teams delegate work to Cowork. Organizations evaluating Cowork should prioritize setting spending caps and credit-visibility tooling now, independent of whichever model Microsoft adds later for cost optimization.
What to watch
Whether Microsoft names a confirmed lower-cost model (DeepSeek or otherwise) and on what timeline; enterprise reaction to a Chinese-origin model if DeepSeek is chosen, given data-governance and procurement scrutiny that typically accompanies such choices; and early usage-based billing data once per-task credit costs become visible to customers, which Microsoft said would arrive "soon after" general availability.
Key Points
- 1Microsoft's spokesperson told Computerworld DeepSeek is 'one of several models under consideration' for Copilot Cowork, not a confirmed decision.
- 2Copilot Cowork went GA June 16 with usage-based Copilot Credits on top of its $20-30/user/month license, billed by model use, context, and runtime.
- 3The confirmed GA lineup is Anthropic's Opus 4.8/Sonnet 4.6; OpenAI's GPT 5.5 and Microsoft's own Cowork 1 are Frontier-only. No lower-cost model has shipped.
Scoring Rationale
Updated with Microsoft's own on-record spokesperson statement (via Computerworld) clarifying that DeepSeek remains one option under consideration rather than a settled decision, plus confirmed general-availability pricing mechanics and current model lineup - correcting the more speculative framing implied by the original single-source Axios report. The usage-based billing shift is a genuinely notable, already-confirmed change for enterprise AI cost governance; the DeepSeek angle itself remains unconfirmed. Score held at 6.9, reflecting confirmed strategic significance (multi-model, metered pricing) balanced against the still-unresolved DeepSeek specifics.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
