Breaking

Microsoft Weighs DeepSeek V4 as Cheaper Copilot Cowork Option

June 16, 2026 at 21:41 EDT

Foundation Models
AI Agents
Sales & Business

Microsoft is considering adding a hosted version of DeepSeek's DeepSeek V4 model on Azure as a lower-cost option for its enterprise agentic feature "Copilot Cowork" as it shifts to usage-based pricing, according to a report by Axios.

June 16, 2026 · Microsoft 365 Copilot

Microsoft Eyes a China-Made Model to Cut the Cost of Its AI Agent

As Copilot Cowork shifts to usage-based pricing, Microsoft says it is exploring an Azure-hosted, fine-tuned DeepSeek V4 as a cheaper alternative to Anthropic and OpenAI models — a low-cost option due within weeks.

$0.435

per M input tokens (API example)

token context length

80.6%

SWE-bench Verified

93.5

LiveCodeBench

Why cost matters: input vs. output pricing

DeepSeek V4 Pro API example — per million tokens

$0.435

Input

$0.87

Output (2×)

Cowork's token-heavy agentic tasks make a low-priced model attractive against pricier Claude-class options.

How a DeepSeek option would work

Fine-tuned DeepSeek V4
offered purely as a low-cost option

→

Hosted on Azure
data stays in Microsoft's cloud

→

Enterprise controls
security, compliance, data residency

Seen as a plus

"Very cheap" with Claude-class performance
Affordable even under extended use
Azure hosting can meet enterprise security needs

Cause for caution

Repetition, reduced consistency on some tasks
Privacy concerns for a Chinese-made model
National security risk questions

Still undecided

Scope and release timing remain undetermined. Microsoft is expected to finalize its model options within weeks — and has not announced whether it will ultimately adopt DeepSeek.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.