Microsoft Copilot Cowork hits GA worldwide with multi-model support

Microsoft made Copilot Cowork generally available worldwide, with CEO Satya Nadella announcing on LinkedIn (9,623 likes) that 'every organization can put long-running agents to work on complex, multi-step tasks, grounded in your organization's unique knowledge and know-how.' The headline GA feature is multi-model support, repositioning Copilot from a single-model assistant into an enterprise agent platform.

Copilot EVP Charles Lamanna detailed GA improvements across model choice, extensibility through plugins, browser automation, and cost-management controls, following months of preview in Microsoft's Frontier program. Currently Cowork runs on models from Anthropic and OpenAI, with reports indicating Microsoft is considering a fine-tuned DeepSeek V4 as a lower-cost option and preparing its own in-house Co-work 1 model.

The multi-model and cost-control emphasis is strategically loaded. Reports this week also detailed Microsoft migrating developers off Claude Code to its in-house Copilot CLI, and Amazon and Walmart curbing employee AI spend — part of a broader 'great coding reset' theme where enterprises scrutinize per-token AI costs. Offering cheaper model routing (potentially DeepSeek) directly addresses customer cost anxiety while reducing Microsoft's dependence on Anthropic and OpenAI inference bills.

Competitively, Cowork squares off against AWS's agentic stack (Quick, AgentCore, DevOps Agent, all updated this week at Summit NY) and Google's Vertex agents. The differentiator Microsoft is betting on is grounding in Microsoft 365 data and Work IQ APIs. The caveat: the same week's M365 Copilot EchoLeak/SearchLeak zero-click vulnerability is a stark reminder that deeply data-grounded agents widen the prompt-injection attack surface. Enterprises will weigh productivity gains against that expanding security exposure as Cowork rolls out broadly.