xAI's Grok models roll out to Microsoft Copilot

Nadella framed it simply — 'More model choice coming to Copilot. Welcome Grok!' — and Musk amplified with a terse 'Grok available in Microsoft Copilot.' The move slots Grok alongside OpenAI models and Microsoft's own MAI systems as a user-selectable option, extending a multi-model strategy Microsoft has pursued aggressively as its OpenAI relationship acquired escape hatches.

Mechanically, this is model selection within the consumer and enterprise Copilot surfaces rather than a deep platform bet: users pick Grok for a query the way they'd pick a different engine. It builds on Grok 4.1 Fast's earlier availability in Copilot Studio for US builders, so the plumbing was partly in place. Grok itself runs on recent generations including Grok 4.6, which shipped August 12 with a 500K-token context and an 'xhigh' reasoning tier.

Competitively, this is Microsoft continuing to hedge. The same week, reports detailed deepening Microsoft–Anthropic ties, with Azure selling Anthropic frontier models to enterprises. Offering Grok, Claude and OpenAI models under one roof lets Microsoft position Copilot as a neutral aggregation layer rather than a single-vendor front end — valuable leverage as OpenAI's own infrastructure strains under GPT-6 Astra demand.

Open questions: enterprise customers will scrutinize data-handling and content-moderation differences given Grok's looser guardrails, and pricing wasn't detailed. Cross-model comparisons circulating this week also noted Grok 4.6 raised cached-input pricing from $0.30 to $0.50 per million tokens, a reminder that 'more choice' comes with its own cost math.