Pentagon deploys xAI's Grok for Government to 1.7M users via Starshield AI

xAI has landed a major government deployment: the Department of War's Starshield AI initiative put Grok for Government in front of 1.7 million users as of August 31. It joins OpenAI's ChatGPT Mil and Google's Gemini on Pentagon platforms, cementing a multi-vendor frontier-AI footprint across the U.S. defense establishment within a single week.

The underlying model is Grok 4.6, released around August 12 with a 500,000-token context window built for coding and agentic tasks, available via API and Microsoft Foundry. AWS also added Grok 4.6 to Bedrock with cross-region inference. On price, community reaction has praised Grok 4.6 as a cost disruptor at roughly $8 per 1M combined tokens versus $30-60 for Opus/Sonnet — though developers flag a long-context billing cliff that doubles pricing past 200K tokens as friction for agent workloads. Grok 4.7 has slipped into early September, promising further token efficiency.

xAI also released Grok Imagine Image 2.0, a designer-grade image model claiming #2 Arena rankings for text-to-image generation and editing, and launched Grok Bot, a $300/month persistent cloud agent service. The government win is significant validation for a lab that has moved fast on both capability and cost, but the scale of the deployment — 1.7M users on defense infrastructure — will intensify scrutiny of Grok's safety guardrails and content behavior in high-stakes contexts, an area where xAI has historically taken a lighter-touch approach than rivals.