More than 1.7 million enrolled against a workforce of more than 3 million is about 57 percent of the department signed up [2][3][1], leaving roughly 1.3 million still to come [2], and nine months to get there implies an average near 189,000 new accounts a month [4][3]. That rate is the asset OpenAI and xAI acquired on Monday, and it is not one a court can transfer. Anthropic's July contract with the Pentagon was worth $200 million [9]; set against the 1.7 million people already on the portal, that is roughly $118 an enrolled user [4], and the more interesting version of the comparison is that Anthropic's distribution cost is being paid in absence, not in dollars.
What makes absence expensive sits in the product descriptions. ChatGPT Mil starts with chat, files, projects and custom GPTs aimed at document-heavy unclassified work in planning, policy, logistics and administration [10], while the Grok release lists three reasoning modes and reusable playbooks explicitly meant to capture institutional knowledge [11]. Institutional knowledge held in a vendor's format is a switching cost with a monthly accrual rate, and a logistics shop that has written its acquisition market research playbook once does not rewrite it because the model list grew by one.
Read a different way, though, this comparison gets complicated on three counts. Enrollment is not usage, and the department has published sign-ups rather than sessions [2], so 57 percent may be a headcount exercise rather than a set of habits. Accreditation may be the scarce thing rather than placement, and Anthropic says it was the first frontier lab to run models on classified US networks, supporting warfighters since June 2024 [12], a tier at which a Controlled Unclassified Information portal is beside the point. And the Pentagon framed the multi-model marketplace as insurance against getting locked into one supplier [13], a stated policy that would make a later Claude listing administrative rather than negotiated.
Then there is the ruling, which the cryptopolitan account cannot hold steady: it reports that Judge Rita Lin issued a final 59-page ruling on August 27, 2026 striking down the supply-chain risk designation, and also that Lin has not issued a final ruling [8]. One of those is wrong, and the gap decides whether Anthropic is negotiating from a vacated designation or from a hearing in which the judge said the crackdown "doesn't seem to be really tailored to a stated national security concern" [14].
This is probably wrong, but the trade looks deliberate on Anthropic's side: written assurances that its models would not be used in fully autonomous weaponry or mass domestic surveillance, against a Pentagon demand for unrestricted access for all lawful purposes and a February 27 deadline it let pass [6]. The real cost of that position is not the $200 million; it is the playbook base Anthropic is not accruing while three competitors compound theirs. The thesis fails on speed: if the D.C. Circuit rules and Claude appears on GenAI.mil within weeks with usage following [5], then IL5 was the gate all along and the 1.7 million was an accounting figure.
