xAI strikes GSA deal for Grok after weeks of speculation 

Elon Musk’s xAI firm has reached an agreement with the General Services Administration to offer its artificial intelligence models to the government at a nominal cost, following reported delays in the wake of company controversies. 

Under the deal, federal agencies can buy xAI’s flagship model, Grok 4, and Grok 4 Fast for 42 cents until March 2027, the GSA announced Thursday. Grok 4 Fast is the company’s latest advancement in “cost-efficient reasoning models,” according to the xAI website. 

Should an agency purchase the models, the 42-cent offer lasts for 18 months — the longest term for a OneGov agreement to date, the agency emphasized. 

xAI joins the wave of technology companies offering their AI products at a severely discounted rate through the GSA’s so-called OneGov strategy. Under the initiative, the GSA aims to work directly with IT manufacturers to consolidate IT purchasing as the Trump administration pushes for AI adoption in the government. 

The agreement comes weeks after reporting from Wired revealed xAI was supposed to land a deal similar to GSA’s partnerships with competitors like OpenAI and Anthropic, but plans fell through after its chatbot espoused antisemitic and pro-Hitler content when responding to inquiries on the social media platform X.

A source told FedScoop last month that there was a June 4 meeting between GSA and xAI representatives, though Musk — a former ally and adviser to President Donald Trump —  did not attend.

Since then, the agency’s AI safety team, with help from data scientists in GSA’s Office of the Chief Data Officer, conducted an analysis of Grok 3 and Grok 4 and attempted to reproduce the reported issues, a GSA official told FedScoop. The team tested for any systemic biases and attempted to identify the root cause of the safety issues, and gave these results to the CDO’s office, which determined Grok 4 could be used within the AI safety team’s guidelines, the official added. 

The GSA official said the agency is “continuously monitoring and conducting safety benchmarks.” 

This builds upon comments from Zach Whitman, GSA’s chief AI officer, who told FedScoop last month that GSA created a process for approving “families” of AI models and establishing a new AI safety team to evaluate the models for different tasks. This includes a method of red-teaming to test models’ performance. 

As part of the GSA agreement, xAI is committing a dedicated group of engineers to assist agencies as they implement Grok into their workflows, and will offer introductory training programs and custom enablement over time, the agency said.

The deal also gives agencies an “upgrade path” to FedRAMP and DOD Impact Levels-aligned enterprise subscriptions for expanded features and rate limits.

In a statement shared by GSA, Musk said xAI has the “most powerful AI compute and most capable AI models in the world,” while xAI co-founder Ross Nordeen said the company will work to “deeply understand the needs of our government” to make America the “world leader” in AI.  

Josh Gruenbaum, commissioner of the GSA’s Federal Acquisition Service, said “widespread access to advanced AI models is essential” to government efficiency and accountability. 

FedScoop first reported on GSA’s interest in Grok in July, days before xAI announced the launch of “Grok for Government,” which is described as a suite of products tailored for U.S. government customers. 

xAI’s competitor OpenAI and Anthropic are each offering government-tailored AI models for $1 per agency per year through GSA OneGov deals, while Microsoft, Meta and Google have struck similar deals to sell their products for less than a dollar. Notably, all of these models are also available on GSA’s new evaluation platform, USAi, but xAI was not listed as an option as of publication time. 

A GSA spokesperson emphasized to FedScoop last month that the agency is considering all AI companies “equally” for government contracts and that the “absence of a particular model should not be interpreted as exclusion or a final determination.”

Concerns have been raised in recent weeks about the use of Grok in the government following the chatbot’s antisemitic responses. This occurred after an instruction was apparently added to Grok’s system prompt, directing it to “not shy away” from certain claims. The instructions were later removed. 

A group of more than 30 advocacy organizations signed a letter to the Office of Management and Budget arguing the use of Grok goes against the Trump administration’s efforts to prevent “ideological bias” in AI models.

Several Democratic lawmakers also voiced concerns to GSA following FedScoop’s reporting in July.