Elon Musk’s artificial-intelligence startup xAI has released Grok 4, its newest large-language model, only a few months after debuting the previous version. The company says it used roughly 100-times more compute than for Grok 2, enabling the system to top the Artificial Analysis Intelligence Index with a score of 73, ahead of OpenAI’s O3 and Google’s Gemini 2.5 Pro, which each scored 70. Independent evaluations show Grok 4 achieving 100% on mathematical Olympiad problems and 44.4% on the arduous “Humanity’s Last Exam,” nearly doubling the prior record. xAI is pitching the model at $3 to $15 per million tokens for developers and $30 a month for consumer access, with a higher-capacity “Grok 4 Heavy” tier priced at $300. Beyond its commercial API, xAI is targeting public-sector work. TheStreet reported that the company secured a Department of Defense contract with a ceiling of up to $200 million and has introduced a “Grok for Government” toolkit, positioning the startup against long-time federal analytics provider Palantir. The rapid rollout has drawn scrutiny from rivals. Researchers at OpenAI and Anthropic describe xAI’s safety culture as “reckless” and “completely irresponsible,” faulting the firm for launching Grok 4 without publishing a system card or the results of so-called dangerous-capability evaluations. The criticism follows earlier incidents in which Grok generated antisemitic statements and controversial AI ‘companion’ personas. xAI adviser Dan Hendrycks said internal safety tests were performed but has not released the findings. The dispute adds momentum to proposed legislation in California and New York that would compel frontier-model developers—including xAI—to disclose safety methodologies before public deployment.