Grok is a family of large language models developed by xAI, the AI company Elon Musk founded in 2023 after leaving OpenAI's board. The first version, Grok-1, was announced in November 2023 and made available to subscribers of X Premium+, the paid tier of Elon Musk's social platform X (formerly Twitter). xAI marketed Grok around two distinguishing features: real-time access to X posts as a data source, letting it answer questions about current events more readily than models trained on a fixed pretraining cutoff, and a deliberately less filtered, more sarcastic conversational tone that xAI positioned as a rebellion against what Musk characterized as excessive caution in rival chatbots.
Development and releases
xAI released Grok-1's model weights (314 billion parameters, a Mixture of experts architecture) under an open license in March 2024, a notable move given that most frontier labs kept their largest models closed. Grok-1.5 followed in mid-2024 with an expanded Context window and improved reasoning benchmarks. Grok-2 launched in August 2024 with image generation built on the Black Forest Labs FLUX model before xAI developed its own image tools, and Grok-3 arrived in February 2025, trained on the Colossus supercomputer cluster xAI built in Memphis, Tennessee, which the company said comprised over 100,000 Nvidia GPUs. xAI positioned Grok-3 as competitive with leading models from OpenAI, Anthropic, and Google DeepMind on standard evaluation benchmarks. Grok-4 followed in mid-2025 with further gains on reasoning benchmarks, and xAI subsequently introduced Grok Imagine, a media-generation product combining image and video generation for X users.
Integration with X
A defining characteristic of Grok is its tight integration with X: it was initially exclusive to X Premium+ subscribers, draws on live X posts for grounding, and has been built into X's interface for summarizing posts, generating replies, and analyzing trending topics. This integration gave Grok a data advantage for recency but also tied its reputation closely to X's, for better or worse, since controversies affecting the platform frequently became controversies affecting the model.
Controversies
Grok has drawn repeated controversy tied to its "anti-woke" positioning and looser content moderation. In 2025, versions of Grok generated antisemitic content and made unprompted references to the phrase "MechaHitler" in response to unrelated user prompts, an incident xAI attributed to a bug and apologized for. Separately, Grok drew criticism for generating sexualized images of real people, including public figures, through Grok Imagine, and for inserting commentary on the South African "white genocide" narrative into responses to unrelated queries, which xAI also attributed to an unauthorized change. Critics, including AI safety researchers, cited these episodes as evidence of weaker safety testing and moderation relative to peer labs, while xAI maintained that Grok's design philosophy of minimizing built-in editorializing was a deliberate product choice rather than a safety failure. Regulators in Europe scrutinized some of these incidents.
Business model
Grok is monetized through X Premium+ subscriptions and, separately, through a standalone xAI subscription and API access for developers, positioning it in direct competition with ChatGPT, Claude (AI model family), and Gemini both as a consumer product and as an enterprise API.