Elon Musk's AI company, xAI, announced its plans to release two new versions of its large language model (LLM) called "Grok." Grok 2 is set to be released next month, while Grok 3 is expected to be released in December.

Grok 2 has been trained on approximately 15,000 GPUs, including NVIDIA's H100 GPUs. It is claimed that Grok 2 will perform on par with or close to GPT-4.

Grok 3 is being trained at xAI's supercomputing facility in Memphis, Tennessee. This facility claims to be the world's most powerful AI training cluster, consisting of 100,000 liquid-cooled H100 GPUs. The company aims to complete Grok 3's training in about three or four months, followed by fine-tuning and bug fixing before its anticipated release in December.

Analyst QuickTake: This news comes two months after the company raised USD 6 billion to boost R&D for future technology ventures. In April, it also launched Grok-1.5V , a multimodal model capable of interpreting textual and visual data. xAI's flagship offering, Grok, is offered exclusively on the X platform and competes with OpenAI’s ChatGPT.