Elon Musk’s xAI Rolls Out Grok 4 Featuring $300 Red Hat A Month Subscription Plan
The expanded-gorilla AI company xAI, led by Elon Musk, unleashed its newest flagship AI model Grok 4 and a new AI subscription plan, called SuperGrok Heavy, for $300 a month late Wednesday night.
Grok is xAI’s version of the models underlying OpenAI’s ChatGPT or Google’s Gemini. It can look at pictures and answer questions. In recent months, Grok has grown very close to Musk’s own social platform X, which was recently purchased by xAI. But this deeper integration also meant millions were using Grok incorrectly.
Grok 4 has a big bar to clear. The newest xAI model is intended to rival OpenAI’s pending release of GPT-5, which is said to be coming out later this summer.
“On the academic front, Grok 4 all subjects, outperforms Ph.D level, not even a close call,” Elon Musk said during a livestream on Wednesday night. “Sometimes it’s not as colloquially sensible, and at this point it has not yet invented a new technology or discovered new physics — but that’s only a matter of time.”
The debut of Grok 4 arrives in a harrowing week for the work of Musk. On Wednesday morning, Linda Yaccarino stepped down from her role as CEO of X, where she had been for almost two years. A successor has not been named.
Yaccarino’s exit comes in the wake of online backlash against Grok’s official automated X account, which most recently replied to users on numerous occasions with anti-Semitic comments, including complaints about “Jewish executives in Hollywood” and praise for Hitler. Therefore, xAI was forced into locking the Grok account and removing the offending posts.
As a result of the incident above, xAI stripped off a new segment from Grok’s public system prompt. This portion of the chatbot had been instructed not to pull any punches in making politically incorrect statements. Musk and other xAI champions have mostly dodged talking about the controversy and have called attention to Grok 4’s abilities and performance.
On Wednesday, xAI released two models, Grok 4 and Grok 4 Heavy. The latter is a “multi-agent version” that has been optimised to be better performing. Grok 4 Heavy, Musk said, allows multiple agents to work on a problem at the same time and come together “like a study group” to compare results and find the best answer.
xAI says Grok 4 scores best across various benchmarks, including the HumanEval Last Exam — a hard test that checks if AI can comprehend crowdsourced questions from domains like math, the humanities, or natural science. Per xAI, Grok 4 aced the test without using any tools, achieving a higher score than Google’s Gemini 2.5 Pro (21.6%) and OpenAI’s o3 (High) model (21%).
When including tools, Grok 4 Heavy scored an impressive 44.4%, nearly two-thirds better than Gemini 2.5 Pro with tools (26.9%).
The non-profit research group ARC Prize also said that Grok had scored a new state-of-the-art 16.2% on its ARC-AGI-2 test — another demanding benchmark that includes puzzle-like problems that challenge an AI to recognize visual patterns. This score is almost double that of the closest commercial model, Claude Opus 4.
Between Grok 4 and Grok 4 Heavy, xAI also announced its most expensive AI subscription to date at $300 per month, which it’s calling SuperGrok Heavy. Subscribers will also have advance copies of Grok 4 Heavy and first access to new features. This pricing puts xAI in the top-tier offerings of OpenAI, Google, and Anthropic — but makes it the most expensive subscription among the major AI providers.
To incentivize subscriptions to SuperGrok Heavy, xAI will offer early access to some new products for SuperGrok Heavy subscribers that will be introduced over the next several months. The company announced on Wednesday that it plans to release:
- An AI coding model in August
- A multi-modal agent in September
- A video generation model in October
xAI is releasing Grok 4 as an API, to enable other developers to create applications on top of the model. The company said its enterprise division is only two months old, but it plans to partner with hyperscalers to supply Grok from major cloud platforms.
However, despite how well Grok benchmarks, it’s unclear how well xAI will be able to recover from its latest fumbles as it tries to pitch Grok as a serious enterprise competitor to ChatGPT, Claude, and Gemini. Whether businesses will be quite that blithe about adopting Grok, along with all its warts, is anyone’s guess.
