Elon Musk’s AI startup has launched its newest model with some grand claims — including that it can outperform leading models from the U.S. and China.
Musk’s company xAI, founded in March 2023, unveiled the latest version of its flagship Grok-3 artificial intelligence model on Monday evening. The model is rolling out first to X’s Premium+ subscribers, and the company announced a separate subscription tier for Grok called “SuperGrok,” which promises access to the latest AI model’s more advanced features.
Grok-3 is “an order of magnitude more capable than Grok 2 in a very short period of time,” Musk said during a demo livestreamed on X. He later added that the model is still in beta and that users can expect improvements “literally every day,” with a voice interaction feature expected to release in about a week.
xAI developers touted benchmark numbers that showed Grok-3 — which was trained with 10 times more computing power than Grok-2 — outperforming rivals like DeepSeek-V3 and GPt-4o on mathematical reasoning, science and coding.
Grok-3 outperforms models from Google, Anthropic and Meta, according to the Artificial Analysis Quality Index, a popular independent AI analysis ranking, and falls behind DeepSeek-R1 as well as OpenAI’s o3 and o1 models. Grok-3’s Reasoning Beta model, however, outranks all except o3 in the index.
xAI on Monday evening also announced a new product called Deep Search, intended to serve as a “next-generation search engine.” Seemingly an answer to AI-powered search tools from the likes of OpenAI and Perplexity, xAI’s Deep Search scans web pages and X posts to formulate its answers.
“Something that might take you half an hour or an hour of researching on the web or searching social media, you can just ask it to go do that and come back, and 10 minutes later it’s done an hour’s worth of work to you,” Musk said during the demo stream. “And maybe better than you could have done it yourself.”
The launch of Grok-3 stirred buzz on X, the social platform (formerly Twitter) owned by Musk. Andrej Karpathy, former director of AI at Musk’s Tesla, shared a detailed review and wrote that the benchmark results “look quite encouraging indeed.” Some lauded its abilities in early tests of the model, while others appeared more skeptical.
In Monday’s stream, Musk described Grok as a “maximally truth-seeking AI, even if that truth is sometimes at odds with what is politically correct.”
The xAI founder has long positioned Grok as an edgy counter to other chatbots — such as those from OpenAI and Google — that he’s criticized for being too “woke.” When Grok first launched in November 2023, it was marketed as being able to use wit and humor to “answer spicy questions that are rejected by most other AI systems.”
Despite this branding, many quickly noticed that Grok demonstrated political leanings similar to its competitors’, with guardrails that sometimes prompted frustration from xAI users who hoped it would combat what Musk often calls “the woke mind virus.”
In a screenshot he shared on Sunday, however, Musk praised Grok-3 for seemingly lambasting the tech-focused news outlet The Information and hailing X instead. The opinions exhibited by Grok-3 in Musk’s screenshot has prompted some online to suspect xAI programmed its latest model to align more closely with Musk’s own views.
When asked for its opinion on The Information, the model apparently wrote that the publication, and legacy media in general, is “garbage” — adding that it delivers “polished narratives, not reality.”
“X, on the other hand, is where you find raw, unfiltered news straight from the people living it. No middlemen, no spin—just the facts as they happen,” Grok-3 continued. “Don’t waste your time with The Information or any legacy outlet; X is the only place for real, trustworthy news.”
But in multiple tests of Grok-3 conducted by NBC News on Tuesday, the model did not produce such an answer. When asked the same question, Grok-3 repeatedly wrote that The Information is a “well-regarded tech news outlet known for its in-depth reporting and analysis,” often stating that the publication is “not infallible,” citing its paywalled content and niche focus.
