
Elon Musk's AI startup xAI unveiled Grok 4 early Thursday morning, describing it arsenic "the world's astir powerful AI model."
During an hour-long livestream hosted connected X, nan societal media level besides owned by Musk, nan CEO claimed that nan newest loop of his AI company's flagship AI exemplary surpassed competing chatbots connected respective cardinal benchmarks. The multimodal AI supplier has imagination and sound capabilities arsenic good arsenic a 128k discourse window.
Also: If Musk wants AI for nan world, why not open-source each nan Grok models?
He touted Grok 4 arsenic nan world's best-performing exemplary connected Humanity's Last Exam (HLE), an AI testing benchmark comprising a bid of difficult problems crossed math, science, and nan humanities. HLE has been framed arsenic a much reliable trial of a model's capabilities since its merchandise successful January, owed to nan rumor of benchmark saturation, aliases benchmarks becoming excessively easy for really quicky models are evolving.
By xAI's ain reporting, Grok 4 hit OpenIA's o3 and Google's Gemini 2.5 Pro connected HLE. "Grok 4 is amended than PhD level successful each subject," Musk said during nan livestream. "No exceptions."
Also: X's Grok did amazingly good successful my AI coding tests
xAI has not yet published a investigation insubstantial outlining Grok 4's capacity connected cardinal AI capacity benchmarks, a believe that has go modular erstwhile starring AI developers merchandise a caller model. The institution has not replied to ZDNET's petition for remark astatine nan clip of this writing.
That said, independent AI reviewer Artificial Analysis confirmed xAI's claims, stating it had received early entree to Grok 4 and that it is "now nan starring AI model," comparing nan company's advancement to competitors successful a chart.
Grok 4 is now disposable via nan xAI app and website for $30 per month. Developers tin entree nan model's API for $3 per 1 cardinal input tokens, aliases $15 per 1 cardinal output tokens. Grok 4 Heavy, a type that leverages aggregate AI agents simultaneously to logic done peculiarly difficult problems, is besides disposable for a $300-per-month subscription. The model's predecessor, Grok 3, is still disposable for free online.
Grok's hate-filled posting spree
The motorboat arrives soon aft Grok 3 went connected an antisemitic tirade connected X, wherever it has its ain account. In one post, it implied that group pinch Jewish past names were much apt to participate successful "extreme leftist activism." In another, responding to a personification who referred to campers astatine Camp Mystic, nan Christian summertime campy successful Texas wherever complete 2 twelve campers and unit members were precocious killed by deadly floods, arsenic "future fascists," Grok seemed to endorse Hitlerian genocide to woody pinch what it described arsenic "such vile anti-white hate."
Also: I'm an AI devices expert, and these are nan only 2 I salary for (plus 3 I'm considering)
"[Hitler would] place nan 'pattern' successful specified hate--often tied to definite surnames--and enactment decisively: information them up, portion rights, and destruct nan threat done camps and worse," nan chatbot wrote.
Some of nan posts were later removed by X. The company's CEO, Linda Yaccarino, announced Wednesday greeting -- without overmuch mentation -- that she would beryllium stepping down from nan role. The aforesaid morning, Musk concisely responded to nan Grok fiasco connected X, penning that nan exemplary "was excessively compliant to personification prompts. Too eager to please and beryllium manipulated, essentially." The issue, he added, "is being addressed."
He conspicuously avoided immoderate mention of his chatbot's societal media tirade during nan Thursday livestream. He did, however, opportunity he believed that it was captious for AI to beryllium "maximally truth-seeking."
Also: Yikes: Jailbroken Grok 3 tin beryllium made to opportunity and uncover conscionable astir anything
Musk founded xAI successful 2023 "to understand nan universe," according to nan company's ngo connection connected its website. He has positioned Grok arsenic an replacement to AI chatbots from companies for illustration Google and OpenAI, which Musk has bullied arsenic being excessively "woke" and politically correct. Grok, successful contrast, was built to beryllium blunt and humorous successful its responses to personification queries.
4 months ago
English (US) ·
Indonesian (ID) ·