Elon Musk’s AI firm, xAI, late on Monday launched its newest flagship AI mannequin, Grok 3, and unveiled new capabilities for the Grok iOS and internet apps.
Grok, xAI’s reply to fashions like OpenAI’s GPT-4o and Google’s Gemini, can analyze pictures and reply to questions, and powers plenty of options on Musk’s social community, X. Grok 3, which has been in improvement for a number of months, was optimistically slated for launch in 2024, however missed that deadline.
Monday’s is an formidable launch.
xAI has been utilizing an unlimited information middle in Memphis containing round 200,000 GPUs to coach Grok 3. In a publish on X, Musk claimed that Grok 3 was developed with “10x” extra computing energy than its predecessor, Grok 2, utilizing an expanded coaching information set that ostensibly consists of filings from court docket circumstances.
“Grok 3 is an order of magnitude extra succesful than Grok 2,” Musk stated throughout a live-streamed presentation on Monday. “[It’s a] maximally truth-seeking AI, even when that reality is typically at odds with what’s politically right.”
Grok 3 is a household of fashions, to be exact. A smaller model of Grok 3, Grok 3 mini, responds to questions extra shortly at the price of some accuracy. Not all of the fashions and associated options of Grok 3 can be found but (some are in beta), however they started rolling out on Monday.
xAI claims Grok 3 beats GPT-4o on benchmarks together with AIME (which evaluates a mannequin’s efficiency on a sampling of math questions) and GPQA (which assesses fashions utilizing PhD-level physics, biology, and chemistry issues). An early model of Grok 3 additionally scored competitively in Chatbot Area, a crowdsourced check that pits totally different AI fashions in opposition to one another and has customers vote on their most popular responses, in keeping with xAI.
Two fashions within the new Grok 3 household, Grok 3 Reasoning and Grok 3 mini Reasoning, can fastidiously “suppose via” issues, much like “reasoning” fashions like OpenAI’s o3-mini and Chinese language AI firm DeepSeek’s R1. Reasoning fashions attempt to fact-check themselves earlier than giving out outcomes, which helps them keep away from among the pitfalls that usually journey up fashions.
xAI claims that Grok 3 Reasoning surpasses one of the best model of o3-mini — o3-mini-high — on a number of fashionable benchmarks, together with a more moderen arithmetic benchmark referred to as AIME 2025.
These reasoning fashions might be accessed by way of the Grok app. Customers can ask Grok 3 to “Suppose,” or — for harder queries — leverage “Huge Mind” mode for reasoning that employs extra computing. xAI describes the reasoning fashions as greatest suited to arithmetic, science, and programming questions.
Musk stated among the reasoning fashions’ “ideas” are obscured within the Grok app to stop distillation, a technique utilized by AI mannequin builders to extract data from different fashions. Just lately, DeepSeek was accused of distilling OpenAI’s fashions to create its personal.
Grok’s reasoning fashions underpin a brand new characteristic within the Grok app referred to as DeepSearch, xAI’s reply to AI-powered analysis instruments like OpenAI’s deep analysis. DeepSearch scans the web and X to investigate info and ship an summary in response to a query.
Subscribers to X’s Premium+ tier ($50 per thirty days) will get entry to Grok 3 first, and different options might be gated behind a brand new plan that xAI’s calling SuperGrok. Priced at $30 per thirty days or $300 per yr (if leaks are to be believed), SuperGrok unlocks extra reasoning and DeepSearch queries, and throws in limitless picture technology.
Sooner or later — as quickly as a few week from now — the Grok app will achieve a “voice mode,” Musk stated, which is able to give Grok fashions a synthesized voice. A couple of weeks after that, Grok 3 fashions might be accessible by way of xAI’s enterprise API, together with the DeepSearch functionality.
xAI plans to open-source Grok 2 within the coming months, Musk stated.
“Our normal strategy is that we are going to open-source the final model [of Grok] when the subsequent model is totally out,” he continued. “When Grok 3 is mature and steady, which might be inside a couple of months, then we’ll open-source Grok 2.”
When Musk introduced Grok roughly two years in the past, he pitched the AI mannequin as edgy, unfiltered, and anti-“woke” — on the whole, keen to reply controversial questions different AI programs gained’t. He delivered on a few of that promise. Instructed to be vulgar, for instance, Grok and Grok 2 would fortunately oblige, spewing colourful language you possible wouldn’t hear from ChatGPT.
However Grok fashions previous to Grok 3 hedged on political topics and wouldn’t cross sure boundaries. In truth, one examine discovered that Grok leaned to the political left on subjects like transgender rights, range applications, and inequality.
Musk has blamed the conduct on Grok’s coaching information — public internet pages — and pledged to “shift Grok nearer to politically impartial.” It’s not but clear whether or not xAI has achieved that aim, and what the results is likely to be.
