Elon Musk’s xAI introduced its newest synthetic intelligence flagship mannequin, Grok 3, on Monday. The corporate additionally introduced Grok 3 mini, a scaled-back model, plus the addition of DeepSearch, a brand new software that the corporate calls a next-generation search engine. xAI has added new performance for Grok 3 net and cell apps and a subscription service particularly for Grok customers, dubbed SuperGrok.
“We’re very excited to current Grok 3, which is, we predict, an order of magnitude extra succesful than Grok 2 in a really quick time frame,” Musk mentioned throughout xAI’s livestream on X. Grok 3 was skilled on 200,000 Nvidia H100 GPUs, which is double that of Grok 2. The staff mentioned it took 92 days to increase its Memphis-based supercomputer, dubbed Colossus, to accommodate coaching for Grok 3.
Musk mentioned throughout the presentation that Grok 3 boasts 15 occasions extra laptop energy than Grok 2, although in a earlier X submit, he mentioned 10 occasions extra. The mannequin was skilled on data ranging from consumer posts on X to court docket paperwork.
Grok 3 faces stiff competitors from OpenAI, Google and Anthropic, all of which have launched new AI fashions already in 2025 or are planning to take action.
Google Gemini 2.0 added helpful performance earlier in February, whereas OpenAI plans to unify all its AI fashions when GPT-5 launches someday later in 2025. In the meantime, Anthropic’s subsequent new AI mannequin could possibly be weeks away.
How one can strive Grok 3
Grok 3 is rolling out to X Premium Plus members beginning on Feb. 18, and they’ll have unique entry for now, together with entry to DeepSearch. It is a small win for these customers, as X not too long ago elevated the worth of Premium Plus from $16 to $22.
Finally, xAI will launch SuperGrok, a subscription service particularly for Grok 3 that can embrace DeepSearch, greater picture technology limits and entry to Grok 3 mini options like Suppose. Costs weren’t proven throughout the presentation.
Benchmark assessments evaluating Grok 3 and Grok 3 mini to the competitors.
‘Scary good’
“Grok 3 has very highly effective reasoning capabilities,” Musk mentioned throughout a Feb. 13 interview with CNBC. “Within the assessments that we have executed up to now, Grok 3 is outperforming something that is been launched that we’re conscious of. At occasions, I feel Grok 3 is type of scary good.”
In fact, there isn’t any solution to check Grok 3’s capabilities but. And the staff talked about that Grok could reply with what it believes is the reality — there was in any other case no point out of Grok 2’s penchant for outputting hallucinations or how typically Grok 3 will achieve this.
Grok 3 is proven conducting a DeepSearch about SpaceX’s subsequent Starship launch.
DeepSearch
Along with Grok 3, the xAI staff introduced DeepSearch, described as the primary technology of Grok 3 brokers that enable customers to ask questions and obtain solutions. It is referred to within the livestream as a “next-generation search engine.”
OpenAI and Google have related agent-based searches. Each are known as Deep Analysis.
DeepSearch exhibits customers the person steps Grok 3 goes by, from desirous about the query to analysis after which lastly the reply. The demo took round a minute and included 15 X posts and 32 net pages as references.
One other function of DeepSearch is the flexibility to view Grok 3’s reasoning. After asking one query about March Insanity, the staff used the function to point out how Grok 3 got here to its conclusions.
“In all probability, I ought to have a look at staff rankings, their efficiency throughout the common season, any accidents or key participant statuses, and possibly some historic knowledge on how they carry out in tournaments,” Grok 3 revealed when requested.
This follows Chinese language AI firm DeepSeek — which launched its platforms in January — together with the reasoning course of whereas answering queries.
Grok 3 conducting a DeepSearch about SpaceX’s subsequent Starship launch.
Grok vs. rivals
Grok shouldn’t be alone, and one professional instructed CNET it could have some catching as much as do.
“Whereas high-ranking benchmark outcomes are spectacular, it says extra concerning the normal motion in direction of fashions with reasoning capabilities — or agentic AI — than it does concerning the particular person competitiveness of fashions,” mentioned Paul Schell, trade analyst for ABI Analysis. “We think about the broader go-to-market, which encompasses the advertising technique in addition to partnerships that mannequin builders have solid. Right here, Anthropic and OpenAI are forward of Grok.”
One distinction between Grok and rivals, Schell sais, is that “Grok 3’s DeepSearch readily accesses real-time data from the web, not like OpenAI’s deep analysis product.”
However he additionally notes that OpenAI’s product “seems to enter extra depth,” when looking out, taking as much as half-hour to finish a job.
