Elon Musk’s company, xAI, has introduced Grok-3, their latest flagship AI model. Initial tests suggest that Grok-3 could perform better than competitors like OpenAI and DeepSeek from China in areas like math, science, and coding.
During a live demonstration on X, Musk shared, “We’re very excited to present Grok 3, which is, we think, an order of magnitude more capable than Grok 2 in a very short period of time.”
New Features and Capabilities
Alongside Grok-3, xAI announced “Deep Search,” described as a “next generation search engine.” Grok-3, which was delayed beyond its intended 2024 release, has been developed using a vast data center in Memphis, equipped with approximately 200,000 GPUs. Musk noted that Grok-3 was trained with “10x” more computing power than its predecessor and an expanded dataset, including legal documents.
Performance and Model Family
Musk described Grok-3 as, “[It’s a] maximally truth-seeking AI, even if that truth is sometimes at odds with what is politically correct.” Grok-3 isn’t a single model but a suite, including Grok-3 mini, which prioritizes speed over accuracy. The rollout of these models began on Monday, though not all features are fully available, with some still in beta.
Grok-3 reportedly outperforms OpenAI’s GPT-4o in several benchmarks, including AIME for math and GPQA for complex science problems. In crowdsourced testing via Chatbot Arena, an early version of Grok-3 performed competitively.
Reasoning and Specialized Models
The Grok-3 family includes specialized reasoning models like Grok-3 Reasoning and Grok-3 mini Reasoning, which aim to “think through” problems, similar to other advanced models.
These models fact-check themselves, aiming for higher accuracy. Musk mentioned that some thought processes are not displayed to prevent other AIs from learning directly from Grok-3’s methods.
Integration and Future Plans
The reasoning capabilities feed into a new feature called DeepSearch, which offers research assistance by scanning the internet and X. Within a week, Grok-3 will introduce a “voice mode” for more interactive communication. A few weeks later, enterprise users can access Grok-3 and DeepSearch via xAI’s API.
Musk also revealed plans to open-source Grok-2 once Grok-3 stabilizes, adhering to xAI’s strategy of releasing previous versions as new ones are fully deployed.
Availability and Pricing
Grok-3 will be available to premium subscribers of X starting Tuesday in the U.S., with access also via a separate subscription for its web and app platforms. X’s Premium+ tier subscribers ($50/month) get first access.
A new plan, SuperGrok, expected to cost $30 per month or $300 annually, will unlock advanced features like additional reasoning queries and unlimited image generation.