Elon Musk’s xAI Unveils Grok 3: A Bold Leap in AI Development

Late Monday night, Elon Musk’s AI company, xAI, officially launched its latest flagship AI model, Grok 3, introducing new features to the Grok app on iOS and the web.

Grok is xAI’s answer to OpenAI’s GPT-4o and Google’s Gemini, designed to analyze images, answer complex queries, and support multiple functionalities on Musk’s social platform, X (formerly Twitter). Grok 3 had been in development for several months and was initially slated for release in 2024 but faced delays.

Grok 3: An Ambitious Milestone

xAI trained Grok 3 in a massive data center in Memphis, housing approximately 200,000 GPUs. According to Musk, Grok 3 boasts 10 times the computational power of its predecessor, Grok 2, and has been trained on an expanded dataset, including legal case documents.

“The capabilities of Grok 3 surpass Grok 2 by an order of magnitude,” Musk stated during a live demonstration on Monday. “It is an AI that prioritizes truth, even when that truth conflicts with political correctness.”

Unlike a single monolithic model, Grok 3 is a family of AI models. One variant, Grok 3 Mini, provides faster response times at the expense of some accuracy. While not all versions have been released, xAI has begun rolling out Grok 3 incrementally.

Grok 3 Outperforms GPT-4o in Benchmarks

xAI claims that Grok 3 has surpassed OpenAI’s GPT-4o in several benchmark tests, including AIME (for mathematical problem-solving) and GPQA (which evaluates PhD-level physics, biology, and chemistry). Early versions of Grok 3 have also performed well in Chatbot Arena, a competitive AI ranking platform where users vote for the best AI-generated responses.

Grok 3 Reasoning: Advanced Logical Thinking

Grok 3 introduces two specialized reasoning models: Grok 3 Reasoning and Grok 3 Mini Reasoning. These models focus on rigorous logical analysis, akin to OpenAI’s o3-mini and DeepSeek’s R1 from China. Unlike conventional models, reasoning models conduct self-fact-checking before generating responses, reducing errors commonly found in AI outputs.

xAI states that Grok 3 Reasoning has outperformed OpenAI’s highest-tier o3-mini-high model in multiple tests, including the latest AIME 2025 mathematics benchmark.

The reasoning models are accessible via the Grok app, where users can enable “Big Brain” mode for complex queries, leveraging additional computational resources. xAI positions these models as particularly well-suited for mathematics, science, and programming.

To protect intellectual property, some of Grok’s reasoning processes are hidden from users to prevent knowledge extraction through distillation techniques—a method AI developers use to extract insights from competing models. Recently, China’s DeepSeek was accused of training its AI by distilling OpenAI’s models.

DeepSearch: xAI’s Answer to OpenAI’s Deep Research

In addition to its reasoning capabilities, Grok 3 debuts DeepSearch, an advanced AI research tool designed to compete with OpenAI’s Deep Research. DeepSearch can scan the internet and X platform, analyze information, and generate comprehensive summaries.

Grok 3 Subscription and Pricing

Users subscribed to X Premium+ receive priority access to Grok 3, while additional premium features require a SuperGrok subscription.

SuperGrok Pricing: $30/month or $300/year
SuperGrok Benefits: More reasoning queries, DeepSearch access, and unlimited image generation

Upcoming Features and Open-Source Plans

Voice Mode: A speech-based interaction feature will launch within a week.
Enterprise API: In the coming weeks, xAI will integrate Grok 3 into its enterprise API, including DeepSearch functionality.
Open-Sourcing Grok 2: Within a few months, xAI plans to open-source Grok 2 once Grok 3 becomes stable.

“Our overarching strategy is to open-source the previous Grok version once the new version is fully deployed,” Musk explained. “As soon as Grok 3 reaches maturity, we will release Grok 2 as open-source.”

Grok’s Unique Positioning: Challenging Political Correctness

When Musk first introduced Grok nearly two years ago, he positioned it as an unfiltered, politically incorrect AI willing to tackle controversial topics that other AI models shy away from.

In some aspects, Grok has lived up to this claim. For instance, previous versions of Grok had no restrictions on using profanity, unlike ChatGPT, which imposes stricter content moderation.

However, earlier versions of Grok still exhibited political biases. A study found that Grok’s responses leaned left on issues such as transgender rights, diversity initiatives, and social inequality. Musk attributed this bias to the model’s training data, which primarily consists of publicly available web content, and vowed to make Grok “more politically neutral.” Whether Grok 3 achieves this remains to be seen.

Conclusion: The Significance of Grok 3

The launch of Grok 3 represents a significant milestone for xAI in the AI race, with key advancements including:

10x computational power compared to Grok 2
New reasoning models optimized for STEM applications
DeepSearch AI research tool
Faster response times via Grok 3 Mini
Upcoming support for voice interaction and an enterprise API
Planned open sourcing of Grok 2

With Grok 3, Musk’s AI ecosystem is rapidly evolving, posing a challenge to AI leaders like OpenAI and Google. Whether Grok will truly deliver on its promise of absolute truth-seeking AI and carve out a place in the competitive AI landscape remains an open question.

Join us at NEPCON China 2025 and seize this unique opportunity to connect with global industry leaders, explore cutting-edge technologies, and expand your business into Southeast Asia's thriving markets！

https://ali2.infosalons.com.cn/reg/NEPCON25SH/registeren/login?type=T07LFW

If interested in exhibiting at NEPCON China, please contact

Ms. Julia Gu

Tel: +86 21-2231-7010

Email: [email protected]

If interested in visiting NEPCON China, please contact

Mr. Walden Li

Email: [email protected], [email protected]

Mobile/WhatsApp/WeChat: +86 136-5125-1335

For TAP service, please contact:

Mr. Alan Li

Email: [email protected]