Elon Musk's xAI Has Made Grok 3 Official. But Not Everybody Will Be Able To Access It Right Now

According to xAI, Grok 3 surpasses GPT-4o in multiple benchmarking tests and also outperforms Claude 3.5 Sonnet, DeepSeek-V3, and Gemini-2 Pro in internal testing.

Musk and the xAI team announced Grok 3 via a livestream on X.

Source : X/@omarsar0

Elon Musk’s artificial intelligence company, xAI, has introduced the latest Grok 3 family of AI models, enhancing their capabilities with advanced reasoning features and new tools like DeepSearch and voice mode. The announcement, made during a live-streamed event hosted by xAI engineers and Musk himself, marks a significant upgrade from the previous Grok 2 models. If you are already excited, note that you won't be able to access Grok 3 right away, unless you are a premium user of X. There's another subscription paywall on top of that as well.

https://t.co/HTK4u2aYRu

— Elon Musk (@elonmusk) February 18, 2025

Grok 3: A Leap In AI Reasoning

Musk described the Grok 3 models as “an order of magnitude more capable than Grok 2,” emphasising xAI’s rapid development efforts. He revealed that the company’s newly built data centre had its first 100,000 GPUs operational in just 122 days, with capacity doubling in another 92 days.

The Grok 3 series features multiple large language models (LLMs) tailored for different applications. Among them are Grok 3, a standard version, and Grok 3 mini, a smaller yet faster model. Additionally, xAI introduced Grok 3 Reasoning and Grok 3 mini Reasoning, both designed for enhanced reasoning tasks using test-time compute-based methods. While the rollout for these models has started, some versions remain in the beta phase.

To prevent knowledge extraction from its reasoning models, Musk noted that some chain-of-thought (CoT) steps would be concealed, mitigating the risk of distillation — where synthetic data from one AI model is used to train another, often smaller, model. This is a noted difference when compared to DeepSeek, which shows detailed CoT to queries. 

Performance Benchmarks & Competitive Edge

According to xAI, Grok 3 surpasses GPT-4o in multiple benchmarking tests, including the American Invitational Mathematics Examination (AIME), Graduate-Level Google-Proof Q&A (GPQA), and LiveCodeBench.

Grok 3 appears to be a state-of-the-art frontier model. This is a huge accomplishment, especially considering how late in the game they started.

Congrats @ibab, @elonmusk, and the rest of the @xai team. Can’t wait to start building on it. pic.twitter.com/fYMkwKYmDD

— Amjad Masad (@amasad) February 18, 2025

The model also reportedly outperforms Claude 3.5 Sonnet, DeepSeek-V3, and Gemini-2 Pro in internal testing. Additionally, the Grok 3 Reasoning models claim to outperform OpenAI’s o3 mini model.

New Features: DeepSearch & Voice Mode

Two major features accompany the Grok 3 launch. DeepSearch, xAI’s response to recent advancements by OpenAI and Google, enables deeper internet and X platform analysis to generate comprehensive reports on complex queries.

Meanwhile, the voice mode will allow the AI model to provide verbal responses, although this feature will roll out at a later date. Musk hinted that it could be available “as soon as a week from now.”

New SuperGrok Subscription Tier: How Much Does It Cost? How To Access?

Alongside the AI model launch, xAI introduced a premium subscription tier named SuperGrok. While pricing details remain undisclosed, this tier offers users access to exclusive features, including DeepSearch, an advanced Think (reasoning) mode, and higher image generation limits. SuperGrok members will also receive early access to upcoming tools.

Users subscribed to X Premium Plus will gain access to Grok 3, but other advanced features will be restricted to SuperGrok subscribers.

Musk also reaffirmed xAI’s commitment to open-sourcing previous AI models. Once Grok 3 reaches full stability, which is expected in the coming months, xAI plans to release Grok 2 as an open-source model.

With the Grok 3 series, xAI continues to push the boundaries of AI development, bringing sophisticated reasoning capabilities and cutting-edge features to its growing user base.