Grok 3: The next-gen ‘truth-seeking’ AI model

---

xAI Unveils Grok 3: A Powerful AI Model with Advanced Reasoning and Image Analysis

xAI introduced its latest artificial intelligence model, Grok 3, on Monday, bringing enhanced capabilities such as image analysis and improved question-answering.

The company developed Grok 3 using a massive data center housing approximately 200,000 GPUs. According to xAI’s founder, Elon Musk, this initiative utilized ten times the computing power of its predecessor, Grok 2, incorporating a significantly expanded dataset—including legal case filings.

Musk described Grok 3 as a "maximally truth-seeking AI," emphasizing that it prioritizes accuracy over political correctness.

A Family of Grok 3 Models

The Grok 3 release includes multiple versions tailored to different needs. For instance, Grok 3 Mini offers faster response times, while the reasoning-focused models—Grok 3 Reasoning and Grok 3 Mini Reasoning—aim to simulate human-like thought processes. These models actively analyze problems and verify their answers, reducing errors. Similar to OpenAI’s o3-mini and DeepSeek’s R1, Grok 3 Reasoning models attempt to fact-check themselves, improving reliability.

Grok 3 vs. the Competition: Benchmark Results

xAI claims Grok 3 outperforms OpenAI’s GPT-4o in certain key benchmarks, including AIME and GPQA, which assess problem-solving across subjects like mathematics, physics, biology, and chemistry.

Early rankings place Grok 3 at the top of Chatbot Arena, a user-driven AI evaluation platform, making it the first model to surpass a score of 1400.

The company also asserts that Grok 3 Reasoning outperforms competing models from Google, DeepSeek, and OpenAI in a range of reasoning benchmarks.

These advanced reasoning models are now available in the Grok app, where users can activate features like "Think" or the high-powered "Big Brain" mode for complex problem-solving. xAI has positioned them as particularly useful for STEM fields, including coding, mathematics, and scientific research.

Protecting Intellectual Property from AI Distillation

One of the more controversial aspects of AI development is "distillation," where rival companies extract knowledge from proprietary models. To counter this, xAI has deliberately obscured certain internal processes within Grok 3 to safeguard its technology.

The issue of distillation gained attention recently when Chinese AI company DeepSeek was accused of leveraging OpenAI’s models to build its own system, R-1.

Introducing DeepSearch: AI-Powered Internet Analysis

A major new feature accompanying Grok 3 is DeepSearch. This tool scans the internet—including Musk’s social platform, X—to compile and synthesize information for users. By leveraging Grok models, DeepSearch aims to provide more comprehensive and accurate responses.

Access and Subscription Plans

Grok 3 is currently accessible through X’s Premium+ subscription, priced at $50 per month (~£41), offering priority access to the latest AI features.

Additionally, xAI is launching a SuperGrok subscription at $30 per month or $300 per year. This tier provides enhanced reasoning capabilities, increased DeepSearch queries, and unlimited image generation.

In the coming weeks, xAI plans to roll out a voice interaction feature similar to Gemini Live, allowing users to engage with Grok through synthesized speech. Musk also announced that Grok 3 models will soon be available via an enterprise-ready API, including DeepSearch functionality.

Open-Source Plans for Grok 2

While Grok 3 remains proprietary for now, xAI intends to open-source its predecessor, Grok 2, once the latest model reaches maturity. Musk stated that this could happen within a few months, reinforcing xAI’s commitment to transparency and open development.

A Controversial “Truth-Seeking” AI

From its inception, Grok has been positioned as an unfiltered AI model, willing to tackle politically sensitive topics. Musk has previously described it as “anti-woke,” setting it apart from more cautious competitors.

However, earlier versions of Grok sometimes exhibited political biases, prompting xAI to refine its approach. Musk now claims efforts are underway to make Grok more politically neutral. Whether Grok 3 succeeds in striking this balance remains to be seen.

The Future of Grok 3

As AI competition intensifies, xAI’s latest advancements push the boundaries of reasoning and information synthesis. However, questions remain about bias, transparency, and the broader ethical implications of deploying such technology.

With OpenAI, Google, and DeepSeek continually refining their models, Grok 3’s success will depend on its ability to deliver accurate, fast, and reliable responses while navigating the challenges of public perception and responsible AI use.

Would you like any additional refinements?