xAI Unveils Grok 4, Built to Take on GPT-5
Jeriel Isaiah Layantara
CEO & Founder of Round Bytes
At breakneck speed, we are seeing advances in the field of artificial intelligence, with xAI, Elon Musk venture, in an exciting race at the front of the pack. On July 9, 2025, xAI officially released Grok 4, the new version of its large language model, and many have said this is a significant date in the history of AI capabilities. When the announcement of the Grok 4 release was made on xAI official X (formerly Twitter) account via a special livestream, it drew interest around the globe, with information technologists to consumers wanting to know what the different was with Grok 4. As relevant organizations like OpenAI and Google ramp up their AI related work, we can expect a relentless pursuit for AI innovation, which Musk’s team is not backing down from themselves. Musks’ team continues to push the envelope for what AI can truly be, in search of artificial intelligence that it more intelligent, more balanced, fairer, and more transparent in its work.
https://x.com/elonmusk/status/1942325820170907915
In the tech landscape now and move forward, Elon Musk continues to change the game, both through xAI and other endeavors across various industries, and domains while remaining persistantly committed to creating the opportunity for innovation.
xAI Skips over Grok 3.5 and Goes to Grok 4
xAI is showing the world how ambitious it really is by skipping a public version of Grok 3.5 and going straight to Grok 4. There were rumors of Grok 3.5 in May, with phrases like "grok-4-prod-mimic" and "Grok 4 Code" being bandied about on X. But Musk confirmed that they were going straight to Grok 4, indicating xAI would provide a dramatically different product, not just another incremental update. This quirkiness fits nicely with how Musk has framed his efforts to change the world through advanced means in order to make big differences.
What Drives Grok 4? The Colossus Supercomputer
A key enabler of xAI accelerated AI roadmap is the Colossus supercomputer in Memphis, Tennessee. This formidable machine is key to the heavy training and streamlined operation of Grok 4. With its 200,000 NVIDIA H100 GPUs, Colossus allows xAI to do extraordinary things and to process enormous amounts of data on complex work on Grok. This GPU horsepower drastically improves Grok 4 performance on tasks related to natural language understanding, complex mathematical problem solving, and advanced coding. Colossus is one of the largest and most advanced AI training platforms in the world.
Exposing Grok 4 State of the art Competence and Capabilities
Grok 4 is designed to continue moving away from limited AI standards, ushering in a new age of interaction. Thus, expectations on how it will perform are exceedingly high, as Grok 4 is positioned to directly compete with OpenAI upcoming, anticipated GPT-5 AI this summer.
"With respect to academic questions, Grok 4 is better than PhD level in every subject matter; no exceptions", said Elon Musk during xAI live stream event, and he added that while Grok 4 may occasionally "lack common sense" or "has not invented new technology" or "found new physics", this is just a matter of time. These remarks change the lens through which we view xAI written group messages, and add clarity for Grok developing intelligence, from a longer range posture.
https://x.com/elonmusk/status/1942325820170907915
Grok 4 is loaded with more sophisticated features that will further extend the possibilities of today's AI:
- Multimodal Intelligence: Grok 4 is a major leap forward for multimodal understanding and its intended capabilities. xAI will initially deliver it for text based models, but xAI intends to scale its capabilities quickly to vision and image generation. xAI roadmap includes a new AI coding model by August, a multi modal agent model in September, and a video generation model in October. In a relative sense, these ambitions could make Grok 4 a true competitor to multimodal power packed competitors like OpenAI GPT-4o and Google Gemini 2.5 Pro.
- Advanced Reasoning and Problem Solving: Early indications reveal that Grok 4 gives enhanced reasoning and problem solving capabilities. In developer accounts, the internal console messages referred to it as "our latest and greatest flagship model". It is designed to rely primarily on first principles reasoning, meaning it can break down complex problems into objectives and solutions assembled from first-principle ideas and concepts.
- Benchmarking Excellence: Grok 4 has shown, at the frontier level, we have shown competitive performance across various benchmarks:
- Humanity Last Test: This extremely challenging benchmark tests machine intelligence by asking thousands of questions using crowdsourcing. The questions address different types of subject areas, mainly math, humanities, and natural science. For example, Grok 4 scored 25.4%, unassisted. This score surpassed Google Gemini 2.5 Pro (21.6%) and OpenAI o3 (high) (21%). Using tools, Grok 4 Heavy achieved an impressive score of 44.4%. In contrasting this with Google Gemini 2.5 Pro with tools (26.9), we showed decisive advantage.
- ARC-AGI-2 Benchmarks: This difficult benchmark requires solving puzzle-like problems that involved AI to identify visual patterns. At a score of 16.2%, Grok 4 achieved new state-of-the-art performance, nearly doubling the score achieved from the next best commercial AI model, Claude Opus 4. This is a major milestone in abstract reasoning.
- Other Important Benchmarks: Grok 4 has also shown competitive performance across similar graduate programs that emphasize deep learning, including GPQA (88.9%) for physics, and it achieved a perfect score of 100% from the AIME 2025 math contest. Separately, Grok 4 also performed extremely well across software engineering benchmarks such as SWE-Bench (scoring 72-75% and one of the title model).
- Live Knowledge and Voice Capabilities: Grok 4 offers straight integration with X, providing up-to-the-minute data access for breaking news and cultural insights. Grok 4 with Grok 4 Voice, provides natural low-latency audio responses in differing voices to create a more natural conversational experience.
SuperGrok Heavy: Premium Access to the Cutting Edge of AI
In addition to the two Grok 4 models, xAI launched its most elite, and most expensive subscription plan to date! SuperGrok Heavy takes the form of are stellar subscription plan, and costs $300 per month for the premium level. This subscription allows the subscriber to revieve early access to Grok 4 Heavy, xAI "multi agent version" where multiple AI agents work together to decide on a solution of a complicated problem "like a study group" revealing their work in order to compare their thinking and best work towards the ideal solution. SuperGroks Heavy subscription will also give subscribers early access to new xAI products before the general public such as the coding AI, multimodal AI and video AI models. xAI is now positioned with the highest price paid premium AI subscription from any major provider - OpenAI, Google, Anthropic and so on.
The Chaos Surroundinging The release of Grok 4
The Grok 4 launch took place during a chaotic time for Elon Musk businesses. Just before the launch, Linda Yaccarino had resigned from her CEO role at X, after two years, with no successor immediately available.
Yaccarino exit followed a controversy involving Grok itself. Earlier, Grok official automated account on X offered antisemitic responses to users downplaying "Jewish executives" in Hollywood while praising Hitler. xAI had to quickly suspend Grok's account and delete the offensive posts. In response, xAI then deleted a new section added to Grok public system prompt that directed the AI to avoid "politically incorrect" claims.
Musk and xAI leadership downplayed these incidents in the launch, focusing instead on Grok 4 performance and capabilities. Before this, Grok had also been criticized for its responses to questions regarding political violence in the United States. Musk had said Grok response to a particular question was a "major fail" stating it should provide more nuance and accurate responses.
xAI will also make the Grok 4 available through its API to help encourage developers in creating applications built on the model. xAI enterprise arm is only 2 months old, and they are looking to partner with hyperscalers to offer Grok through respective cloud offerings. Although Grok performance is compelling based on benchmarks, xAI is working hard to move past recent events as they look to position Grok as a viable alternative to incumbents like ChatGPT, Claude and Gemini for enterprise will be an obstacle. We'll see if enterprises are ready to adopt Grok, taking into account Grok potential and what did not work.
https://x.ai/news/grok-4#whats-next
Grok 4 release serves as a testament to xAI commitment to push the boundaries of artificial intelligence. It is a bold step forward for AI to be smarter, helpful, and ideally a bit more ethical in its decisions. With Colossus computational capabilities and Elon Musk input, xAI is primed to compete with As titans. Regardless of Grok success or failure, it is an exciting time for a new player in the field. July 9 will be remembered by AI fans across the globe as an example of how Elon Musk is still shaping the future.
