Musk launches Grok 4, and a $300/month subscription plan
What's the story
Elon Musk's xAI has unveiled its latest set of artificial intelligence (AI) models, Grok 4 and Grok 4 Heavy. The latter is described as the company's "multi-agent version" designed for enhanced performance According to Musk, Grok 4 Heavy deploys several agents to tackle a problem in parallel, then compares their outputs "like a study group" to determine the best solution. The launch was announced along with a new subscription plan called SuperGrok Heavy.
Enhanced features
Grok 4 trained on xAI's Colossus supercomputer
Grok 4 has been trained on xAI's Colossus supercomputer for advanced, scientist-grade reasoning. The model promises improved logical reasoning and text generation. It also comes with Grok 4 Code, a specialized variant designed to write, debug, and explain code more efficiently. This feature is similar to tools like GitHub Copilot or OpenAI's GPT-4 Code Interpreter. Musk says you can just copy and paste your entire source code into the query box, and Grok 4 will handle the rest.
Twitter Post
Take a look at Musk's post
You can cut & paste your entire source code file into the query entry box on https://t.co/EqiIFyHFlo and @Grok 4 will fix it for you!
— Elon Musk (@elonmusk) July 10, 2025
This is what everyone @xAI does. Works better than Cursor.
Benchmark results
Beating Gemini on humanity's last exam
Grok 4 has shown impressive performance on several benchmarks, including Humanity's Last Exam. The test measures an AI's ability to answer thousands of crowdsourced questions across different subjects. Without "tools," Grok 4 scored 25.4% on the exam, beating Google's Gemini 2.5 Pro (21.6%) and OpenAI's o3 (high), which scored 21%. With "tools," Grok 4 Heavy managed a score of 44.4%, outperforming Gemini with tools, which scored significantly lower than Grok 4 Heavy's score at 26.9%.
Test achievement
Setting new record on Arc Prize benchmark
The non-profit Arc Prize has reported that Grok 4 has achieved a new state-of-the-art score on its ARC-AGI-2 test. This difficult benchmark consists of puzzle-like problems where an AI has to identify visual patterns. Grok 4 scored an impressive 16.2%, nearly double the score of the next best commercial AI model, Claude Opus 4 by Anthropic.
Twitter Post
Core announcements and benchmark scores
Insane that Elon Musk has pulled it off again, absolutely crushing the AI wars with Grok 4.
— Deedy (@deedydas) July 10, 2025
Summarizing the core announcements:
— Post-training RL spend == pretraining spend
— $3/M input told, $15/M output toks, 256k context, price 2x beyond 128k
— #1 on Humanity’s Last Exam… pic.twitter.com/eHfz4XjGu7
Multimodal support
Competing for multimodal capabilities
Grok 4 can handle not just text but also images and possibly videos. Musk said this was one of their biggest weaknesses, and better multimodal capabilities would bring the model closer to competition with OpenAI's GPT-5 and Gemini 2.5 Pro. The model also comes with Grok 4 Voice, which offers a natural, human-like voice with fewer interruptions.
Web integration
DeepSearch pulls live data from the web
Like its predecessors, Grok 4 comes with DeepSearch, a tool that pulls live data from the web. This feature especially focuses on X to provide up-to-date results during chats. One of the biggest differentiators for Grok has been its understanding of internet culture. The new model is being tuned to interpret memes, slang, and humor with high accuracy.
Market positioning
Launch amid concerns over content moderation
The launch of Grok 4 comes amid a backlash over racist responses from earlier versions, raising questions about xAI's content moderation. Despite this, Musk remains confident in his AI model's capabilities. "We've run out of test questions to ask," he said during the launch. The release also puts Musk directly against OpenAI's GPT-5 and Anthropic's Claude 4 Opus in the AI race.
A new plan
SuperGrok Heavy: A new subscription plan
Along with Grok 4 and Grok 4 Heavy, xAI has also launched its most expensive AI subscription plan yet, SuperGrok Heavy. The plan costs $300 per month and gives subscribers early access to Grok 4 Heavy as well as new features. It is similar to ultra-premium tiers offered by OpenAI, Google, and Anthropic, but now holds the title of the most expensive subscription among major AI providers.