Summary of Key Points
The release of Grok 4.6 has enabled Musk’s AI model to join the elite ranks of top large-scale models in the United States, with capabilities on par with those of OpenAI and Anthropic’s flagship products. In terms of programming performance, it even outperforms some of them, and its API prices are more competitive. However, Grok still exhibits instability when handling complex, lengthy tasks that require continuous execution (it tends to fail towards the end). Musk has invested billions of dollars in AI, and Grok must maintain its leading position to convert this investment into revenue from various sources such as model sales, API usage, and computing power, rather than merely becoming a provider of computing resources. The competition among large-scale models has shifted from “who is smarter” to “who can complete tasks more reliably at a lower cost.” Although SpaceXAI possesses complementary resources in terms of computing power, data, and distribution, it still needs to address issues related to organizational boundaries, corporate trust, and the reliability of long-term tasks.
1. Grok 4.6: In the Elite League, but Still Struggling with Complex Projects
Grok 4.6 is now considered a “quasi-top-tier” model in the AI community:
- Strong Capabilities: It ranks only about one point behind OpenAI’s Sol and Anthropic’s Fable 5 on comprehensive intelligence benchmarks, and outperforms Sol in programming tests.
- Competitive Prices: Its API costs $2 per million characters input and $6 per million characters output, which is more than half the cost of Sol’s $5.69 per task.
- Clear Weaknesses: When dealing with complex, lengthy tasks (such as expanding a 3-minute video into a 90-minute movie or completing multi-step programming projects), Grok performs well in the initial stages and has a solid framework, but it tends to make mistakes and become unstable towards the end.
For example, Musk mentioned that he wants Grok to produce a long film based on “The Odyssey” by the end of the year. While Grok can create a production schedule and cost estimate, even Musk acknowledges that completing the 90-minute film in 208 days is challenging due to the difficulty of maintaining consistency across thousands of shots, which would result in high costs for rework.
2. Musk’s Bold Investment in AI
Musk’s investment in AI is unprecedented:
- Rapid Spending: In the first half of 2026, SpaceX’s AI division spent $23.6 billion, accounting for 86% of the company’s total expenses.
- A Bet on the Future: SpaceX estimates the AI market to be worth $26.5 trillion—13 times larger than its rocket and satellite business—and expects AI revenues to exceed the combined revenue of all other businesses by September 2026, contributing 99% of the company’s value in four to five years.
- Grok’s Critical Role: If Grok can stay at the forefront, SpaceX will be able to generate revenue from multiple sources: model sales (APIs/subscriptions), application services (such as AI features within X), and computing power (renting out GPUs). Conversely, if it falls behind, all the investment could be lost.
Therefore, Musk is pushing for rapid updates to Grok: version 4.6 was released in just 35 days, and version 4.7 (with 2.1 trillion parameters) is expected in a few weeks. This progress relies on a combination of scaling computing power (building a cluster of 100,000 GPUs in 122 days), feeding the model with real-world data (from X and Cursor’s programming activities), and rapid engineering iterations.
3. The Shift in Large-Model Competition
The focus of competition has shifted from “who is the most intelligent” to “who can complete tasks reliably at a lower cost.” While previous models were compared based on their intelligence, the current gap in capabilities among top models is only a few percentage points. The key factors now include:
- Task Completion: Customers are concerned with whether a model can complete a task efficiently without the need for rework.
- Cost Control: Companies are engaging in price wars; OpenAI has reduced the prices of some models by 80%, and DeepSeek’s Flash version costs only $0.14 per million tokens input, which is lower than Grok’s API rates.
- Model Specialization: OpenAI uses a tiered model approach, with Sol handling complex tasks, Terra handling routine work, and Luna performing repetitive tasks to reduce overall costs. Anthropic maintains high prices due to its strength in long-term tasks (such as writing lengthy documents), but it must prove the value of its models.
In simple terms, when customers choose an AI model, they consider not only its capabilities but also its reliability and cost-effectiveness.
4. SpaceXAI’s Advantages
SpaceXAI has a favorable advantage because its various components can complement each other:
- Computing Power: Its Colossus cluster consists of 200,000 GPUs, enabling fast training.
- Data: X provides real-time information and user feedback, while Cursor offers programming data for model training.
- Distribution: The X platform reaches hundreds of millions of users, and Cursor is a popular tool among developers.
However, there are also challenges:
- Organizational Barriers: Data sharing between Tesla (autonomous driving, robotics) and SpaceXAI’s AI models is limited by privacy and regulatory constraints.
- Corporate Trust: Corporate clients are concerned about security and stability; since Grok has just joined the elite ranks, it has yet to build sufficient trust.
- Sales Channels: SpaceXAI lacks a mature corporate sales network like Google Cloud or Microsoft Azure, making it difficult to secure large-scale contracts.
Just as Musk’s goal of having Grok produce a “The Odyssey” film is ambitious, there are many hurdles to overcome to ensure the smooth and reliable completion of such a complex project.
In Conclusion
Grok 4.6 offers hope for Musk’s AI efforts, but to become a successful commercial entity, SpaceXAI must overcome challenges related to task reliability, gaining corporate trust, and establishing a comprehensive resource ecosystem—similar to Odysseus’ journey home, which was full of challenges.