Summary of Key Points
Two months after fading from the mainstream model competition, Elon Musk has returned with Grok 4.5. This time, he has abandoned the approach of competing on who has the strongest parameters and instead focused on practical cost-effectiveness: offering near-top-tier model capabilities at a lower cost and faster speed, targeting real-world productivity scenarios for businesses. Additionally, this is SpaceX's first product since integrating the programming tool Cursor, marking a new beginning for Musk's AI business following the dissolution of xAI, as it seeks to make a breakthrough in the corporate tools market.
1. Musk Becomes More Pragmatic: No Longer Competing on “Strongest Models,” but on “Usefulness and Low Cost”
In the past, AI companies would boast that their models were the “best in the world,” but Musk has changed his strategy. He openly acknowledges that Grok 4.5 is not as powerful as Anthropic’s Claude Fable, emphasizing that “most tasks do not require such top-tier capabilities.” For example, tests have shown that Grok 4.5 can achieve results comparable to Fable at one-seventeenth of the cost. Musk says they are more concerned with practicality in the real world—it’s more important for Tesla and SpaceX’s engineers to find a tool that is useful than to be number one on testing lists.
In simple terms, while before companies competed on who could process data the fastest, now Musk is focusing on whether a model can get the job done with the least amount of resources (in this case, computational power).
2. How Good Is the Cost-Effectiveness?
The cost-effectiveness of Grok 4.5 is evident in three aspects:
1. Low Cost: The pricing is $2 per million input tokens (units of text that the model processes) and $6 for the output, which is cheaper than many similar models. For example, in real-world work tasks, Grok 4.5 only costs $0.49 to complete a task, compared to China’s Zhipu GLM-5.2 and Kimi K2.6; programming tasks cost about $2.5, which is one-fifth of Claude Fable 5 and half of GPT-5.5.
2. Fast Speed: It can generate 80 tokens per second, significantly faster than other models, meaning users get responses more quickly.
3. Efficient Use of Tokens: Grok 4.5 uses fewer tokens to complete the same tasks. For instance, it only needs about 16,000 output tokens for a programming task, which is three-quarters less than Claude Opus 4.8, resulting in lower costs.
In layman’s terms: Grok 4.5 does the same job for less money and faster, without wasting resources.
3. Acquiring Cursor to Fill a Weakness: Grok 4.5 Targets Corporate Work Tools
Previously, Musk’s AI team had a weakness in programming. Therefore, in June of this year, SpaceX acquired the programming tool Cursor. Grok 4.5 is the first product to integrate Cursor’s capabilities:
- During training, it used trillions of real development data from Cursor, learning not only how to code but also how developers modify code, use tools, and collaborate with AI—essentially infusing the model with programmers’ “work experience.”
- Grok 4.5 now targets real office scenarios such as working with Excel spreadsheets, creating PPTs, and handling complex software projects.
This means Grok is no longer just a chatbot; it has evolved into a productivity assistant for corporate employees, helping with tasks like writing code, generating reports, and making presentations.
4. A New Start for the Team After Major Changes: SpaceX Needs AI Success Stories
There have been significant changes behind Grok’s development. From February 2025 to March 2026, key members of xAI (Musk’s previous AI company) left one after another, and the entire founding team departed. In May of this year, xAI was dissolved, and Grok was integrated into SpaceX as a sub-product of SpaceXAI.
Grok 4.5 is the first product created by the restructured SpaceXAI team, which is crucial for Musk. He needs to prove to investors that SpaceX can excel in AI, just as it does with rockets. With AI being a hot topic in the tech industry, SpaceX needs more AI-related achievements to support its valuation and future growth.
5. The Industry Shifts from “Rankings” to “Task Costs”
In the past two years, AI companies focused on testing rankings to determine which models were the best. This year, however, the focus has shifted to cost per task—how much it costs to complete a real-world task. Developers now say that “task cost” is the most important factor when deploying AI. For example, if a model can achieve 95% of the performance of top models at a low cost, it’s more useful than one that ranks first but is extremely expensive, as most business tasks are not the most challenging. Grok 4.5 hits this balance perfectly, making it very attractive to businesses.
5. Can Grok 4.5 Shake Up the Market?
Grok 4.5’s cost-effectiveness is impressive, but whether it can truly change the market landscape depends on whether companies and developers adopt it. Currently, leading models like OpenAI and Anthropic already hold a large market share. Whether Musk’s “practical” approach will succeed remains to be seen based on user feedback. Nevertheless, this comeback at least ensures that Musk is back in the AI game, no longer just an observer.