Elon Musk’s AI assistant has also started hiring Claude to get things done!
On October 7, Musk announced on X that the Grok Bot will select the most suitable backend model for each specific task, explicitly naming Claude Opus 5.5, Midjourney, and Suno.
The standard he provided is simple and straightforward:
Use whichever is most likely to give you the best results.
Use whoever works best.
Netizen ExiledonMoon joked that even Musk’s AI assistant has started asking Claude for help with tasks.
Another user, Token Maxxer, pointed out: Musk’s list also includes “other leading APIs.” OpenAI was not named, but it was not explicitly excluded either.
This also leaves an open question: Will OpenAI’s models also become allies for Grok Bot?
At the very least, Claude has already started working.
According to the same report by 9to5Mac, the Grok Bot is now available with Claude Opus 5.5.
You assign the task to Musk's AI assistant, but the actual work might be done by Claude.
While competing with Anthropic for users, why is Musk willing to call on Claude?
Leverage the opponent's model to attract your own users.
From Musk’s emphasis on the “best outcome,” it’s clear he’s thinking practically:
Let Grok Bot handle the user’s tasks first; there’s no need for your own model to manage every step.
If a task can be performed better with Claude, integrating it offers the opportunity to reduce user churn caused by poor experience.
More importantly, even if Claude is doing the work behind the scenes, users can still interact with, assign tasks to, and track progress through the Grok Bot.
Anthropic provides some model capabilities, while Grok Bot organizes these capabilities and delivers the results. As long as tasks are completed well, users have reason to continue using this assistant.
Therefore, invoking the opponent's model can also be a strategy to attract users.
Even if part of the work is completed by Claude, users may still open the Grok Bot first for their next task.
The Grok Bot wants to take on an entire task.
The Grok Bot mentioned by Musk was launched in beta on August 11.
This smart assistant comes with a cloud computer that can perform tasks on behalf of the user.
According to the official description, bots can use tools, log into applications, and handle tasks across software; multiple bots can also communicate with each other, share task context, and collaborate分工.
Work continues even after you step away from your computer.
Multiple Grok bots hand over tasks and report progress.
Elon Musk mentioning Grok Bot does not mean that the Grok chat feature on X will fully switch to Claude.
In addition to Claude, the list includes the image generation tool Midjourney and the music generation tool Suno, spanning text, images, and music—reflecting a practical need: completing an entire task often requires the collaboration of multiple capabilities.
For example, creating a short video requires writing copy, preparing a thumbnail, and adding background music.
Ask a few AIs for help, and the assets might be ready soon. But explaining requirements, moving files, and coordinating revisions often still require you to switch between different windows.
If the Grok Bot can invoke the appropriate models based on tasks to connect these steps, it has the potential to take over some of your coordination work.
Users simply want to switch less between different models and tools, and not have to explain the same requirement multiple times.
I’ll even save you the trouble of choosing which model to use.
Will taking on more models cause users to suffer from decision fatigue?
Open the assistant, but first think again: Should I use Claude or Grok for coding? Which tool should I switch to for creating diagrams?
The official documentation for the Grok Bot has already outlined the design concept:
Model selection is managed by the platform; there is no user-facing model selection menu, and the actual model combination may change over time.
In other words, you assign the task, and the system selects the model for you.
You don’t need to research who recently upgraded just to write a piece of code, nor do you need to sift through the feature lists of every tool just to create a chart.
However, this does not mean users don't need to know what they used and how much they spent.
The official documentation states that usage analytics will show which model was actually used, and charges will be calculated based on the model used.
This information helps users understand where their money is going and determine whether this task is worthwhile.
Is it reliable to let AI choose the model?
After leaving the choice to the system, the question arises: What are the advantages of having AI select models for users, and is this kind of selection reliable?
There is already precedent for one model assigning tasks to other models.
As early as 2023, HuggingGPT attempted to have ChatGPT break down user tasks, select appropriate models based on their descriptions on Hugging Face, and then have the system invoke these models to execute the tasks and aggregate the results.
OpenAI's recently released public beta of the Decisions API also provides tools for such decisions.
The developer provides task information and candidate options; GPT-6 Luna returns a judgment, and the application proceeds to the next step based on that judgment.
When used in a multi-model assistant, it can help determine which model should handle the current task.
The advantage of automatic selection is especially evident when a task requires repeatedly invoking the model.
Simple information organization, complex reasoning, and final text refinement require different levels of model capability. If the system can progressively assess these needs, it can deploy stronger, more expensive models precisely where they are truly required.
Research has verified the cost benefits of such practices.
The 2024 RouteLLM study shows that, in the MT-Bench evaluation, distributing requests between GPT-4 and Mixtral enables the system to achieve approximately 95% of GPT-4’s evaluation performance while reducing costs by more than 85% compared to using GPT-4 exclusively.
Model allocation performance of RouteLLM on MT-Bench.
However, this result compares automated allocation with fixed use of GPT-4; it does not fully prove that AI is better than humans at selecting models, nor does it equate to Grok Bot’s actual performance. More public information is needed to determine how Grok Bot specifically selects models.
Moreover, there is no one-size-fits-all answer to what is "suitable."
Some people are in a hurry, some want to save money, and others are willing to spend more time and cost for better results. When generating images, the standards for a temporary placeholder and a set of ready-to-publish posters are different.
The AI assistant must first understand these requirements before deciding which capabilities to invoke. Otherwise, even the most powerful model may fail to deliver the results the user wants.
For users, whether automatic selection truly saves effort depends on factoring in the time spent checking, revising, and redoing tasks.
Who can become the user's first choice?
In this competition, there's also OpenAI, which has just launched Dots.
On September 29, OpenAI released the 24/7 agent Dots, which comes with a built-in cloud computer, can connect to applications, and continues working even after the user has logged off.
Dots has updated the promotional materials according to the new requirements.
It is competing with Grok Bot for the same opportunity: to become the AI assistant users are willing to entrust with tasks over the long term.
However, competition among assistants may coexist with collaboration in model invocation.
If the Grok Bot integrates with OpenAI’s models in the future, users might delegate tasks to Musk’s assistant, with OpenAI’s model handling part of the work. OpenAI provides the model capabilities, while both assistants continue to compete for users.
Currently, this is still just a possibility. Musk explicitly named Claude this time, but it has not been confirmed whether OpenAI's model will be integrated.
When different assistants have the opportunity to invoke similar or identical models, who better understands the user and can continue from the previous task may become a key factor in the decision.
If an assistant already knows your writing preferences, image style, and can locate previous files, you’ll naturally think of it first when you have a new task.
Switching to another assistant may mean re-explaining the context, reconnecting tools, and re-aligning on workflows.
Integrated Claude to add callable model capabilities to Grok Bot.
Next, these capabilities must be translated into tangible benefits for users:
The requirements have changed—can we still make adjustments? Work is halfway done—can we continue moving forward? The delivered results still require how much effort from the user to finalize?
These experiences will influence which user the task is entrusted to next time.
Reference materials:
https://x.com/testingcatalog/status/2107803665457189061
https://x.com/elonmusk/status/2107724314451878104
https://www.theinformation.com/briefings/musk-says-spacex-will-sometimes-use-rival-models-power-grok-bot
https://developers.openai.com/api/docs/guides/decisions?utm_source=chatgpt.com
This article is from the WeChat public account "New Intelligence Yuan" (ID: AI_era), author: ASI Revelation; editor: Yuan Yu.