It's the first time in the history of the world.
In the meantime, Musk's AI assistant has begun to ask Claude to work!
On October 7, Musk announced on X that Grock Bot would select the most suitable rear-end model based on the specific task and would be directly named Claude Opus 5.5, Midjourney and Suno.
The criteria he gave were simple and straightforward:
Which one is most likely to bring you the best results?
Who wants to use, who wants to use.
Exiledon Moon, an AI assistant for Musk, also started looking for Claude to help.
Another networker, Token Maxxer, warned that Musk's list also includes "other leading APIs". OpenAI is simply not branded and has not been explicitly excluded.
This also leaves a dilemma: will the OpenAI model also serve as an outlet for Grock Bot?
At least Claude is addicted.
According to 9to5Mac reports on the same day, Grock Bot is now available with Claude Opus 5.5.
You've given the task to Musk's AI assistant, and it's possible that Claude is behind it.
Why is Musk willing to do this while competing with Anthropic for users and calling Claude?
Borrowing models from opponentsFighting for your usersAccording to Musk's emphasis on "best results", he considers the following to be realistic:
First, let Grock Bot do the user's job without having to have their own model do all the work.
If a task allows Claude to do better and access it, there is a chance to reduce the number of users switching to other products due to a poor experience.
More importantly, even if Claude is behind it, where users switch needs, connect tools, follow tasks, it can still be Grock Bot.
Anthropic provides some modeling capabilities, and Grock Bot is responsible for organizing these capabilities and delivering results.
Therefore, invoking an opponent's model can also be a means of gaining users.
Even if some of the work is done by Claude, the next time a user has a task, they may open the Grock Bot first.
Grok Bot wants to be next.It's the whole thing.Grok Bot, mentioned by Musk, launched a beta version on August 11.
This smart assistant carries a cloud-based computer and can perform tasks on top of it for users.
According to the official description, bots can use tools, login applications, and cross-software processing; multiple bots can also communicate with each other, share task contexts, and collaborate.
After the user leaves the computer, the work can continue.
Several Grok Bot communication missions, reporting progress.
This time, Musk mentioned Grock Bot, and not the same as the Grock chat feature on X, it should be completely replaced with Claude.
In addition to Claude, the image generation tool Midjourney and the music generation tool Suno, which spans text, images, and music, respond to a very real need: to do the whole thing, it often takes several abilities to collaborate.
For example, make a short video, write a document, prepare a cover, and accompany a piece of music.
Looking for some AI help, material may be available soon. But interpreting requests, moving files, coordinating modifications, you often have to go back and forth between different windows.
If Grok Bot can summon the right model according to the task and connect these links together, you will have the opportunity to coordinate the next part for you.
What the user wants is simple: switching back and forth between several models and tools, without having to explain the same needs multiple times.
Which model to chooseI want to save for you.The more models are received, the more users will be forced to "select difficulty"?
When you open your assistant, you need to think again: should I write code with Claude or Grok?
Grok Bot's official documentation has given design ideas:
Model selection is managed by the platform, without a user-oriented model selection menu, and the actual model combinations used may change over time.
In other words, you switch tasks, and the system chooses models for you.
You don't have to study who's upgraded recently to write a piece of code; you don't have to go through a list of tools to create a chart.
However, that doesn't mean users don't need to know what they're using or how much they're spending.
The official document states that the usage analysis shows which model is actually used and the cost is calculated according to the model used.
This information can help users understand where money is spent and whether the task is budgeted.
Let AI choose a modelDo you think it's unreliable?After giving the choice to the system, the question arises: what is the advantage of allowing AI to choose a human model that is not reliable?
There is already a precedent for assigning work to one model to another.
As early as 2023, HuggingGPT attempted to have ChatGPT dismantle user tasks, select suitable models based on the model on the Hugging Face, and then call these models to perform tasks by the system.
OpenAI's recently released Decisions API also provides tools for this type of choice.
The developer gives the task information and the candidate options, which are returned to judgment by GPT-6 Luna, and the application takes the next step according to the judgment.
It can be used in multiple model assistants to assist in deciding which model to assign to the current task.
The advantages of automated selection, especially when a task requires repeated calls to a model.
Simple information ordering, complex reasoning, and the final word coloring are not the same as the requirements for model capability. If the system can be judged gradually, it has the opportunity to use stronger, more expensive models where it really needs to be.
Research has confirmed the cost-effectiveness of this approach.
A 2024 RouteLLM study showed that in MT-Bench ratings, by allocating requests between GPT-4 and Mixtral, the system can reduce the cost of using GPT-4 as a whole by more than 85% when it reaches about 95% of GPT-4 ratings performance.
RouteLLM is the model allocation effect in MT-Bench.
However, this comparison between automated allocation and fixed use of GPT-4 does not fully demonstrate that AI is more likely to select models than humans, nor does it match Grok Bot's test scores, how Grok Bot specifically selects models, and requires more public information to judge.
Moreover, there is no one-size-fits-all answer.
Some people want to save time, some people want to save money, and some people are willing to spend more time and money in exchange for better results.
The AI assistant must first figure out these requirements and then decide which capabilities to call. Otherwise, the model will be stronger and may not produce the results the user wants.
For users, the automated choice of whether or not to pay attention to the time of checking, modifying and returning to work must also be taken into account.
Who can be the first choice of users?In this competition, there is also OpenAI, which has just launched Dots.
On September 29, OpenAI released the all-weather smart body Dots, also a cloud-based computer, which connects applications and continues to work after users leave.
Dots modified the release material according to the new requirements.
It competes with Grock Bot for the same opportunity: to become an AI assistant who is willing to give tasks to its users in the long run.
However, competition between assistants may also exist alongside collaboration on model calls.
In the future, if Groc Bot accesses the OpenAI model, the user can hand over the task to Musk's assistant, and the OpenAI model does some of the work. OpenAI provides model capabilities, and the two assistants continue to compete for the user.
For the time being, this is only a possibility. Musk has made it clear that Claude has not yet confirmed whether he has access to the OpenAI model.
When different assistants have the opportunity to call on similar or even identical models, who understands the user better, who can then complete the previous job, may be the key to influencing the choice.
If an assistant already knows your file preferences, image style, and can find the previous file, it's easier to think about it the next time you have a task.
Switching to other assistants may mean switching backgrounds, connecting tools, and refining the workflow.
Joining Claude adds to the callable modeling capabilities of the Grock Bot.
The next step is to translate these capabilities into benefits for users to feel:
The demands have changed and can't continue to change; the work is half done and can't continue to progress; the results have been delivered and the amount of effort required by the user to finish...
These experiences will affect who the user assigns the task to next.