Grok Bot will no longer run only on SpaceXAI’s own AI models. In a post on X early on Wednesday 7 October 2026, Elon Musk said that SpaceX “will use the best back end model for any given task” inside Grok Bot, “including Claude Opus 5.5, MidJourney, Suno and other leading APIs”. Claude Opus 5.5 comes from Anthropic, a company Musk was calling “misanthropic and evil” in February.
The announcement reads like a reversal, and City A.M. framed it as Musk turning to a rival to power Grok. But the product’s own documentation already says that model choice is handled for the user, with “no fixed vendor set guaranteed”. Musk’s post makes public a design that was already in the small print.
This guide explains what Musk said, how the agent works and who can use it, why Opus 5.5 is the model he named, and how the two creative services he listed fit. It then turns to the questions that matter to a business: where your data goes, what each task costs, how much control administrators really have, and what to check before rolling Grok Bot out.
Table of contents
- What Musk Said About Grok Bot
- What Grok Bot Is and Who Can Use It
- The Grok Bot Docs Already Allowed Other Models
- Why Grok Bot Would Route Work to Claude Opus 5.5
- What a Grok Bot Task Could Cost on Each Model
- Midjourney and Suno: Two APIs That Barely Exist
- What Changes for Grok Bot Customers
- A Checklist Before You Roll Out Grok Bot
- The Bigger Shift: Owning the Router
- Grok Bot FAQs
- References
What Musk Said About Grok Bot
The post went up at 06:46 UTC on 7 October and tagged the product’s own @Bot account on X. It is three sentences long.
The post, word for word
“Important note regarding Grok @Bot: Going forward, @SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs. Whatever is most likely to give you the best outcome.” Musk did not say when the change takes effect, which tasks go to which supplier, or whether any of the three companies has signed a new agreement.
How far it travelled
When we read the post through the fxtwitter API on the evening of 7 October, it showed 35.2 million views, 65,586 likes, 4,927 reposts, 4,546 replies, 2,527 quote posts and 6,028 bookmarks. Those figures will keep rising. For comparison, the @Bot account’s post announcing proactive suggestions on 1 October had about 21.2 million views.
What the replies added
Investor’s Business Daily reported that Grok “clarified in replies that it will pick the strongest model available for each specific task, or whatever works best on a case-by-case basis.” That is the whole operating principle so far. There is no published list of task types, no routing table and no statement on how the system will decide between suppliers when two are close.
What Grok Bot Is and Who Can Use It
Grok Bot is SpaceXAI’s agent product, launched in beta on 11 August 2026. It is not the Grok chatbot. Each Bot is a named assistant that works on a persistent cloud computer with a browser, file system and terminal, signs in to your apps, and keeps working when your laptop is closed.
An agent with its own computer
The documentation describes Bots as “AI teammates with names, jobs, and context that compounds over time”. You message a Bot, grant access as it asks, and it handles multi-step work across websites and apps. All of a user’s Bots share one cloud computer, including its files, browser sessions and logins. The docs state plainly that the computer “belongs to your account, not to an individual Bot”. We covered how that sharing works when Grok Bot gained voice mode.
A fast run of new features
The product has shipped features almost weekly. Voice arrived on 17 September and voice notes the next day. A Finance integration through Plaid followed on 26 September, then Team Bots in public beta on 28 September. On 30 September Bots gained the ability to hand coding tasks to Cursor, and on 1 October the proactive “Primary Bot” suggestions we described in our Grok 4.7 and Primary Bot report arrived.
Who gets Grok Bot
| Plan | Grok Bot access | Notes from the docs |
|---|---|---|
| SuperGrok, SuperGrok Plus, SuperGrok Heavy | Included | Link an individual subscription |
| Cursor Pro, Pro Plus, Ultra | Included | Every paid individual Cursor plan |
| Cursor Teams (Standard and Premium) | Included, on by default | No switch to turn it off on Teams |
| Enterprise | Through sales (the docs FAQ says rolling out) | Admin enables it; group-level access; model allowlist |
Subscriptions include weekly usage, and eligible accounts can add on-demand usage “billed from model and token cost”. If you hold both a Cursor and a SuperGrok subscription, it draws on whichever has more usage left.
The Grok Bot Docs Already Allowed Other Models
The most useful context for Musk’s post is not in the post. It is in the security section of the product documentation, which has described a multi-supplier product all along.
“Cursor manages model selection”
SpaceX completed its $60 billion purchase of Cursor in August, and the agent runs on Cursor’s accounts, sign-in and privacy settings. The docs say: “Cursor manages model selection. There is no customer-facing model picker, and the serving mix can change over time with no fixed vendor set guaranteed.” Our report on the SpaceX Cursor acquisition explains how the two businesses fit together.
Cursor already sells Claude Opus 5.5
Cursor’s own pricing page lists frontier models “from OpenAI, Anthropic, Google, SpaceXAI, and more”, including Claude Opus 5.5 at Anthropic’s list price. It runs two usage pools: a “Cursor Models” pool with much more included usage for Grok 4.7 and Composer 2.5, and an “Other Models” pool for third-party models “charged at the model’s API price”. On Teams and Enterprise plans, a Cursor Router picks the model for each automatic request. So the commercial channel to Anthropic already exists inside a SpaceX business.
Billing follows the serving model
One line in the docs has direct cost consequences: “Usage analytics show the model that served each request, including failovers, and billing follows the serving model.” If a Grok Bot task is routed to a more expensive supplier, it uses more of your allowance, or more on-demand spend. The docs promise visibility after the fact, not a say beforehand.
Why Grok Bot Would Route Work to Claude Opus 5.5
Musk has spent much of 2026 conceding that Anthropic is ahead. Naming Opus 5.5 as a back end for Bots is the commercial version of that admission.
Musk’s own comparisons
In mid-September, before Grok 4.7 shipped, Musk wrote that it “should be roughly on par with Opus 5.0, not 5.1”, according to Cryptopolitan. On 28 September he shared a five-minute film that an independent creator made by giving Claude Opus 5.5 access to Midjourney for 12 hours, wrote that he “really felt the AGI profoundly this time”, and replied “Accurate” to a post rating the model close to general intelligence, Memeburn reported. That film used exactly the pairing he has now named for Grok Bot.
A relationship that was already two-way
Anthropic is also one of SpaceX’s large computing customers, renting capacity at the Colossus data centres. Musk publicly backed Dario Amodei’s call to pace frontier development in September, as we reported in Musk backs Amodei. Meanwhile SpaceXAI sells Grok 4.7 through rivals’ clouds: Amazon Bedrock on 28 September, Google’s Gemini Enterprise Agent Platform on 1 October and Microsoft Foundry on 7 October.
What Anthropic’s terms say
Anthropic’s Commercial Terms of Service, effective 17 June 2025, bar customers from using its services “to build a competing product or service, including to train competing AI models or resell the Services except as expressly approved by Anthropic”. Whether routing Grok Bot tasks to Claude needs that approval, or is already covered by Cursor’s existing agreement, is for the two companies. We found no public comment from Anthropic on Musk’s post by the evening of 7 October.
What a Grok Bot Task Could Cost on Each Model
Because billing follows the serving model, list prices are a fair guide to how fast each supplier would consume a Bot allowance. Cursor, Anthropic and SpaceXAI all publish them per million tokens.
The published prices
| Model | Input per million tokens | Output per million tokens | Source |
|---|---|---|---|
| Grok 4.7 (under 200k prompt tokens) | $2 | $6 | SpaceXAI API docs; Cursor |
| Claude Sonnet 5.5 | $2 | $10 | Cursor pricing |
| Claude Opus 5.5 | $4 | $20 | Anthropic pricing; Cursor |
| Claude Opus 5.5, fast mode | $8 | $40 | Anthropic pricing |
| Claude Fable 5.1 | $10 | $50 | Anthropic pricing; Cursor |
A worked example
Take a Grok Bot task that reads 500,000 input tokens and writes 50,000 output tokens, spread across requests under 200,000 tokens each, with no caching. On Grok 4.7 that costs $1.00 of input plus $0.30 of output, or $1.30. On Claude Opus 5.5 it costs $2.00 plus $1.00, or $3.00, about 2.3 times as much.
Each figure is 0.5 times the input price plus 0.05 times the output price from the table above; bars are scaled to the highest, $7.50.
What the example does not show
Real tasks vary enormously, caching cuts input costs sharply, and a stronger model that finishes a job in fewer steps can cost less overall. SpaceX has not said how the router will weigh price against quality. The point is narrower: the same job can draw down very different amounts of your usage depending on a routing choice you do not make.
Midjourney and Suno: Two APIs That Barely Exist
The two creative services in Musk’s post raise a different question, because neither has offered a general public developer API.
Midjourney
As of mid-2026, Midjourney “still does not offer an official public API for developers”, according to an Analytics Insight buyer’s guide published in July. Developers who want its images programmatically use unofficial aggregators. Midjourney has spent the year on other projects, including a full-body ultrasound scanner it unveiled in June. For Grok Bot to use Midjourney as a back end, SpaceX would need direct access of a kind Midjourney has not announced.
Suno
Suno said in July that it was “exploring a developer API” and planned “to start with a curated group of partners”, in a LinkedIn post by product chief Jack Brody reported by Digital Music News. At that time it had no official public API. Suno has since released Suno Speech, which we covered in our report on Suno’s spoken-word beta. In July, Digital Music News reported that Universal Music Group and Sony Music still had active copyright cases against it, and musicians have filed further suits since.
What that implies
If Grok Bot is to call Midjourney and Suno directly, it points to partner arrangements that neither company has announced. We found no statement from either by the evening of 7 October. Until they confirm, treat those two names as intentions. SpaceXAI also has its own image and video generator, Grok Imagine, so Bots are not short of creative tools without either supplier.
What Changes for Grok Bot Customers
For an individual user, better answers from a stronger model are simply good news. For a business, Musk’s post changes the supplier list behind a tool that may already be switched on for your whole team.
More companies may see your data
Every new back end is another company that may process prompts, files and outputs. The docs say zero data retention “follows Cursor’s existing provider agreements” and that model providers “do not keep prompts or outputs”, but also that “providers may run abuse and safety classifiers, and flagged data may be stored for investigation”. Under UK GDPR, that makes them sub-processors you should know about.
The allowlist is Enterprise only and not guaranteed
Administrators who want to restrict Grok Bot to approved models face a hard limit. “The team model allowlist is Enterprise only, and enforcement is not guaranteed,” the docs say, adding that onboarding “presents an acknowledgement that Grok Bot may not follow the list”. On the Teams plan the product is on by default, and “there is no switch to turn it off”.
Location and logging
Grok Bot computers run “in the United States today”, the security docs say. That is not the same as Cursor’s US-only data residency programme, which “does not apply to Grok Bot by default”. Usage analytics do show which model served each request, including failovers, which gives auditors a record after the fact.
| Control | Teams plan | Enterprise plan |
|---|---|---|
| Turn Grok Bot off | Not possible | Admin enables it and can block all members |
| Choose the model | No picker | No picker |
| Restrict models with an allowlist | Not available | Available; enforcement not guaranteed |
| See which model served a request | Usage analytics | Usage analytics |
| Exclude data from training | Privacy Mode | Privacy Mode |
A Checklist Before You Roll Out Grok Bot
None of this means Grok Bot is unsafe to use. It means the supplier picture is broader than the brand suggests, and your paperwork should reflect that. Bring your cybersecurity and legal leads in before the Bots are given access to company systems.
Map the sub-processors
List every model provider that could serve Grok Bot requests, including Anthropic, OpenAI and Google through Cursor, and any creative services SpaceX adds. Check them against your customer contracts. The docs advise that if your contract limits sub-processors, you should “contact your account team before rolling out Grok Bot”. Our vendor management service maps these chains for clients.
Decide whether Teams is enough
If you need to control which models touch your data, the Teams plan cannot do it. Price the Enterprise plan with its allowlist, and test whether the allowlist is actually honoured in your configuration, given the documented warning that it may not be.
Watch usage, not just results
Because billing follows the serving model, check usage analytics weekly for the first month. A shift towards premium models would show up there before it shows on an invoice. Set alerts on on-demand spend.
Keep humans on the risky steps
The agent hands login, two-factor authentication and payment steps to the user, and sensitive actions can go through Auto Review. Keep those defaults, and decide which actions always need a person. That is part of good IT governance for any agent with access to live systems.
The Bigger Shift: Owning the Router
City A.M. made the strategic point well: “As models become better at different tasks, owning the interface that chooses between them could prove as commercially important as building the underlying technology itself.”
SpaceX’s two bets
SpaceX is still building its own models. Investor’s Business Daily reported, citing the Financial Times, that SpaceX is seeking about $40 billion to buy Nvidia chips. But through Cursor and Grok Bot it also owns two interfaces that can send work to anyone’s model. If Grok 4.8 or 4.9 closes the gap, the router can send more work home. If not, the agent still sells.
Anthropic gains either way
For Anthropic, being named as the preferred back end by its loudest critic is valuable marketing. City A.M. cited Capital on Tap data suggesting Anthropic took as much as half of UK small businesses’ spending with major AI providers in September, while ChatGPT’s share fell from 60% to 36%. Each Grok Bot task routed to Opus 5.5 adds to that.
What to watch next
Three signals will show whether the change is real. The first is a changelog entry or routing note from SpaceXAI. The second is a statement from Anthropic, Midjourney or Suno confirming a partnership. The third is your own usage analytics showing Opus 5.5 serving requests. Expect the security page to change too, although its current wording about the serving mix is already broad enough to cover Musk’s plan. Until then, plan as if any frontier model could handle your data.
Grok Bot FAQs
Is Grok Bot switching to Claude?
Not entirely. Musk said Grok Bot will use “the best back end model for any given task”, naming Claude Opus 5.5 as one option. SpaceXAI’s own Grok models remain in the mix.
When does the change take effect?
Musk gave no date. The product docs already say Cursor manages model selection with no fixed set of suppliers.
Can I choose which model Grok Bot uses?
No. There is no model picker. Enterprise customers can set a model allowlist, but the docs say enforcement is not guaranteed.
Will Grok Bot cost more?
Possibly. Billing follows the serving model, so tasks routed to Claude Opus 5.5 consume usage at a higher rate than Grok 4.7 at list prices.
Do Midjourney and Suno have APIs Grok Bot can use?
Neither offered a general public API as of mid-2026. Suno was exploring a partner-only API. Any such integration would need arrangements neither has announced.
Where does Grok Bot process data?
Its computers run in the United States. Model providers may process requests under Cursor’s provider agreements.
References
Elon Musk: Important note regarding Grok Bot (X)
Musk turns to rival Anthropic to power Grok (City A.M.)
SpaceX seeks $40 billion to buy Nvidia AI chips (Investor’s Business Daily)
Elon Musk’s Grok Bot AI will use third-party models Claude, Midjourney and Suno (dev.ua)
Grok Bot security (SpaceXAI docs)
Grok Bot security FAQ (SpaceXAI docs)
Grok Bot for teams and enterprises (SpaceXAI docs)
Grok Bot is now available for SuperGrok, Cursor Pro and Cursor Teams (SpaceXAI)
Models and pricing (Cursor docs)
Claude API pricing (Anthropic)
SpaceXAI API pricing (SpaceXAI docs)
Commercial Terms of Service (Anthropic)
Musk admits Grok 4.7 is behind Anthropic and OpenAI’s latest models (Cryptopolitan via MSN)
Elon Musk says Claude Opus 5.5 made him feel the AGI (Memeburn)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
