The Bench
SpaceXSI, Grok, and Grok Bot: models, plans, and our verdict
SpaceXSI's Grok models and Grok Bot: where they stand, plans and API prices, data controls, and recent changes.
In this piece
SpaceXSI on the bench
- AdoptGrok: Grok 4.7 with live web and X search, in the Grok apps and the API.Strongest at: Real-time research · Cited answers · Market and social signal
- AdoptGrok Bot: Agent teammates on a persistent cloud computer, with skills and routines.Strongest at: Recurring work · Event-driven routines · Slack and GitHub workflows
SpaceXSI, the company formerly known as xAI, builds the Grok models, the Grok apps, and Grok Bot, its persistent agent teammates. Its strengths are real-time information, through search across the web and X, and agents built to keep working after a conversation ends. This page covers where SpaceXSI stands, the models and products that exist today, what they cost, how data is handled, and what changed recently. We checked every source linked here on November 10, 2026, mostly SpaceXSI's documentation at docs.x.ai. Plans and model names change often, so read each figure as of that date.
A note on names. SpaceX announced on February 2, 2026 that it had acquired xAI (announcement). The documentation now titles itself SpaceXSI Docs and calls the company SpaceXSI, and the announcement page's footer names SpaceX Super Intelligence LLC. The developer products keep the older name: the docs still refer to the xAI API and the xAI Python SDK (web search guide), and requests still go to api.x.ai (API overview). We use SpaceXSI for the company and the product names as the docs print them.
Where it stands
Grok and Grok Bot are both on Adopt: ready for serious business use with the controls described below. These are our judgments from hands-on use and SpaceXSI's documentation.
Grok earns Adopt for real-time, sourced information. Grok 4.7, the recommended default for everything except image, video, and voice work, can call web search and X search on the Responses API, and SpaceXSI states that Grok knows nothing of current events beyond its training data unless one of those tools is enabled (models, Grok 4.7 overview). The research controls are inspectable: web search accepts up to five allowed or excluded domains, and responses return citation data a reviewer can check (web search, citations). The caveat sits in that same documentation: citations are not guaranteed on every answer, so sourced output still needs checking.
Grok Bot earns Adopt for agents that work like teammates. Each Bot has a name, a job, and memory (Bots), and works on a persistent cloud computer with a browser, filesystem, and terminal (overview). Routines run in the cloud while your laptop is closed, on a schedule or when an event arrives from Slack, GitHub, Linear, Sentry, PagerDuty, email, or a webhook (skills and routines). Its controls are documented too, including approval rules, a setting that blocks execution on your own computer, and email from routines or other work you did not start held as a draft until someone presses Send (approvals, security, and privacy).
The trade-offs: all of a person's Bots share one cloud computer, so separate Bots are not a security boundary, and Grok Bot chooses the model for each task, which can be a third-party model, with no way to pin one (FAQ). A Cursor team's model allowlist does not govern Grok Bot either (security).
The lineup today
| Surface | Model or product | What it is for | Status on Nov 10, 2026 |
|---|---|---|---|
| xAI API | grok-4.7 | Coding, agentic tasks, knowledge work; 500k context | Recommended default since September 21 |
| xAI API | grok-4.6, grok-4.5, grok-4.3, and three grok-4.20 variants | Earlier models; the 4.3 and 4.20 family is cheaper with a 1M context | Available |
| xAI API | grok-build-0.1 | Agentic coding | Public beta on the API since May 29 |
| xAI API | grok-imagine-image-2.0, grok-imagine-video-1.5, grok-imagine-video-1.5-lite | Image and video generation | grok-imagine-image-quality retired November 2 |
| xAI API | grok-voice-3.0, grok-voice-transcribe-2.0 | Speech to speech and transcription | grok-voice-2.0 deprecated October 14 |
| Grok apps | grok.com, iOS, Android | Chat, Grok Imagine, voice, file uploads, connectors | Free to start; paid plans raise limits |
| Grok Business, Enterprise | Team workspaces on grok.com | Licensed team use with sharing controls | Enterprise adds organizations and SSO |
| Grok Bot | Desktop (macOS, Windows, Linux), iOS, Android, X Chat, @bot on X, Slack | AI teammates on a persistent cloud computer | Launched in beta August 11 |
Sources: models, release notes, Grok Build 0.1 announcement, Grok overview, organizations, Grok Bot FAQ.
The models page recommends Grok 4.7 for everything except image, video, and voice work. The Grok 4.7 announcement names the API, Cursor, and Grok Build, SpaceXSI's coding agent, as launch surfaces. The Grok 4.7 overview adds the US regional endpoint and the OpenRouter, Vercel, and Cloudflare model gateways. Grok 4.7 Fast, the same model on faster infrastructure at twice the token rates, runs only in Cursor and Grok Build (same page). The announcement does not say which model the Grok apps use, so we do not assume the app behaves like an API request.
Plans and pricing
Figures below are list prices as of November 10, 2026, from SpaceXSI's documentation, excluding sales tax.
The X Subscription launched on October 12, 2026 and bundles Grok, Cursor, and X (announcement). New self-serve customers choose from these plans, existing Cursor, SuperGrok, and X Premium plans continue, and switching is optional (pricing).
| Plan | Monthly | Annual | Included usage | Grok Bot |
|---|---|---|---|---|
| Free | $0 | $0 | Limited | Not included |
| Plus | $8 | $80 | Base | Not included |
| Premium | $30 | $300 | 4x Plus | Included |
| Super | $100 | $1,000 | 15x Plus | Included |
| Ultra | $200 | $2,000 | 50x Plus | Included |
Team seats on Premium, Super, and Ultra cost the individual monthly price per seat, and Plus is not offered to teams. One weekly pool covers Grok Bot and SpaceXSI models across Grok and Cursor, while third-party models in Cursor draw on prepaid credits from the first use (pricing). Paid plans include Grok API access, but API calls are billed separately at model rates and do not draw from the pool (x.ai pricing). We could not load an official price list for SuperGrok tiers or Grok Business seats, so we leave those prices out; Grok Business sells SuperGrok and SuperGrok Heavy licenses through the console (license management).
Grok Bot can be reached through several routes (FAQ, plans and billing, teams and enterprises):
| Route | Grok Bot access |
|---|---|
| The X Subscription Premium, Super, Ultra | Included for individuals and team seats |
| Legacy Cursor Pro, Pro+, Ultra, self-serve Teams | Included |
| SuperGrok, SuperGrok Plus, SuperGrok Heavy, or X Premium+ linked to a legacy Cursor account | Usage grant; does not stack with a Cursor plan, and the link is permanent |
| Self-serve Grok Business seat | Sign in with Grok |
| Cursor Enterprise | An admin enables it from the Cursor dashboard |
| SuperGrok Lite, Grok Enterprise | Cannot be linked |
API prices per million tokens, for prompts below 200k tokens (pricing):
| Model | Input | Cached input | Output |
|---|---|---|---|
grok-4.7, grok-4.6 | $2.00 | $0.50 | $6.00 |
grok-4.5 | $2.00 | $0.30 | $6.00 |
grok-4.3 and grok-4.20 models | $1.25 | $0.20 | $2.50 |
grok-build-0.1 | $1.00 | $0.20 | $2.00 |
A prompt of 200k tokens or more doubles every rate for that request. Web search costs $5 per 1,000 calls. X search costs $5 per 1,000 posts and $10 per 1,000 profiles returned, the per-item pricing in effect since September 21, 2026 (X search). Code execution costs $5 per 1,000 calls and collections search $2.50. The Batch API takes 20% off the Grok 4.x models above but not grok-build-0.1, Priority Processing doubles token rates, and the US regional endpoint adds 10% (pricing). Requests still sent to the retired grok-imagine-image-quality slug are now billed at the lower grok-imagine-image-2.0 low-quality rate (migration guide).
Data, privacy, and compliance
Each surface has its own terms. On the API, SpaceXSI says it never trains on inputs or outputs without explicit permission, and stores requests and responses encrypted for 30 days for abuse auditing before deleting them (security FAQ). Zero Data Retention is available team-wide but disables stored features such as the Files and Batch APIs. The same page states SOC 2 Type 2 compliance and does not claim HIPAA compliance (a BAA can be requested). For US-only processing, us.api.x.ai serves only grok-4.7 and grok-4.6, with no image, video, or voice APIs; files, collections, and server-side tools such as web and X search still work there but fall outside the US guarantee (regional endpoints).
Text watermarking is now documented for the API. The output watermarking page, last updated November 9, says it is off by default and that team admins can turn it on per model for grok-4.7 through the grok-4.20 family; some models may be locked on for a team, and every opt-in and opt-out lands in the team audit log. An earlier version of the page said watermarking would be on by default from November 7 (archived November 2). The mark uses Google DeepMind's SynthID Text, adds no visible characters, covers structured outputs and tool-call arguments, and shows up as an x-watermark: true response header. Access to detection is by application. SpaceXSI says it can help with EU AI Act compliance.
For the Grok apps, the security FAQ refers to an "improve the model" toggle in data controls on grok.com and mobile; when it is on, Build mode sessions may be used for model improvement, and Private Chat is unavailable in Build mode. The Grok FAQ says deleted chats and files are removed within standard retention windows without stating their length. Grok Business team workspaces are covered by SpaceXSI's enterprise terms, with custom retention on the Enterprise tier (Grok Business guide). Admins set sharing ceilings (license management), and organizations, including SSO, are Enterprise-only (organizations).
Grok Bot runs on Cursor accounts and follows Cursor's privacy settings. With Privacy Mode on, customer data is not used for training (security), and Privacy Mode (Legacy) blocks Grok Bot entirely (teams and enterprises). Grok Bot computers run in the United States, and US-only data residency is available through an account team. Anysphere, the company behind Cursor, holds ISO/IEC 27001 and ISO/IEC 42001 certifications with Grok Bot in scope. Marking outside content as untrusted reduces but does not eliminate prompt-injection risk, the page notes; since the November 5 release, the Bot also tells you when a page hides instructions for AI agents and that it did not follow them (changelog). Audit logs, network controls, and enforced Auto Review are Enterprise-only. On a self-serve Cursor Teams plan, Grok Bot is on for every member and there is no switch to turn it off, so the rollout plan has to come before the seats. Admins can set retention periods that delete inactive Bots, inactive members' computers, and old messages; each is off by default, and deleted data cannot be recovered (teams and enterprises).
What changed recently
- August 11: Grok Bot launched in beta for SuperGrok and Cursor subscribers on desktop and iOS, with an Enterprise waitlist. The announcement described Bots learning a workflow and saving it as a routine (announcement). The current docs list desktop apps for macOS, Windows, and Linux, mobile apps for iOS and Android, and access through X Chat and Slack.
- August 12: Grok 4.6 reached the API.
- September 21: Grok 4.7 launched on the API, in Cursor, and in Grok Build. X search per-item pricing applies from this date.
- September 25: The
safety_identifierfield lets developers attribute policy violations to an end user. - October 2:
grok-voice-transcribe-1.0reached end of life; version 2.0, the default since September 17, takes its requests. - October 12: The X Subscription launched for new customers. Its docs list Grok Bot in Premium, Super, and Ultra.
- October 14:
grok-voice-3.0arrived andgrok-voice-2.0was deprecated. - October 27: Custom moderation controls added a
guidelinesarray for thensfw,violence, andprofanitycategories (moderation controls). - November 2:
grok-imagine-image-qualityretired. Its requests, and those togrok-imagine-image-pro, are served bygrok-imagine-image-2.0at low quality, $0.01 per image cheaper, with no code change required (migration guide). - November 2: SpaceXSI described Data Bot, a Team Bot its data science team teaches with shared skills and reviewed shared memory, which it estimates handled about 70 hours of the team's work in one week. The estimate is SpaceXSI's own (post).
- November 3 to 6: Grok Bot releases 0.78.4 through 0.79.3. After you approve a flight, stay, or grocery cart and its total, the Bot now pays on its own and asks again only if something material changes. The Bot flags pages that hide instructions for AI agents, admins can set a retention period for inactive members' computers, and Bots can reach MCP servers on a personal Tailscale network (changelog).
- November 9: SpaceXSI changed its API text watermarking to off by default. The watermarking page had said text output would be watermarked by default from November 7 (archived November 2). The version updated November 9 says admins turn it on per model (output watermarking). The release notes do not mention either.
No new text model has appeared since Grok 4.7; grok-4.7 is still the recommended default (models). API entries come from the release notes unless another source is linked.
Where it fits
Grok fits work that depends on what is happening now: market and competitor monitoring, vendor and regulation tracking, and anything where public conversation on X is part of the signal. On the API, grok-4.7 with web and X search suits applications that need current public information with sources a user can open, and SpaceXSI's image, video, and voice models cover media work from the same account (models).
Grok Bot fits teams that want agents running recurring work inside the tools they already use, from scheduled briefs to engineering and operations workflows triggered by Slack, GitHub, or alerts. New customers get it through The X Subscription at Premium and above, and Enterprise admins can turn it on for an organization (Grok Bot FAQ, teams and enterprises).
It is a weaker fit where every task must run on one pinned model, or where HIPAA coverage is required, since SpaceXSI's documentation does not claim it. Teams sharing Bots should treat the shared cloud computer as one environment and connect only the apps each job needs.
What we're watching
- Whether the developer products take the SpaceXSI name; the docs currently mix both.
- Which model the Grok apps run, and whether Grok Bot gains model pinning or allowlists.
- Enterprise unification: some Enterprise accounts cannot yet switch to The X Subscription (overview).
- Teach a task, which is rolling out gradually, and inference hooks, still in beta.
- Moderation controls, where
declineis the only behavior available today. - Published pool sizes for The X Subscription, which the docs describe only as multiples of Plus.
- Text watermarking: whether the off-by-default setting holds, which models a team finds locked on, and whether the release notes record the change.
- A retirement date for
grok-voice-2.0, deprecated since October 14 with no end-of-life date listed. - How much spending Grok Bot does on its own now that an approved cart or booking completes without a second confirmation, and whether admins get a team-wide purchase control.
From verdict to deployment
Putting SpaceXSI to work is the hard part.
We design and build AI into the websites and applications a business runs on, automate workflows with agents that work across its tools, and roll AI out across whole organizations, with every team trained on its real work. We build on whichever lab fits the job, and we stay through launch and adoption.
- WebsitesSites that explain the business and put AI in the customer journey.
- ApplicationsCustom apps, portals, and platforms with AI built into the product.
- Automation & agentsConnected tools and agents that take on work across your systems.
- Rollout & trainingOrganization-wide adoption, with teams trained on their real work.
Put this into practice
Bring the decision you’re working through.
A free discovery session is a chance to review your goal, the tools or website you have, and the help you need. We’ll identify a sensible next step and discuss scope before any project commitment.
Custom AI Systems & Agents →AI Advisory & Training →Book a free call