Trying SuperGrok and Grok Build next to Claude Code
I drive a Tesla, and one day I noticed the car could talk to something called Grok. I used it a couple of times while parked, mostly because it was sitting there, then went home and looked up what xAI even was. That led me into the usual stuff — X, the company, pricing, people arguing about benchmarks. I signed up because a SuperGrok promo made it cheap enough that I didn't have to care about any of that yet.
SuperGrok
SuperGrok is the paid plan on grok.com and the Grok apps. It's separate from X Premium. Normal price is about $30 a month. I got closer to three months for $30. That was the whole decision. I've already got Claude and Cursor on the books; I wasn't looking for a fourth subscription theory. I just wanted enough runway to see if the coding side was real.
Grok Build
Once I was in, what I actually used was Grok Build. You install their CLI and run it in a repo:
curl -fsSL https://x.ai/cli/install.sh | bash
cd your-project
grok
It drops you into a fullscreen terminal UI — plan mode, diffs, tools, the usual agent loop. There's a headless mode too. Some people wire it into VS Code or Cursor; those get language servers from the editor, not from Grok Build. On its own it's still "agent edits files and runs commands," same basic shape as Claude Code.
The phone app I barely care about. ChatGPT, Claude, Grok — for a one-off paste, they're close enough that I use whatever's already open. Build is the only reason SuperGrok lasted past the first week.
I use Claude Code on this blogs monorepo and Cursor at work. More on Claude here: what I actually use Claude Code for. Grok Build sits next to them. Some days I open grok because SuperGrok is already paid and I want a second pass. Some days I don't.
Compared to Claude Code, Codex, and Cursor
| Grok Build | Claude Code | Codex | Cursor Agent | |
|---|---|---|---|---|
| Surface | Terminal TUI | Terminal TUI | Terminal TUI | GUI IDE |
| Entry command | grok |
claude |
codex |
Cursor app |
| Company | xAI | Anthropic | OpenAI | Cursor |
| Models | Grok | Claude | OpenAI coding models | Cursor-routed models |
| Plan, diffs, tools | Yes | Yes | Yes | Yes |
| Project memory | AGENTS.md, skills, MCP | CLAUDE.md, skills, MCP | Agent instructions, MCP | Cursor rules, index |
| Built-in LSP | No | No | No | Yes |
| I use it for | Second agent | Long monorepo work | Rarely (OpenAI stack) | Day-job IDE work |
Claude Code is still what I reach for when a session is going to run long. Tool use feels more reliable, and CLAUDE.md means I don't have to re-explain the monorepo every time. I've already taught that setup my defaults. Grok Build hasn't gotten that investment from me, so it shows — more babysitting, more "no, that package lives over here."
Codex is the same class of thing as Grok Build and Claude Code: a terminal TUI agent on OpenAI's models. I'm not deep on it right now because I'm already paying for SuperGrok and Claude. A third terminal agent only makes sense if work forces OpenAI or a task just goes better there.
Cursor is different. It's a full GUI IDE with an agent inside it. Types, jump-to-def, the file tree, the thing I live in at work. Grok Build doesn't replace that. Claude Code doesn't either. When I need the editor, I use the editor.
What Grok Build has going for it, for me: the model is good enough that I keep patches, the promo made it cheap to try, and I like having another agent that isn't Claude when I'm stuck. The harness is still the weak part. Claude Code is further along on the stuff around the model — the loop I don't have to watch as hard.
Renewal
Tesla is how I heard the name. The promo is how I paid. Grok Build is the only piece that would make me renew when the discount ends.
I'll check the same way I check everything else: did I run grok last week, or did I just go back to Claude Code and Cursor? So far Build still gets opens. I'll decide on the bill when the promo runs out.