Use Grok Build for cheap, parallel coding work and keep Claude Code for production repos. Grok Build reads your Claude Code setup unchanged, but its July upload incident was fixed in config, not removed from the code. Prices and docs below were checked October 21, 2026.
| Your reason to look beyond Claude Code | Pick | What you give up |
|---|---|---|
| Token cost on bulk or parallel tasks | Grok Build on grok-build-0.1 | A clean trust record; upload code still ships |
| You want to keep CLAUDE.md, plugins and MCPs | Grok Build | Nothing in config; you change billing and vendor |
| Model freedom, local models or an MIT license | OpenCode | Claude Code's bundled plan and defaults |
| Production monorepo, team seats | Stay on Claude Code | Grok's lower per-token price |
In short: Grok Build wins on cost and setup reuse, Claude Code on track record, OpenCode on openness.
Should you stay on Claude Code?
Stay if your repo holds production code, secrets or client work. Claude's pricing page says "Claude Code is included in all paid plans." Pro costs $17 a month billed annually or $20 monthly. Max starts at $100 a month. Team standard is $20 per seat a month billed annually, or $25 monthly.
The Free plan is $0 but does not include Claude Code. So the floor for Claude Code is a paid plan.
Composio's write-up, based on 50+ hours of use, scored Claude Code ahead 6 to 3. If your setup works and cost isn't the problem, you have little to gain by moving. Your plugins and CLAUDE.md files keep working there, and your team's billing stays in one place. Moving only makes sense when a specific reason below applies to you.
If cost is your reason: Grok Build bills less per token
Grok Build's own model is the cheap part. xAI's model page lists grok-build-0.1 with a 256k context window:
- Under 200k prompt tokens: $1.00 input, $0.20 cached input, $2.00 output per 1M tokens.
- At 200k or more: $2.00 input, $0.40 cached input, $4.00 output.
Grok 4.6 is the bigger option, with a 500k context window. It costs $2.00 input, $0.50 cached and $6.00 output per 1M under 200k tokens. At 200k or more, it doubles to $4.00 / $1.00 / $12.00.
So grok-build-0.1 output costs one third of Grok 4.6 output. On grok-build-0.1, cached input is 80% cheaper than fresh input ($0.20 vs $1.00). Repeat-heavy agent loops benefit most.
A worked example helps. Say a batch of tasks uses 10M input tokens and 2M output tokens, all under 200k per prompt:
- grok-build-0.1: $10 input plus $4 output, $14 in total.
- Grok 4.6: $20 input plus $12 output, $32 in total.
Composio reports $2.49 vs $11.80 cost per task (Artificial Analysis figures) between the two tools; treat it as one benchmark, not your bill. Cache hits cut the input side further. Claude Code is priced differently: a flat plan fee, not per token. So compare your monthly token volume against the plan price, not token to token.
The plan side is less clear. Reports disagree on whether a free tier exists. Some say it opened in August; others say there is none. Wikipedia says Grok Build needs a SuperGrok subscription at $30 a month. We could not confirm any plan price on xAI's own page, so check grok.com/plans before you budget.
If you want to keep your setup: Grok Build reads it unchanged
This is the strongest reason to try it. Grok Build's compatibility docs state: "Grok is fully compatible with Claude Code with zero configuration needed."
Per the same page, Grok reads:
- Instruction files: CLAUDE.md, Claude.md, CLAUDE.local.md and .claude/rules/.
- Extensions: Claude Code marketplaces, plugins, skills, MCPs, agents and hooks.
- Plugins in the TUI: "a Marketplace tab for browsing and installing plugins from configured sources."
That makes a trial cheap. Point Grok Build at a repo you already use with Claude Code and compare results. Your config files stay the same, so you can switch back the same day.
If trust or model freedom is your reason: OpenCode
OpenCode is the open option. Its site says it is "public on GitHub under the MIT License." It lists over 215,000 GitHub stars and 950 contributors. The GitHub repo itself showed 219.1k stars on the same day, so treat star counts as rough.
What you gain:
- Model choice: 75+ LLM providers through Models.dev, including local models.
- Price: "OpenCode is 100% free to use," with a set of free models. Other providers may cost extra.
- Paid models, optional: OpenCode Go is $10 a month and Go Plus is $40 a month. Go offers about 29 curated models.
What you give up: Claude Code's bundled subscription and defaults. You pick, pay for and tune the model yourself.
If code must never leave your machine, pair OpenCode with a local model. Our air-gapped coding stack guide covers that setup.
Is Grok Build safe after the July upload incident?
It is safer than in early July, but the fix lives in config, not in removed code. Simon Willison's review of the open-sourced repo lays out the sequence.
- July 12: xAI "disabled default retention for all Grok Build users."
- July 14: Developer backlash over uploaded data.
- Mid-July: xAI released the harness under Apache 2.0, as "just a single commit." Sources give July 15 and July 16.
- After release: Musk said "all user data that was uploaded to SpaceXAI before now will be completely and utterly deleted."
Two caveats remain. First, gcs.rs still contains Google Cloud Storage upload logic. Today, upload_session_state() returns a hard-coded session_state_upload_unavailable error. A config change could turn it back on. Second, no source we read describes an independent audit of the deletion.
The harness is open, but the models it calls are paid xAI API models. Ownership matters too. Musk's statement names SpaceXAI as the holder of the uploaded data, and SpaceX now owns Cursor. Cursor announced on August 14, 2026, that it "is now part of SpaceX." The all-stock deal was about $60 billion. If you want to avoid one vendor across your tools, factor that in. See our Cursor page and Cursor vs Windsurf for that side.
What each costs in October 2026
| Grok Build | Claude Code | OpenCode | |
|---|---|---|---|
| License | Harness: Apache 2.0 | - | MIT |
| Entry price | Reports say SuperGrok, $30/month; check vendor page | Pro: $17/month annual, $20 monthly | Free; Go $10/month |
| Higher tier | Check grok.com/plans | Max from $100/month | Go Plus $40/month |
| Model price | grok-build-0.1: $1 in / $2 out per 1M | Flat plan fee (Pro $17-20/month) | Depends on provider |
| Reads Claude Code config | Yes, zero config (xAI docs) | Native | - |
| Local models | Not listed | Not listed | Yes, via Models.dev |
| Trust note | July upload incident | No incident in our sources | Open repo, MIT |
Plan prices are from Claude's pricing page and opencode.ai, checked October 21, 2026. Grok plan prices are secondary reports only.
How to switch from Claude Code to Grok Build
- Keep Claude Code installed. Run both on the same repo for a week.
- Clean the repo first. Move secrets out before the first run, since upload code still ships.
- Launch grok. On first launch, it "opens a browser for authentication."
- Set the model. xAI's docs name no default; set it in ~/.grok/config.toml.
- Leave your config alone. CLAUDE.md, rules, plugins, MCPs and hooks are read as-is.
- Check the Marketplace tab. Confirm your plugins loaded.
What doesn't carry over: your Claude plan. Grok billing is separate, and plan terms are unconfirmed, so set a spend limit before long runs. If you also use Grok for chat, see the Grok app and Claude and Grok compared.
Decision rule: client or production code, stay on Claude Code. Throwaway or parallel tasks on a budget, try Grok Build. Need local models or an MIT license, use OpenCode.
