Free Allowances for AI Coding CLIs
Checked on 2026-08-12. A free login, a paid subscription allowance, API trial credit, and a temporary promotion are different things. Vendors can change all four without changing the CLI binary.
Short Answer: Is “Grok CLI” Free?
There is an official xAI terminal agent, now called Grok Build, and its command is grok. The official installer is:
curl -fsSL https://x.ai/cli/install.sh | bash
On first launch it can authenticate through an xAI account; XAI_API_KEY is the alternative for API-backed use. This should not be confused with unrelated third-party projects named grok-cli on GitHub.
As of this snapshot, xAI explicitly says “Try it in Grok Build for free” on the Grok 4.6 launch page. The same page offers 2× included usage for the first week. That establishes a real free entry point today, but xAI does not publish a durable numerical free quota, reset schedule, or permanence commitment for Grok Build.
Verdict: yes, it is currently possible to try the official Grok Build CLI for free; no, the public evidence does not support calling it a stable, quantified “forever-free tier.” Treat it as free-but-undisclosed and promotion-sensitive.
The ambiguity is not academic. Grok Build launched on 2026-05-25 as an early beta limited to SuperGrok and X Premium Plus subscribers. The 2026-08-12 page broadened the message to free trial access. Old posts that describe either state as permanent are already stale.
Current Comparison
| Tool | Zero-payment path | Published allowance | Authentication | Important boundary |
|---|---|---|---|---|
| Gemini CLI | Yes, stable documented tier | 60 requests/minute and 1,000 requests/day | personal Google account | One prompt may make several model requests; CLI and Code Assist agent mode share and aggregate the daily pool across models. API-key auth also advertises 1,000 requests/day, but it is a separate route. |
| OpenAI Codex CLI | Yes | Free plan exists for “quick coding tasks”; no public numeric Free-plan quota | Sign in with ChatGPT | Current numeric tables start at Plus. API-key mode is pay-as-you-go and has no cloud features. Do not reuse paid-plan numbers for Free. |
| GitHub Copilot CLI | Yes | CLI is included in Copilot Free; agent interactions draw from an undisclosed monthly AI-credit allowance. Separately, Free includes 2,000 IDE code completions/month. | GitHub account with Copilot Free | The 2,000 figure is code completion, not 2,000 CLI prompts. Free uses automatic model selection, and unused credits reset monthly rather than roll over. |
Grok Build (grok) | Yes, currently advertised | Not disclosed; launch promotion says 2× included usage for the first week | xAI account, or XAI_API_KEY | Free entry is confirmed, durable quota is not. API-key use follows xAI API pricing rather than proving a free CLI allowance. |
| Claude Code | No general $0 subscription route documented | Included limits exist on paid Pro/Max/Team/Enterprise plans; API/Console access is metered | paid Claude login, Console/API credentials, or cloud provider | Pro/Max usage is shared with Claude apps. An ANTHROPIC_API_KEY takes precedence and can create API charges. “Included” does not mean free. |
| Qwen Code | Not through the former first-party OAuth tier | None now for Qwen OAuth; discontinued 2026-04-15 | paid Alibaba Coding Plan, provider API key, or local endpoint | The harness is open source and can use free local/third-party models, but that is not a current Qwen-hosted free quota. Old claims of 2,000 requests/day are obsolete. |
What the Numbers Actually Count
The cleanest allowance here is Gemini's, but even it is not “1,000 prompts.” Google states that one prompt can trigger multiple model requests. The daily limit is shared between Gemini CLI and Gemini Code Assist agent mode and aggregated across model families. For practical planning, inspect consumption rather than counting terminal submissions.
Codex, Grok Build, and Copilot increasingly expose compute-weighted limits rather than a fixed message count. A long repository task, tool-heavy loop, or large context can consume much more allowance than a short question. A vendor dashboard is therefore the account-specific source of truth when public documentation gives only a qualitative tier.
Subscription routes and API-key routes must also remain separate:
- signing into Codex with ChatGPT uses the plan allowance; an API key is metered by API tokens;
- signing into Claude Code with a paid Claude account uses shared subscription limits;
ANTHROPIC_API_KEYswitches to API billing; - signing into Grok Build may use included account usage;
XAI_API_KEYinvokes the priced xAI API route; - an open-source harness such as Qwen Code can be free to install while its selected model is local, free promotional, or paid.
Best Zero-Cost Starting Order
- Gemini CLI for the clearest and largest published no-payment allowance.
- Codex CLI for a real free plan when a small, undisclosed allowance is acceptable.
- GitHub Copilot CLI if a GitHub account and automatic model selection fit the task; use the billing dashboard for the actual AI-credit balance.
- Grok Build as a current trial, not as a capacity commitment. Run a small canary before depending on it.
- Open-source BYOK harnesses only after choosing the model endpoint; “free CLI” says nothing about inference cost.
Claude Code belongs in the comparison because it is a major terminal tool, but not in the zero-payment queue. Qwen Code's former OAuth quota should be removed from current recommendation lists.
Verification Checklist
Before starting work that must finish within a free allowance:
- Confirm the exact official product and executable.
- Record the authentication route: consumer login, subscription, API key, or local model.
- Open the provider's usage page or run the CLI's status command where available.
- Check the reset window and whether other products share the pool.
- Disable or remove API-key fallbacks if an accidental charge would be unacceptable.
- Use a disposable repository and a small task first; a successful login does not prove enough remaining capacity for an agent loop.
Update Rule
Recheck this note after 14 days, or immediately when a vendor changes plan names, authentication, or quota UI. Promote a number into the table only when an official source states its unit and reset period. Keep “not disclosed” rather than deriving a quota from anecdotes, frontend flags, or a single account's remaining balance.