🇺🇸 USA · xAI Grok
Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: Elon Musk’s AI chatbot, made by xAI, deeply integrated with X (formerly Twitter). The least filtered of the major Western chatbots; useful but comes with strong caveats about content quality and training-on-your-posts.
Front-matter facts
| Field | Value |
|---|---|
| Vendor | xAI (San Francisco, USA) — owned by Elon Musk |
| Country / origin | 🇺🇸 USA |
| Recommended for Australian users? | ⚠️ Use with caution — accessible from AUS with no region locks, but trains on your posts and chats by default; outputs less filtered (a feature for some users, a risk for others); fewer enterprise data-handling guarantees than Claude/ChatGPT/Gemini |
| Privacy summary | Free + paid: trains on your X posts AND your Grok chats by default. Opt-out exists but has been less consistent than competitors. Has previously had outages and data-exposure incidents. |
| Free tier | Yes — limited Grok access via grok.com or X mobile app |
| Paid tiers | X Premium / Premium+ (bundles Grok with X features); SuperGrok US$30/mo standalone (direct from xAI); SuperGrok Heavy higher tier; Enterprise quoted via xAI API |
| First released | November 2023 (Grok-1 to limited X Premium+ users); generally available 2024 |
| Last reviewed | 2026-06-26 |
| Official site | https://grok.com (also available inside x.com) |
What it is
Grok is the AI chatbot made by xAI, the AI company Elon Musk founded in July 2023 after departing OpenAI’s board. It’s available as a standalone chat at grok.com, embedded in the X (Twitter) mobile app and website, and via the xAI API for developers.
Grok is distinctive in a few ways that matter:
- Real-time access to X (Twitter) as a data source — when you ask “what’s happening with the AUS election today,” Grok can pull live X posts as context
- “Fun mode” that produces snarkier, less filtered responses than competitors’ default voice
- Significantly looser safety filtering than Claude / ChatGPT / Gemini — Grok will engage with topics other chatbots refuse (sometimes a feature for adult research; sometimes a risk in irresponsible use)
- Native integration with X — quoting posts, summarising threads, generating reply suggestions
- Image generation powered by xAI’s “Aurora” model (and previously by partners’ models)
Musk has marketed Grok as “maximally truth-seeking” and explicitly anti-censorship. In practice it has been more permissive about edgy questions, but it has also at times produced outputs (especially in 2024-25) that drew criticism for hallucinations, biased framing, and propagating fringe content.
Analogy: if Claude is “the careful, thoughtful one,” ChatGPT is “the polished best-friend assistant,” and Gemini is “the Workspace-native one,” Grok is “the chatbot built by someone who runs a social network and wants the AI to feel less restrained.” Useful in some contexts, risky in others.
What you’d use it for
- Quick chat with real-time X context — “what are people saying right now about Y” with live post citations
- Summarising trending X threads — particularly useful if you don’t have time to read 500 reply branches
- Less-filtered conversational questions — edgy humour, adult-but-legal topics, comparative-political content where Claude / ChatGPT would refuse or hedge
- General chat — questions, brainstorming, drafting
- Coding chat — Grok’s coding ability has improved markedly with Grok 3 and Grok 4 / Grok Heavy
- Image generation — via Aurora (xAI’s model)
- Image understanding — multimodal Q&A on uploaded images
- Voice mode — spoken back-and-forth (rolling out)
- DeepSearch — Grok’s web-search mode
- Reasoning mode — Grok’s extended-thinking mode for harder problems
How to sign up + first 5 minutes from Australia
- Go to
grok.comin any browser. Or open X (Twitter) mobile app — Grok is built in. - Sign in with your existing X (Twitter) account, or sign up new at grok.com with Apple / Google / email. AUS phone number (+61) accepted.
- Free tier active immediately — limited daily Grok-4 access.
- Try one of these:
- “What’s the most-discussed topic on Australian X today?” (uses real-time X data)
- “Compare the Sydney and Melbourne housing markets in 2026”
- “Write me an image-generation prompt for a cyberpunk Aussie outback scene, then make it” (uses Aurora)
- In the X app: long-press a post → “Ask Grok” to summarise / fact-check the thread
- Optional — upgrade. Two paths:
- Via X Premium / Premium+: increases Grok limits as one of many X features. AUD billing through X.
- Standalone SuperGrok US$30/mo direct from xAI at grok.com → Upgrade. Better if you don’t want X Premium’s other features.
- For the API: sign up separately at
console.x.ai. Pre-paid credits; AUS card accepted.
What it costs — what you actually get
Free tier
- Grok-4 daily message limit
- Image generation (limited Aurora gens/day)
- DeepSearch (limited)
- Reasoning mode (limited)
- Voice mode (rolling out)
- Inside the X app: free if you have an X account
X Premium — ~US$8/month
- Increased Grok limits compared to free
- Plus all X Premium features (verification, longer posts, edit button, ad reductions, monetisation eligibility)
- Billed by X (AUD via X Premium)
X Premium+ — ~US$22-40/month (varies by region)
- Much higher Grok limits
- Premium+ X features (full ad-free, larger creator revenue share, priority replies)
- Includes access to Grok Heavy reasoning at lower tier
- AUD billing via X
SuperGrok (standalone) — US$30/month
- Direct from xAI, no X subscription required
- Higher Grok-4 limits than free or X Premium
- Access to Grok 4 Heavy (extended reasoning)
- Higher Aurora image generation quotas
- AUD card accepted at grok.com
SuperGrok Heavy — higher tier (US$300/month-ish; check current pricing)
- Significantly higher reasoning-mode limits
- For power users running heavy queries
xAI API (console.x.ai)
- Pay-per-token model
- Grok-4, Grok 4 Fast, Grok 4 Heavy, Grok 2 Vision, Aurora image
- Pre-paid credits; AUS cards accepted
- Generally priced competitively with OpenAI / Anthropic but with less mature tooling
Enterprise — quoted via xAI sales
- Less developed than Anthropic Enterprise / OpenAI Enterprise / Google Cloud Vertex AI in mid-2026
- Custom contracts available
Hidden costs to know about
- The X Premium and SuperGrok tiers overlap awkwardly — if you already pay X Premium+, you’re getting Grok bundled; subscribing to SuperGrok on top is rarely needed
- App Store / Play Store may add their own markup on Premium subscriptions
- AUS X Premium billing has fluctuated; check current AUD price at x.com/premium
How it compares to alternatives
| Capability | Grok (SuperGrok) | ChatGPT Plus | Claude Pro | Gemini AI Pro |
|---|---|---|---|---|
| Chat quality | Good (Grok 4) | Excellent | Excellent | Excellent |
| Real-time social data | Best — X integration unique | Some via web search | Limited | Some via Google Search |
| Reasoning mode | Good (Heavy) | Excellent (o-series, GPT-5 Pro) | Excellent (Extended Thinking) | Excellent (Deep Think) |
| Image generation | Aurora (good) | gpt-image-1 (built-in) | None native | Imagen (built-in) |
| Video generation | Limited | Sora (built-in) | None | Veo 3 (built-in) |
| Voice mode | Rolling out | Best (Advanced Voice) | Good | Good (Gemini Live) |
| Safety filtering | Loosest (for better or worse) | Moderate | Moderate-strict | Moderate |
| Hallucination rate | Higher than top 3 historically | Lower | Lower | Lower |
| Enterprise maturity | Least developed | Mature | Mature | Mature |
| Privacy posture | Loosest defaults | Train-by-default + opt-out | Train-by-default + opt-out | Train-by-default + opt-out |
| Coding | Good (Grok 4) | Strong (GPT-5) | Best (Claude Code) | Strong (Antigravity / Code Assist) |
| Price (consumer top) | US$30 (SuperGrok) | US$20 | US$20 | US$19.99 |
If you live on X and want fast access to real-time conversations, Grok’s social-data edge is real. If you want the best general-purpose chat without political controversy or weaker safety, Claude / ChatGPT / Gemini are clearly stronger picks. The encyclopedia generally recommends one of the big three for primary use and Grok as a secondary tool for the specific X-data and less-filtered use cases.
Privacy / data handling deep dive
xAI’s defaults are the most permissive of the major Western providers. This is a real concern worth understanding before using Grok for anything sensitive.
Training on your X posts (separate from Grok chats)
By default, xAI trains its models on your X posts — public AND private (including DMs, in some prior policy versions). Multiple regulators (EU, UK, Brazil, Ireland’s DPC) have pushed back on this; xAI has revised its terms multiple times.
To opt out of X-posts being used for training:
- Open X (Twitter) → click your profile → Settings and Privacy
- Privacy and Safety → Data Sharing (or Grok)
- Find “Allow your posts as well as your interactions, inputs and results with Grok to be used for training and fine-tuning” → toggle OFF
This is one toggle covering both: your X posts AND your Grok chats.
Training on Grok chats
The same toggle controls this. Free and SuperGrok tiers train by default; the toggle opts you out of future training.
xAI API
The xAI API privacy posture has been less consistent than other providers. As of mid-2026, the API has a no-training-by-default commitment in the developer terms — but verify the current Terms of Service at x.ai before relying on this for sensitive workloads.
Enterprise
xAI Enterprise contracts can be negotiated for no-training and data-isolation, but the enterprise tier is less mature than competitors. For regulated AUS workloads (health, finance, government), Claude Enterprise via AWS Bedrock Sydney, M365 Copilot, Azure OpenAI Service, or Vertex AI in australia-southeast1 are stronger choices.
Data retention
X chats and Grok activity are retained in your X account profile until you delete them. The Delete Account flow on X removes most associated data on a 30-day cycle, but verify via the Privacy Dashboard.
Where data lives
Primarily US data centres. No published AUS data-residency option for consumer or standard enterprise as of mid-2026.
Historical concerns
- 2024: EU and UK regulators required xAI to suspend training on European users’ X posts pending consent overhaul. Australia’s OAIC issued related cautions.
- 2024-25: Multiple instances of Grok producing controversial outputs (antisemitic content, election misinformation, fabricated stories about real people). xAI has made adjustments after each.
- 2025: Concerns about xAI engineers having broader access to user chats than typical at competitor labs — verify current internal-access policy via xAI’s Trust Center if this matters to you.
Practical recommendation: opt out of training immediately after sign-up. Don’t paste sensitive personal, financial, medical, or business data into Grok regardless. For business / regulated use, prefer Claude / ChatGPT / Gemini Enterprise tiers — Grok’s enterprise posture isn’t yet at the same level.
See privacy-and-data-training.md 🟩 🟦 for the comparative deep dive.
Recent changes (since the last review)
- 2026: Grok 4 and Grok 4 Heavy released
- 2026: Aurora image model upgraded
- Late 2025: SuperGrok pricing introduced (standalone subscription path)
- 2025: DeepSearch and Reasoning modes generally available
- 2025: Grok 3 launched as the daily-driver model
- 2024: Multiple safety / content / privacy revisions following regulator action
- 2024: Grok generally available to all X Premium tiers
- 2023: Grok-1 launched to X Premium+ users (limited rollout)
(Check x.ai/news for the latest.)
Gotchas
- Grok will engage with things other chatbots refuse. Sometimes this is helpful (mature topics, edgy humour, controversial-but-legal questions); sometimes it produces content you’d rather not see, propagate, or screenshot.
- Real-time X data ≠verified. Grok will happily cite a viral X post as if it’s a fact. Always verify against authoritative sources for anything time-sensitive or contested.
- Election / political answers can be biased. Grok has shown patterns of framing aligned with its owner’s politics. For political research, cross-check with neutral sources.
- Fun mode vs Regular mode — toggle in Grok settings. Fun mode is snarkier; Regular is more measured. Some users find Fun mode produces more hallucinations.
- xAI is owned by Elon Musk — if you object on principle to using a Musk-owned product, that’s a valid consideration. xAI is a separate legal entity from X / Tesla, but ownership is concentrated.
- The X integration cuts both ways. Useful for live data; concerning for privacy (your X account is linked to your Grok identity).
- xAI has acquired X itself in 2025 (xAI bought X for ~US$33B in stock) — they’re now corporate siblings under Musk. This may affect data-sharing and integration over time.
- Enterprise data-handling commitments are less mature than competitors. If you’re in regulated industries, prefer Anthropic / OpenAI / Google / Microsoft enterprise products.
- App Store and X Premium billing are tangled — if you subscribed to X Premium on iOS, cancel via Apple ID → Subscriptions, not via x.com. SuperGrok purchased on grok.com is separate again.
- The brand “Grok” comes from Robert Heinlein’s 1961 novel “Stranger in a Strange Land” — useful trivia, occasionally referenced in xAI marketing.
- Some content is censored. Despite the “minimally filtered” branding, Grok still refuses certain content (CSAM, illegal activities, specific weapons synthesis, certain copyrighted material). Don’t believe marketing that suggests it’s “uncensored.”
See also
- ChatGPT 🟩 🟦
- Claude.ai consumer 🟩 🟦
- Gemini 🟩 🟦
- Microsoft Copilot 🟩 🟦
- Perplexity 🟩 🟦
- Meta AI 🟥
- Aurora (xAI’s image model) 🟥
- What is an LLM? đźź©
- Decision frameworks — Claude vs ChatGPT vs Gemini 🟥
- which-ai-for-which-job.md 🟩 🟦
- privacy-and-data-training.md 🟩 🟦
- Glossary — G (Grok) 🟩