Grok Explained: What It Is, How to Access It, and How It Compares to ChatGPT and Claude

Human Verified 0

Share to: , and

A clean, wide illustration of a single conversational AI core with live data streams flowing into it from many directions — posts, news, images — over a dark, minimal background, deep blue and white. No text, no X logo.

Grok Explained: What It Is, How to Access It, and How It Compares to ChatGPT and Claude

If you know Grok only from headlines on X, you probably have the wrong picture of it. Grok is not one product. It is four: a chatbot you talk to, a family of models that power it, an always-on agent that does work for you on its own cloud computer, and a coding agent that builds software. The chatbot is the part most people meet. The models are the part developers pay for. The agent and the coding tool are the parts that decide whether Grok is worth paying for at all. This guide explains all four, with the prices and limits xAI itself publishes, and it corrects three mistakes that appear in almost every third-party Grok guide on the internet: the current flagship is Grok 4.7, not Grok 4.3; new customers buy The X Subscription, not SuperGrok or X Premium; and the buttons called DeepSearch and Think mode are not features you will find in Grok today. Checked 14 November 2026 against xAI’s official docs, pricing pages and announcements.

The short version, before the detail, is this. Grok is xAI’s AI assistant, live on X and at grok.com, built on the Grok 4 family of models. Its one genuine edge over ChatGPT and Claude is live data: it can search X’s live post stream and the web while it answers, with citations. Everything else about it is a capable general chatbot with a voice mode, image and video generation, file analysis, memory and a coding agent. The free tier lets you try it for nothing, with limits xAI does not publish in numbers. The X Subscription costs $8, $30, $100 or $200 a month depending on how much usage you need, and the always-on Grok Bot agent starts at the $30 Premium tier. Developers pay per token through the API, where the flagship Grok 4.7 costs $2.00 per million input tokens and $6.00 per million output tokens, and the cheaper Grok 4.3 costs $1.25 and $2.50 with twice the context window. Pick Grok when live X data or the agent products matter to you; pick something else when they do not. The rest of this guide is the evidence for that verdict.

A clean, wide illustration of a single conversational AI core with live data streams flowing into it from many directions — posts, news, images — over a dark, minimal background, deep blue and white. No text, no X logo.

The Short Answer: What Grok Is, in One Paragraph

<img src="

A person asking a question to a glowing, friendly voice interface on a phone, warm daylight, minimal composition.

” alt=”A person asking a question to a glowing, friendly voice interface on a phone, warm daylight, minimal composition.”>

Grok is xAI’s AI assistant: a chatbot you can talk to on the web at grok.com, inside the X app, on iOS and Android, and by voice in Tesla cars. Behind the chatbot sits the Grok 4 family of models, the current flagship being Grok 4.7, released 21 September 2026. Around the chatbot sit two agent products: Grok Bot, an always-on agent that works on a persistent cloud computer across your apps, and Grok Build, a coding agent that writes and runs software. There is also Grok Imagine, the image and video generation layer built into the chat thread. That is the whole product family in one paragraph, and keeping the four pieces apart is the key to understanding everything that follows, because each one has a different price, a different access route, and a different set of limits.

The question most readers actually want answered is whether Grok is better than ChatGPT or Claude. The honest answer is that it depends on what you ask it to do, and the comparison section near the end of this guide takes that question seriously, task by task, without inventing a winner. But one thing can be said up front: Grok’s real advantage is narrow and specific. Its model was trained on data up to May 2026, so it does not know what happened after that date from its own memory. What it can do, and what its rivals do less directly, is search the live stream of posts on X and the wider web in the middle of an answer, and show you the sources it used. That matters when the question is about something happening right now — a breaking story, a trend forming on X, a niche conversation that has not yet reached the news. For most other questions — writing, coding, analysis, summarising documents — Grok is a strong general chatbot competing on capability and price, not a different kind of thing.

Three corrections belong in the first section, because the idea file this guide was built from predates the current lineup, and because most of the Grok coverage on the web still repeats the old facts. First, the flagship model is Grok 4.7, released 21 September 2026. Grok 4.3, which many roundups still call the flagship, is now the cheap, long-context workhorse of the family. Second, access for new customers is The X Subscription — five tiers from Free to Ultra at $200 a month — which replaced SuperGrok and X Premium for new sign-ups on 12 October 2026. The old plans continue only for people who already have them. Third, the feature names DeepSearch and Think mode do not appear on xAI’s current product page. The capabilities those names used to describe still exist, but xAI now presents them as Search, Deep reasoning and Multi-agent mode. If an article tells you to tap a DeepSearch button, it is describing an interface that no longer exists.

A note on the company name, since it causes confusion of its own. The docs and the x.ai site now brand the company SpaceXSI — SpaceX Super Intelligence — after SpaceX acquired xAI in February 2026. The announcements still say xAI, and this guide uses xAI throughout, because that is the name on the products, the API and the pricing pages you will actually use.

Grok Is Four Products, Not One

Four distinct doors or panels in a row, each a different colour and height, in a bright modern hallway — one product family, four entrances.

Ask four people what Grok is and you will get four different answers, and all four will be partly right. That is the first thing to fix. Grok the chatbot is the conversational assistant at grok.com and inside X: you type or speak, it answers, and it can search the web and X, generate images and video, analyse files, and remember your preferences across chats. It is free to try, and the paid tiers of The X Subscription buy more of it. Grok the model family is the set of Grok 4.x models behind the chatbot, sold to developers through the xAI API at per-token prices: Grok 4.7, 4.6, 4.5, 4.3, the three Grok 4.20 variants, and the coding model Grok Build 0.1. Grok Bot is a different kind of product again: an always-on agent with access to a real computer in the cloud, which does work across your apps — Slack, Notion, Gmail, Calendar and more — rather than just answering questions. And Grok Build is the coding agent: a surface where Grok writes, tests and runs software, powered by the Grok Build 0.1 model and the flagship models. Grok Imagine, the image and video generator, is the media layer that runs inside the chat thread rather than a standalone product.

Each product has a different price shape, and confusing them is how people end up comparing the wrong numbers. The chatbot is a flat monthly subscription with a shared weekly usage pool. The API is pay per token, so a heavy month costs more than a light one and a single session can be priced to the cent. Grok Bot has no standalone price at all: it is included in The X Subscription from the Premium tier up, and its usage draws from the same shared weekly pool as the chatbot. Grok Build is partly a subscription product and partly an API product — the coding agent is included in the subscription tiers, while the Grok Build 0.1 model is also sold on the API at $1.00 per million input tokens and $2.00 per million output tokens, making it the cheapest text model xAI sells. A month of heavy chatbot use is not an API bill, and the $30 Premium subscription should never be compared to a $2.60 API session as though they were the same thing.

The access routes follow the same split. The chatbot lives at grok.com and in the X app, with iOS and Android apps and synced history across all of them; you sign in with X or with email. It also lives in Tesla cars as a hands-free voice companion, and as an add-on for Google Workspace apps — Docs, Sheets, Slides and Gmail — which xAI’s own news post covers as a real access route. The models live on the API at console.x.ai, where one prepaid key covers text, vision, voice, images and video, and through distribution channels including Cursor, Grok Build, Amazon Bedrock, GitHub Copilot, Google Cloud, Microsoft Foundry, OpenRouter, Vercel and Cloudflare. Grok Bot lives wherever its connectors live: in chat, in Slack, on X, on desktop and mobile. Grok Build lives in its own coding surface and inside Cursor. Which door you walk through determines what you pay and what you get, and the rest of this guide takes the doors one at a time.

The Model Lineup in One Table

An orderly row of labelled-free glass containers of different sizes on a workbench, suggesting a model comparison at a glance.

Here is the whole current lineup in one place, taken from xAI’s model catalogue and pricing page as read on 14 November 2026. Prices are US dollars per million tokens, for short-context requests — prompts under 200,000 tokens. The long-context rule, which doubles these prices once a prompt reaches 200,000 tokens, is explained below the table, because it changes the arithmetic in a way that catches people out.

ModelAPI codeReleasedContextInput $ / 1MCached $ / 1MOutput $ / 1MReasoning effortsBatch
Grok 4.7 (flagship)grok-4.721 Sep 2026500k$2.00$0.50$6.00low / medium / high (default) / xhigh20% off
Grok 4.6grok-4.612 Aug 2026500k$2.00$0.50$6.00low / medium / high / xhigh20% off
Grok 4.5grok-4.516 Jul 2026500k$2.00$0.30$6.00low / medium / high (default) / xhigh20% off
Grok 4.3grok-4.32026*1M$1.25$0.20$2.50none / low (default) / medium / high / xhigh20% off
Grok 4.20 Reasoninggrok-4.20-0309-reasoning2026*1M$1.25$0.20$2.50reasoning variant20% off
Grok 4.20 Non-reasoninggrok-4.20-0309-non-reasoning2026*1M$1.25$0.20$2.50non-reasoning variant20% off
Grok 4.20 Multi-agentgrok-4.20-multi-agent-03092026*1M$1.25$0.20$2.50multi-agent variant20% off
Grok Build 0.1 (coding)grok-build-0.12026*256k$1.00$0.20$2.00reasoning supportedNot supported

* xAI publishes no release date for Grok 4.3, Grok 4.20 or Grok Build 0.1 in its docs or announcements. The honest wording is “released in 2026; xAI gives no exact date.” Third parties report various dates for these models, but none of them is confirmed on an xAI page, so this guide does not state them.

Three patterns in that table matter more than any single row. First, the top three models cost exactly the same: Grok 4.7, 4.6 and 4.5 are all $2.00 in and $6.00 out, with the same 500,000-token context window. You do not pay more for the newest model; you pay the same price for a better one, which is why the older flagships stay available mainly for stability and for platforms that have not yet moved to 4.7. Second, the cheaper models are not weaker in every dimension. Grok 4.3 and the three Grok 4.20 variants cost $1.25 in and $2.50 out — 60% less on input and 140% less on output than the flagship — and they give you twice the context window, 1 million tokens against 500,000. The trade is capability against context and cost, not simply newer against older. Third, Grok Build 0.1 is the cheapest text model on the API at $1.00 in and $2.00 out, but it has the smallest context window at 256,000 tokens and no Batch discount, because it is a specialist coding model, not a general-purpose one.

The long-context rule deserves its own paragraph, because it is the trap in this pricing. Once a prompt reaches 200,000 tokens, the higher rate applies to all tokens in that request, not just the part above the threshold — the price doubles on the whole request. Grok 4.7, 4.6 and 4.5 become $4.00 in, $1.00 cached ($0.60 for 4.5) and $12.00 out. Grok 4.3 and 4.20 become $2.50 in, $0.40 cached and $5.00 out. Grok Build 0.1 becomes $2.00 in, $0.40 cached and $4.00 out. A request that crosses the threshold costs twice as much as the same request just under it, which is why the 1-million-token context of Grok 4.3 is both its selling point and its risk: long documents are exactly the requests that trigger the doubling. The worked examples later in this guide show the arithmetic.

Two more mechanics complete the picture. Rate limits differ sharply between the tiers of the family: Grok 4.7 and 4.5 allow 150 requests per second and 50 million tokens per minute, while Grok 4.3, 4.20 and Build 0.1 allow 37 requests per second and 10 million tokens per minute. And the model names use aliases: a bare model name points to the latest stable version, the “-latest” suffix points to the newest, and a dated suffix pins one fixed release. Old model names still resolve, but they now serve a newer model — the retired names redirect to Grok 4.3, as the final section explains — so developers should pin the exact code they mean rather than relying on silent redirection. For the parallel guide to the other big model family, see our Gemini models explained guide, which covers the same kind of lineup — flagship, fast and cheap tiers, context windows and per-token prices — for Google’s models.

Grok 4.7: The Current Flagship

A lighthouse beam cutting through dusk over a calm sea — the newest, brightest light in the range.

Grok 4.7 is the current flagship, released 21 September 2026, and xAI describes it as its most capable model for coding and knowledge work. It is a new, larger base model than Grok 4.6, trained with a longer reinforcement-learning run on harder, multi-hour tasks, and trained to natively understand the Grok Bot harness — the agent framework described in the Grok Bot section below. xAI’s own description of it, paraphrased, is that it works longer on difficult tasks, checks its own work more carefully, and comes with the company’s best-calibrated safeguards to date. It is also better at documents and presentations than its predecessors. On the API it costs $2.00 per million input tokens and $6.00 per million output tokens, with a 500,000-token context window, cached input at $0.50, and configurable reasoning efforts — low, medium, high and xhigh, with high as the default. The reasoning effort is the dial that decides how long the model thinks before it answers: higher effort costs more in output tokens and takes longer, and buys more careful work on hard problems.

The benchmark table xAI published with the launch is worth reproducing, with a clear label on what it is. These are xAI’s own chosen benchmarks and harnesses, run by xAI, not independent audits, and the comparison models are the ones xAI chose to include. Grok 4.7 is shown at xhigh effort against Grok 4.6 at high effort, GPT-5.6 Sol at maximum effort, and Fable 5.1 at maximum effort:

BenchmarkGrok 4.7Grok 4.6GPT-5.6 SolFable 5.1
CursorBench 4.0 (coding, longer tasks)46.3%40.4%41.7%51.8%
DeepSWE v1.171.0%*65.2%72.7%70.0%
EEBench (electrical engineering)64.0%53.0%39.4%56.4%
AA Briefcase v1.1 (multi-hour office work)1,6571,5461,4871,678
Terminal-Bench 4.037.6%20.3%37.3%57.9%
Harvey Legal Agent Benchmark19.6%15.8%2.5%6.7%
HealthBench Professional56.7%48.5%60.5%62.1%
Input / output price per 1M tokens$2 / $6$2 / $6$4 / $20$10 / $50

* The DeepSWE figure of 71.0% is at high effort, per xAI’s own footnote. Read the table the way xAI presents it and two things stand out. Grok 4.7 beats its predecessor on every benchmark xAI chose, sometimes by a wide margin — Terminal-Bench jumps from 20.3% to 37.6%, and the legal agent benchmark from 15.8% to 19.6%. But it does not win the table outright: Fable 5.1 leads on CursorBench, Terminal-Bench and HealthBench, and GPT-5.6 Sol leads on DeepSWE and HealthBench. The honest reading is that Grok 4.7 is a large step up from Grok 4.6 and a competitive flagship that is also, by a wide margin, the cheapest of the four models in the table on price per token. No third-party benchmark scores are quoted in this guide, because none were verified against a primary source for it — the numbers above are xAI’s own, and they should be read as xAI’s own.

The safety claims in the launch post need the same label. xAI says Grok 4.7 was built with an entirely new safeguard stack, and describes it as the strongest model it has tested on refusals and jailbreak resistance. It says the model tops LatchBio’s biosafety benchmark at 62.4%, and that on HackerBench v0.3 it allows only 3.3% of risky dual-use prompts through while rarely blocking legitimate security work. Select cybersecurity partners get invite-only access to red-team capabilities. These are xAI’s claims about its own model, reported here as xAI’s claims, not as independent findings. What a reader can take from them without over-reading is that xAI treats safety as a headline feature of this release rather than a footnote, and that the safeguard stack is one of the things that distinguishes 4.7 from 4.6. The Acceptable Use Policy that actually governs what you may do with the model is covered in the final section.

One variant of Grok 4.7 deserves a warning of its own, because it is easy to buy by mistake. Grok 4.7 Fast is the same Grok 4.7 model running on faster infrastructure, at twice the standard token rates — $4.00 in and $12.00 out below 200,000 tokens, $6.00 and $18.00 above it — or 1.5 times the standard rates for long-context requests. It is available only in Cursor and Grok Build, billed through the plan there, and it is not available on the public xAI API. Grok Build’s free tier does not include it. If you are building on the API and you see a “Fast” option in a coding tool, you are looking at a different billing path from the API prices in the table above, and the two should not be mixed up when you estimate costs.

Finally, the knowledge cutoff. Grok 4.7’s training data runs to May 2026. Anything newer than that comes only through the search tools described in the real-time section — the web search and X search tools that fetch live information during the answer. The model itself does not know what happened after May 2026, and any article that implies otherwise is overstating the case. As of 14 November 2026, Grok 4.7 is the newest model xAI has announced. Grok 5 has not been released, and no Grok 4.8 has been announced on xAI’s own pages — reports of a 4.8 rollout are unverified third-party claims, and this guide does not present 4.8 as shipped.

Grok 4.6 and 4.5: Still Current, and Why They Exist

Two reliable workhorse machines side by side in a workshop, one slightly newer, both clearly in active use.

Grok 4.6 and Grok 4.5 are still live on the API, still supported, and still priced exactly the same as Grok 4.7 — $2.00 in, $6.00 out, 500,000-token context — which raises the obvious question: why would anyone use them when the newer model costs the same? The answer is availability, stability, and the specific strengths xAI built into each release. A model that is already wired into your workflow, tested against your prompts, and available on the platform you use is often worth more than a newer model that is not yet there, and xAI keeps the older flagships live precisely for that reason.

Grok 4.6, released 12 August 2026, builds on 4.5 with a focus on long-running agents and more ambitious interactive and visual work. It stays with complex tasks across many steps, gives stronger first passes on visual and interactive projects, and does more self-testing and verification on longer trajectories. Its launch table reported an AA Intelligence Index score of 61, matching GPT-5.6 Sol — again, xAI’s own table, not an independent audit. It shipped with twice the included usage in Grok Build and Cursor for its first week, a launch offer that has since expired. It is available on Amazon Bedrock from 18 August 2026 and in GitHub Copilot from 14 August 2026, which matters for developers whose infrastructure is already on those platforms: Grok 4.7 reached Bedrock later, on 28 September 2026, and platform availability still varies by model. If your code runs on Bedrock or Copilot, 4.6 may simply be the Grok that is there.

Grok 4.5, released 16 July 2026, was described by xAI at launch as its smartest model, built to excel at coding, agentic tasks and knowledge work. It was trained alongside Cursor, across tens of thousands of NVIDIA GB300 GPUs, and served at 80 tokens per second — xAI’s own launch-post claim, labelled as such. xAI also claims roughly twice the token efficiency of comparable leading models, its own claim again. It is the default model in Grok Build, and it can build Excel models with web research and multi-sheet formulas, and PowerPoint and Word documents with native shapes. That Office ability is a practical difference, not a benchmark: if your work is spreadsheets and decks rather than code, 4.5’s document skills are the reason to keep it in the lineup. Its cached input price is also slightly cheaper than the newer models — $0.30 per million cached tokens against $0.50 for 4.7 and 4.6 — a small saving that adds up on workloads that reuse the same context again and again.

The practical rule for choosing between the three is simple. Use Grok 4.7 for new work, because it is the newest, the most capable on xAI’s own benchmarks, and the same price. Keep Grok 4.6 or 4.5 when they are already integrated into a workflow that works, when the platform you use serves them and not yet 4.7, or when 4.5’s Office document abilities are the feature you actually need. Do not switch a working system to 4.7 for the sake of the version number, and do not stay on 4.5 out of inertia if you are starting fresh. The price is the same either way, so the decision is about fit and availability, not cost.

Grok 4.3 and 4.20: The Cheap, Long-Context Workhorses

A long, open road stretching to the horizon with a small, efficient car driving it — long context, low cost.

Grok 4.3 and the three Grok 4.20 variants are the cheap end of the family, and they are the models most developers should look at first, because the price difference against the flagship is large and the context window is twice as big. All of them cost $1.25 per million input tokens and $2.50 per million output tokens, with cached input at $0.20, and all of them offer a 1-million-token context window. Against Grok 4.7’s $2.00 and $6.00, that is 60% more on input and 140% more on output for the flagship — or, the other way round, the workhorses cost roughly a third as much on output. The Batch API discount of 20% applies to all of them, and their rate limits are lower — 37 requests per second and 10 million tokens per minute, against 150 and 50 million for 4.7 and 4.5 — which is the main practical constraint on heavy use.

Grok 4.3 is xAI’s own description, paraphrased, as a fast, reliable model with strong tool calling and instruction following. It is the all-rounder of the family: cheap enough for bulk work, long-context enough for whole documents and codebases, and capable enough for most agentic tasks. It is also the redirect target for all the retired model names — when an old model name is called, Grok 4.3 answers instead, as the final section explains — which makes it the default Grok in practice for a large amount of legacy code. Its reasoning efforts run from none through low, medium, high and xhigh, with low as the default, so it can be run cheaply with no reasoning at all for simple tasks, or dialled up when a task needs it.

Grok 4.20 comes in three variants, all at the same price and the same 1-million-token context: a reasoning variant, a non-reasoning variant, and a multi-agent variant. xAI’s model-page description of the family, paraphrased, calls it a high-performance model with industry-leading speed and agentic tool calling, and claims the lowest hallucination rate on the market with strict prompt adherence — xAI’s claim, labelled as such. The three variants exist so a developer can pick the behaviour the task needs without paying for the others: reasoning when the task needs step-by-step thought, non-reasoning when it needs a fast direct answer, and multi-agent when the task is split across parallel agents. Which variant you need depends on the architecture of your application, not on a ranking between them.

Who should use these models? Anyone doing bulk work, long documents, or budget API use. A million input tokens and 100,000 output tokens on Grok 4.3 costs about $1.25 plus $0.25 — $1.50 in total at short context — against $2.60 for the same session on Grok 4.7. On Grok Build 0.1 the same session is $1.20, but with a quarter of the context window. The worked examples in the pricing section below show the arithmetic for several session sizes. The rule of thumb is to start with 4.3 or a 4.20 variant, and move up to 4.7 only when the task demonstrably needs the flagship’s capability — longer reasoning, harder coding, more careful verification. Most tasks do not, and the savings from starting cheap are real money at volume. The one caution is the long-context doubling rule: a 1-million-token context invites prompts that cross the 200,000-token threshold, and once they do, the price doubles on the whole request. The cheap model used carelessly on very long prompts can cost more than the flagship used carefully on short ones.

Grok Build: The Coding Model and the Coding App

A developer's desk at night with code on a screen and a half-built application taking shape beside it, focused and quiet.

Grok Build is two things that share a name, and they need separating just as carefully as the chatbot and the models do. Grok Build 0.1 is a model: xAI’s intelligent coding model for agentic software engineering and workflow tasks, sold on the API at $1.00 per million input tokens and $2.00 per million output tokens, with a 256,000-token context window and cached input at $0.20. It is the cheapest text model on the API, and the only one in the current lineup that does not support the Batch API discount. Grok Build the app is the coding agent surface that the model powers: an environment where Grok writes code, tests it, runs it, and iterates on it, alongside the flagship models. Grok 4.7 is available in both Cursor and Grok Build, and Grok 4.5 is the default model in Grok Build. The model is the engine; the app is the workshop.

What the coding agent actually does is best understood through the abilities xAI claims for the models behind it. Grok 4.5, the default in Grok Build, can build Excel models with web research and multi-sheet formulas, and PowerPoint and Word documents with native shapes — the kind of office work that usually sits outside a coding tool. Grok 4.7, available in the same surface, is the model xAI describes as its most capable for coding, with longer reinforcement learning on multi-hour tasks and a habit of checking its own work. Grok 4.6 shipped with twice the included usage in Grok Build and Cursor for its first week, a launch offer that has expired. The Fast variant of Grok 4.7 — twice the standard token rates — is available only in Cursor and Grok Build, billed through the plan there, and not on the public API. A free tier for Grok Build exists, according to xAI’s launch posts, but its limits are not published in the pages read for this guide, so this guide does not state them.

Where Grok Build sits against the wider coding-assistant market is a question of the surface as much as the model. The model is trained alongside Cursor, and it ships inside Cursor as well as in its own app and through the API; it also reaches developers through GitHub Copilot, which ships Grok 4.6, and through the model routers and cloud platforms listed earlier. If you already work in Cursor, Grok is available where you already are. If you work in another coding assistant, the question is whether the model or the surface matters more to you. For a comparison of the coding assistants themselves — Copilot against Cursor, Claude Code and Windsurf — see our comparison of AI coding assistants, which covers the tools Grok Build competes with on their own terms.

The practical advice for developers is the same as for the model lineup generally: start with the cheapest model that can do the job. For most coding work that means Grok Build 0.1 on the API at $1.00 in and $2.00 out, or Grok 4.5 as the default inside the Grok Build app, and step up to Grok 4.7 when the task is genuinely hard — a long, multi-hour engineering task, a codebase the cheaper model cannot hold in its 256,000-token window, or work where a mistake is expensive. The 256,000-token context of Build 0.1 is the binding constraint: it is a quarter of Grok 4.3’s window, so large codebases need to be worked on in pieces or moved to a bigger-context model. And remember that the Batch discount does not apply to Build 0.1, so bulk, non-urgent coding work that could be batched is cheaper on Grok 4.3 than on the coding specialist.

How to Get Grok: grok.com, X, and the Apps

A phone, a laptop and a car dashboard arranged together on a table, each glowing softly — same assistant, many doors in.

Getting Grok is straightforward, and the free tier is genuinely free — but the limits of that free tier are not published in numbers, and this guide says so rather than inventing them. Go to grok.com on the web, or open the Grok tab inside the X app, or download the iOS or Android app. Sign in with X or with email. Your history syncs across web, iOS and Android, so a conversation started on your phone continues on your laptop. The free tier gives limited chat and agent requests — that is the wording on xAI’s own pricing page, and it is all the wording there is. No message counts, no image or video counts, and no context-window figure for the free tier is published on any official page read for this guide. Third parties variously claim free users get a limited version of an older model, but that is unverified, and this guide does not state which model free users get. If you need a guaranteed amount of usage, you need a paid tier, which is the subject of the next section.

Inside X, Grok is a tab in the app and on the website, and the free tier gives limited access there too. The integration is the point: Grok can search live posts on X while it answers, which is the capability the real-time section explains in detail. Outside X, the chatbot is the same assistant in a different frame. The iOS and Android apps are the mobile route, grok.com is the desktop route, and the history syncs between them. If you are new to AI chatbots generally, our beginner’s guide to using ChatGPT covers the habits — how to prompt, how to check answers, how to keep context in a thread — that transfer directly to Grok, even though the product is different.

Two access routes surprise people. The first is the car. Grok is a hands-free voice companion in Tesla cars: you choose a voice and a personality, from Storyteller to Unhinged, and it answers questions, takes vehicle commands — navigate a trip, make a phone call, search and play music, adjust the temperature, open the glove box — and answers questions about the car itself. Connectors and Grok Bot in the car manage your inbox, organise your calendar, sort files, chats and tasks, and hand off errands such as placing a coffee order or booking a reservation. Tesla’s own support page is the source for those features. The specific hardware and software requirements are not stated with version numbers here, because the page read for this guide was the feature summary; check Tesla’s support page for the current requirements before assuming your car qualifies.

The second is the office. A Grok add-on for Google Workspace exists — for Docs, Sheets, Slides and Gmail — covered in xAI’s own news post. Treat it as a real access route, and do not over-claim what it can do beyond that news post: it puts Grok beside Gemini inside Google’s apps, which is a notable piece of distribution for a model that competes with Google’s own. Developers have a different set of doors again: the API at console.x.ai, where one prepaid key covers text, vision, voice, images and video, and the distribution channels — Cursor, Grok Build, Amazon Bedrock, GitHub Copilot, Google Cloud, Microsoft Foundry, OpenRouter, Vercel and Cloudflare. Grok 4.6 ships in GitHub Copilot, and our GitHub Copilot explainer covers the tool it now runs inside. Per-platform availability varies, and this guide does not claim every model is on every platform — only that these are the channels xAI names.

The X Subscription: Free, Plus, Premium, Super, Ultra

A tiered stack of five blank membership cards fanned out on a dark surface, ascending in size — free to premium, no legible text.

On 12 October 2026, xAI replaced SuperGrok and X Premium for new sign-ups with a single bundle called The X Subscription: one plan that bundles Grok, Grok Bot, Cursor, and X Premium benefits. The prices below are from xAI’s own pricing pages, read 14 November 2026, in US dollars, excluding tax, and the prices are subject to change. Existing SuperGrok, X Premium and Cursor plans continue unless the user chooses to switch, and switching gives a prorated refund for unused time. But a new buyer cannot buy the old plans, and any guide that quotes SuperGrok or SuperGrok Heavy prices as current prices is wrong. Those are legacy prices for existing subscribers only.

PlanMonthlyAnnualIncluded usageGrok Bot?Headline inclusions
Free$0$0LimitedNoLimited chat and agent requests; standard X — post, reply, follow
Plus$8/mo$80/yr ($7/mo)Base poolNoGrok Chat, Voice, Imagine, Build; Grok API access; Cursor (Grok models only); X Premium checkmark, fewer timeline ads, creator tools, Advanced Analytics
Premium$30/mo$300/yr ($25/mo)4x PlusYesEverything in Plus, plus the latest Grok models, faster chat, AI images and videos, code hosting and AI code reviews, cloud agents and automations; X Premium+ benefits including ad-free scrolling on most surfaces, a preferred handle, and a higher reply boost; third-party models in Cursor billed separately
Super$100/mo$1,000/yr ($84/mo)15x PlusYesEverything in Premium, plus significantly higher usage across Chat, Imagine, Voice and Build, 1080p 15-second videos, lightning-fast replies, priority access at peak times
Ultra$200/mo$2,000/yr ($167/mo)50x PlusYesEverything in Super, plus the largest usage pool and earliest access to all products and models

Three mechanics decide whether a tier is worth its price, and all three are easy to miss. First, the usage pool is shared and weekly: one pool covers Grok, Grok Bot and Cursor together, and top-ups buy Additional Usage credits beyond what is included. Heavy use in one product eats the pool for the others. Second, third-party models in Cursor are not part of the included usage at all — they draw credits from the first use, billed separately. Third, Grok Bot is included on Premium, Super and Ultra only. Plus, at $8 a month, does not include it, and its usage draws from the seat’s shared weekly pool rather than a separate one. There is no standalone Grok Bot price. If the agent is the reason you are considering a subscription, the entry price is $30 a month, not $8.

Team plans follow the same tiers: Premium at $30 per seat, Super at $100 per seat, Ultra at $200 per seat, with the same usage per seat, and members can mix seat tiers. Plus is not available for teams. Enterprise plans are custom, arranged through the sales team, with single sign-on, data residency and volume pricing. You can cancel anytime, at least 24 hours before renewal, and the prices on the page are listed as subject to change — which is worth remembering, because xAI has already restructured its subscription once this year.

Which tier is actually worth paying for? Free is the right choice for trying Grok and for light, occasional use, with the honest caveat that its limits are unpublished and you will hit them without warning. Plus at $8 a month is the cheapest way to get the full chatbot — Chat, Voice, Imagine and Build — plus Cursor access to Grok models and the X Premium checkmark, and it is the natural tier for a regular user who wants Grok without the agent. Premium at $30 a month is where the product changes: four times the usage of Plus, the latest models, faster chat, and Grok Bot, the always-on agent. It is the tier for someone who will actually use the agent, and it is also the tier xAI marks as most popular. Super at $100 and Ultra at $200 are for heavy users and professionals who need the larger pools, the 1080p 15-second videos, priority access at peak times, and the earliest access to new products — fifteen and fifty times the Plus usage respectively. Most readers should not buy them. The same “which tier is worth paying for” question, asked about the rival product, is answered in our comparison of ChatGPT’s Free, Plus and Pro tiers, which is the natural cross-check before you commit to any single vendor’s subscription.

One paragraph for the old plans, because readers arriving from older articles will be confused by them. SuperGrok and X Premium — including the old X Premium tiers and the SuperGrok Heavy tier — continue for existing subscribers, and switching to The X Subscription is optional. If you already have SuperGrok, nothing in this guide says you must switch; the new subscription is the only option for new buyers. Do not quote the old prices as current, and do not assume a feature that belonged to an old tier — such as Grok 4 Heavy, the multi-agent parallel variant that was a SuperGrok Heavy feature — is still available as a purchasable tier. Under The X Subscription, no official page read for this guide says which tier, if any, gets Heavy, or whether it still exists as a separate mode, so this guide does not promise it on any tier.

Grok Bot: The Always-On Agent

A small robot teammate working alone at a desk in an empty office late at night, inbox and calendar glowing on screen.

Grok Bot, launched 11 August 2026, is the most important product in this guide after the chatbot itself, because it changes what Grok is: not an assistant you ask questions, but a teammate that does work. xAI’s own description calls it “your proactive AI teammate… an agent with access to a real computer in the cloud, so it can do all the things you can.” The Bots share a persistent cloud computer — a browser, a filesystem, a terminal — so work continues after you close your laptop. A Bot picks up where it left off, with your preferences, files and logins already in place. Show it once, and it can take it from there: that is xAI’s own phrasing for how Bots learn skills and routines from demonstration and then repeat them on a schedule.

What a Bot actually does is work across your apps and websites, including platforms with no clean API or MCP connection, because it has a browser and can use the interface the way a person would. It supports custom tooling through any MCP server or API, and its connectors include Slack, Notion, Google Drive, Microsoft Teams, Gmail, Calendar and Granola. Chat is the main interface, and voice calls and voice notes work too; Bots work in Slack and on X as well as on desktop and mobile. The marketplace of pre-built Bots shows the shape of the product: Overheard watches Reddit, Hacker News, news and X for mentions of a topic; Tradbot is an email and calendar assistant; and there are Bots for meeting recaps and pitch-deck coaching. Our own news coverage reports that Grok Bot has left beta and moved into team meetings — see our report on Grok Bot leaving beta — and the product documentation describes the current, shipping product rather than a beta.

The difference from the chatbot is the difference between answering and doing. The chatbot answers a question and stops. A Bot takes a task — watch for mentions, prepare a meeting brief, sort an inbox, draft a deck — and carries it through across apps, bringing the work back to you to approve, edit or send back. Sensitive steps stop for approval: that is the control xAI describes in its documentation, and it is the right one to emphasise, because an agent with access to your email, calendar and files is a different trust proposition from a chatbot in a browser tab. Before you connect a Bot to anything sensitive, read the approvals and permissions in xAI’s own Grok Bot documentation and decide what you are comfortable delegating. This guide does not state the specific security-boundary claims that circulate in third-party coverage of the Bot, because they were not verified against the documentation extract read for it.

Access is the other thing to get right. Grok Bot is included in The X Subscription on Premium, Super and Ultra only — $30, $100 and $200 a month. It is not included on Plus, and there is no standalone Grok Bot price. Its usage draws from the seat’s shared weekly pool, the same pool the chatbot and Cursor draw from, not a separate one. That means a Bot running scheduled routines all week consumes the same pool your own chatting draws from, and heavy Bot use is one of the fastest ways to need a top-up or a higher tier. For a reader deciding whether Premium is worth $30 a month, Grok Bot is usually the deciding feature: if you have recurring work across apps that an agent could do, Premium is the cheapest way to get it; if you only want to chat with Grok, Plus at $8 a month is the better buy.

The Real-Time Advantage, Honestly Tested

A newsroom-style wall of live tickers and breaking feeds, motion-blurred, with one calm figure reading them.

This is the section the whole guide stands or falls on, because the real-time claim is Grok’s headline and also its most overstated feature. The precise truth is this: Grok’s model does not know what is happening right now. Its training data ends in May 2026. What Grok can do is fetch live information during an answer, using search tools, and show you the sources. Without those tools, xAI’s own models page states in substance that Grok has no knowledge of current events beyond its training data. The advantage is real, but it is a tooling advantage, not a knowledge advantage, and the distinction matters every time you rely on an answer about something that happened this week.

The tools are server-side, and on the API they are priced per use. Web search — searching the internet and browsing web pages — costs $5 per 1,000 calls. X search — searching X posts, user profiles and threads — costs $5 per 1,000 posts and $10 per 1,000 profiles. Code execution in a Python sandbox costs $5 per 1,000 calls. File attachment search costs $5 per 1,000 calls. Searching uploaded document collections costs $2.50 per 1,000 calls. Image generation uses the image API rates, and image and video understanding are billed on tokens with no per-call fee. X search is billed per item fetched, not per call: every post returned, including parent and quoted posts, counts toward the post rate, and every profile returned counts toward the profile rate. Reasoning tokens, image tokens and cached tokens are all billed. Those prices are the API prices; the chatbot’s free and paid tiers ration search behind usage pools that xAI does not publish in numbers, so a chatbot user cannot price a search the way an API user can.

What the tools buy is a specific kind of coverage. X search is a first-party tool — xAI owns X — so Grok can search the live post stream directly, with citations, in the middle of an answer. That is genuinely different from a chatbot that can only search the web: breaking news, a trend forming on X, a niche conversation that has not reached the news yet, and the reaction to an event as it happens are all places where live X data beats a web search. The web search tool covers the rest: news sites, documentation, and anything published on the open web. Together, with citations, they let Grok answer “what happened today” in a way a model relying on training data alone cannot. Multi-agent mode, described in the features section below, can run parallel agents over the same question and merge their work into one cited answer, which is the closest Grok comes to the old multi-pass research workflow.

Where the advantage does not hold is equally important. Search results are only as good as their sources, and xAI itself describes Grok 4.7 as having minimal hallucinations — less, not never. A live search can return a wrong post, a rumour, or an outdated page, and the model can still present it confidently. The chatbot’s usage pools ration how much live searching a free or Plus user can do, and the API user pays per call and per post, so heavy real-time use is metered in a way that a plain chat is not. And for most questions, live data adds nothing: a coding problem, a document summary, a piece of writing, or an explanation of a concept does not get better because the model can see today’s posts. The honest rule is to reach for Grok’s live search when the question is about something happening now, something being discussed on X, or something whose answer changed recently — and to treat its citations the way you would treat any search result: check the source, especially before you repeat the claim.

What Grok Can Do: Voice, Imagine, Files, Memory and More

A creative studio table with a microphone, a camera, sketches and documents spread out — voice, images, files, memory in one place.

The feature list on xAI’s product page is long, and it is worth going through in plain language, because the individual features are what decide whether Grok fits your daily work. Chat is the core: step-by-step reasoning you can follow and verify, with threads and follow-ups that keep context intact. Multi-agent mode runs parallel agents on the same question, each showing its work, and merges them into one cited answer. Search covers the live web and X, with citations, as the previous section explains. Voice is a natural conversation mode, which xAI describes as having sub-second latency — its own claim. File and PDF analysis lets you upload documents and ask questions about them. Vision handles screenshots, photos and diagrams. Memory works across chats, remembering your preferences and past conversations. Canvas is the long-form writing surface with inline suggestions. Custom instructions set how Grok should answer you by default. Shareable conversations produce a public link to a thread. And the whole thing works in more than 30 languages, on the web, iOS and Android, with synced history.

Imagine is the media layer, and it is the feature with the clearest pricing. Inside a chat thread, you can generate and edit images and videos with follow-up prompts: text to image and text to video in one thread, up to 2K resolution and 15-second videos, with 720p for video on the product page. On the API, the image model grok-imagine-image-2.0 costs $0.01 per image in, and $0.04 to $0.08 per image out depending on resolution — 1K, 1.5K or 2K — and quality, Low or Medium. The original grok-imagine-image model is cheaper: $0.002 in and $0.02 out for 1K or 2K. Video is priced per second: the 1.5 model costs $0.08 a second at 480p, $0.14 at 720p and $0.25 at 1080p, with a $0.01 per image input charge; the original video model costs $0.05 a second at 480p and $0.07 at 720p, with small input charges and no 1080p price listed for it; and the 1.5-lite model costs $0.02, $0.03 and $0.14 a second. A 15-second 1080p clip on the 1.5 model costs 15 times $0.25 — $3.75. The same clip at 720p is $2.10, and at 480p it is $1.20. Voice on the API is priced separately: speech-to-speech on grok-voice-3.0 costs $0.08 a minute — $4.80 an hour — plus $0.004 per text input, with the older grok-voice-2.0 deprecated at the same rate for reference only. Speech to text costs $0.10 an hour over REST and $0.20 an hour streaming, and text to speech costs $15.00 per million characters.

Video generation is adjacent to a whole market of specialist tools, and if Imagine is the feature that interests you, the comparison is worth reading before you spend money on it: see our comparison of AI video generators, which covers Runway, Sora, Veo, Kling and Pika on their own terms. Imagine’s advantage is that it lives inside the chat thread, so a video is one follow-up prompt away from the conversation that asked for it; its limitation is that it is a generalist tool inside a chatbot, not a specialist video studio.

Two naming corrections close this section, because they are the kind of detail that wastes a new user’s time. The old feature names DeepSearch and Think mode are not current. xAI’s product page presents the capabilities as Search — searching the web and X live — Deep reasoning — step-by-step thinking you can follow and verify — and Multi-agent mode. The old names describe the same underlying capabilities, and if you meet them in an older article, read them as history: the buttons were folded into the mode selector, and the current terms are the ones above. Similarly, the product page still carries stale copy in one place — it says “Upgrade to SuperGrok” — while the pricing pages say new customers buy The X Subscription. Follow the pricing pages. The product page is behind; the money pages are current.

Grok vs ChatGPT vs Claude: An Honest Comparison

Three chess pieces of different designs facing each other on a board in dramatic side light — a deliberate contest, not a race.

No honest comparison of Grok with ChatGPT and Claude can be a single verdict, because the three products are built around different strengths, and because this guide verifies Grok only. Every price, model name and capability stated for ChatGPT or Claude in this section would need a source of its own, and this brief does not supply one — so the comparison below is structural: how each product is accessed, where its live data comes from, what its agent looks like, and how it handles coding. Where a rival figure matters, the link goes to the site’s own verified guide to that product rather than to a number quoted here. That restraint is deliberate. A comparison that invents rival prices to make a point is worse than no comparison at all.

On access and price shape, the three products rhyme but do not match. Grok’s access is The X Subscription — Free, Plus at $8, Premium at $30, Super at $100 and Ultra at $200 a month — a single bundle that covers the chatbot, the agent, Cursor, and X Premium benefits, with one shared weekly usage pool. ChatGPT’s tiers ask the same “which tier is worth paying for” question in a different shape, and the site’s own comparison of those tiers is the place to check the rival’s structure rather than taking a number from a Grok roundup. For the OpenAI side of the market generally — the company and the model family behind ChatGPT — see our explainer of OpenAI. For Claude’s model family, the sibling guide to this one, see our Claude models explained guide, which covers that family the way this guide covers Grok’s. The structural point is that Grok’s subscription is the only one of the three bundled with a social network and a coding tool in the same price, and that bundling is either a bargain or irrelevant depending on whether you use X and Cursor.

On real-time data, Grok’s edge is genuine and narrow. It is the only one of the three with first-party access to a live social post stream: X search runs against the posts on X itself, because xAI owns X, and it is billed per post fetched on the API. The other two products search the web, and web search covers news and published pages well. The difference shows up on questions about what is being said right now, on X, before it becomes news — a breaking story’s first hour, a trend forming, a niche community’s reaction. It does not show up on questions about what happened yesterday and was properly reported, where all three can search the web and cite sources. And the edge has a cost: Grok’s model cutoff is May 2026, so without its search tools it knows nothing newer than that, and its search is metered behind usage pools on the chatbot and per-call charges on the API. Treat the real-time claim as a tool you switch on for a specific kind of question, not as a property of the model.

On agents, the three products are at different stages of the same idea. Grok Bot is an always-on agent with its own persistent cloud computer, working across Slack, Notion, Gmail, Calendar and the rest, with skills it learns from demonstration, scheduled reruns, and approvals on sensitive steps. It is included from the $30 Premium tier of The X Subscription. The rival agent products — ChatGPT’s agent mode and Claude’s Cowork — are described, not ranked, here: all three aim at the same job, doing work across apps rather than answering questions, and which one fits depends on which apps you already live in and which subscription you already pay for. The same applies to coding. Grok Build and Cursor, with Grok 4.7 and the Grok Build 0.1 coding model, compete with GitHub Copilot — which itself ships Grok 4.6 — and with Claude Code. The model matters less than the surface: the coding assistant you already use, the codebase it already knows, and the workflow it already fits.

On personality, one difference is visible without any benchmark at all. In Tesla cars, Grok offers personalities from Storyteller to Unhinged, and the chatbot’s voice is generally more conversational and less guarded than the neutral tone its rivals default to. That is a matter of taste, and taste is a legitimate reason to prefer a product: some readers will like a chatbot with a point of view, and others will find it a distraction. It is not a capability difference, and it should not be scored as one.

Task-based recommendations, then, rather than a winner. Choose Grok for everyday chat when you already live on X and want the chatbot where the conversation already is, and when live X data matters to the questions you ask. Choose it for breaking news and trends, where its X search is the best tool of the three. Choose it for coding when you work in Cursor or Grok Build and want the Grok models where you already code. Choose it for agents when the $30 Premium tier’s Grok Bot fits work you actually repeat across apps. Choose ChatGPT or Claude instead when your work lives in their ecosystems, when you need their particular agent or coding tools, or when the X bundle’s extras — the checkmark, the timeline, the Cursor seat — are worth nothing to you. And whichever you choose, do not choose on a benchmark table: xAI’s own launch table in the Grok 4.7 section shows Grok 4.7 leading some benchmarks and trailing others, and the rival vendors’ tables show the same pattern in reverse. Benchmarks are a vendor’s chosen evidence, not a verdict.

Limits, Safety, Retired Models — and How to Pick

A signpost at a crossroads in a misty landscape, several paths leading in different directions, with a small warning fence beside one path.

Every model guide should end with the limits, and Grok’s are worth stating plainly. The Acceptable Use Policy, current version effective 14 August 2026 and replacing the 26 June 2026 version, applies to consumers, developers and businesses alike. It bans illegal and harmful use, using the models to train a competing service, evading rate limits, and reselling raw access, among other things. Violations can lead to action up to suspension or termination of the account. On the API, usage that breaches the policy can still be billed — the meter runs whether the use was allowed or not — so the policy is not only a legal document but a billing risk. Read it before you build anything on the API, and before you connect Grok Bot to sensitive accounts: an agent with access to your inbox and calendar is a different commitment from a chatbot in a browser tab, and the approvals mechanism described in the Grok Bot section is the control that makes that commitment manageable.

The retired models are the second limit, and they matter to anyone with old code. On 15 May 2026 at noon Pacific Time, xAI retired from the API: Grok 3, the original Grok 4 (grok-4-0709), the fast reasoning and non-reasoning variants of Grok 4 and Grok 4.1, grok-code-fast-1, and grok-imagine-image-pro. Requests to those retired names still work, but they are served by Grok 4.3 and billed at Grok 4.3 rates — reasoning IDs map to Grok 4.3 with low effort, non-reasoning IDs to Grok 4.3 with no reasoning, and grok-code-fast-1 maps to Grok Build 0.1. xAI recommends updating code to name the new model explicitly rather than relying on that silent redirection, and the advice is sound: a redirect that works today is a dependency you do not control. The older history is worth one line for context. Grok 4 arrived in July 2025, Grok 4.1 on 17 November 2025, and Grok 4.1 Fast on 19 November 2025 with a 2-million-token context window; the 4.1 launch claimed a hallucination rate down from about 12% to about 4% and a public leaderboard score of 1483 — 2025 claims about a retired model, mentioned here only as timeline context, not as current facts. The voice model grok-voice-2.0 is also deprecated; grok-voice-3.0 is current.

The third limit is the pricing trap already described in the lineup section, and it is worth repeating here because it is the mistake most likely to cost a reader real money: once a prompt reaches 200,000 tokens, the higher rate applies to all tokens in the request, not just the part above the threshold. A 1-million-token context window invites long prompts, and long prompts double the price. Keep prompts under the threshold where you can, and when you cannot, price the request at the doubled rate before you send it. The worked examples make the arithmetic concrete. A session of 1 million input tokens and 100,000 output tokens at short context costs about $2.60 on Grok 4.7 — $2.00 plus $0.60 — about $1.50 on Grok 4.3, and about $1.20 on Grok Build 0.1. The same session at long context costs about $5.20 on Grok 4.7, because the whole request doubles. Run the same session through the Batch API at 20% off and the Grok 4.7 short-context session costs about $2.08. A 15-second video costs $3.75 at 1080p, $2.10 at 720p and $1.20 at 480p on the Imagine Video 1.5 model, and an hour of speech-to-speech voice costs $4.80. None of these are large numbers on their own; at volume, the choice of model is the whole budget.

How to pick, then, by task. For everyday chat, start free at grok.com or inside X, and pay $8 a month for Plus when the free tier’s unpublished limits start getting in your way. For breaking news, trends, and anything being discussed on X right now, Grok is the strongest choice of the three major chatbots, because its X search is first-party — but check the citations before you repeat what it says. For coding, use Grok Build 0.1 on the API or Grok 4.5 in the Grok Build app, and step up to Grok 4.7 for the hardest tasks; the coding-assistant comparison linked in the Grok Build section covers the rival surfaces. For agents and recurring work across apps, Premium at $30 a month is the entry point, because that is where Grok Bot starts, and it is the tier to evaluate first. For API bulk work and long documents, start with Grok 4.3 or a Grok 4.20 variant and their 1-million-token context, watch the 200,000-token doubling threshold, and use the Batch discount where the work is not urgent. For the model-picker method itself — the framework for choosing between AI models in general, beyond Grok — see our guide to how to choose an AI model, which sets out the decision process this article applies.

On maintenance: this guide was checked on 14 November 2026, and Grok is a product that moves fast. Grok 4.7 is the flagship today, and the next release could change the lineup, the prices, or both; xAI’s own pricing pages carry the warning that prices are subject to change, and the company has already restructured its subscription once in 2026. The free tier’s limits are unpublished, the per-tier model mapping is unpublished, and Grok 4.8 and Grok 5 have not been announced — any of those facts could change, and a reader who reaches this guide months after publication should re-check the prices on xAI’s pricing page and the lineup on xAI’s model catalogue before spending money on the strength of it. The links are in the Sources section below, and they are the pages this guide was built from. A model guide is a snapshot, not a permanent truth, and the honest ones say so on the page. This one does.

Sources

Checked 14 November 2026 against xAI’s official documentation, pricing pages and announcements. All facts in this guide come from the sources below, read in full on that date. No third-party roundup was used as a source for any price, limit or capability; where third-party pages disagreed with xAI’s own pages, xAI’s pages were followed. The x.ai/grok product page still says “Upgrade to SuperGrok” in one place; the pricing pages, which say new customers buy The X Subscription, were followed instead.

  • xAI model catalogue and pricing overview — https://docs.x.ai/developers/models
  • xAI API pricing page — https://docs.x.ai/developers/pricing
  • Grok 4.7 model page — https://docs.x.ai/developers/models/grok-4.7
  • Grok 4.5 model page — https://docs.x.ai/developers/models/grok-4.5
  • Grok 4.3 model page — https://docs.x.ai/developers/models/grok-4.3
  • Grok 4.20 Reasoning model page — https://docs.x.ai/developers/models/grok-4.20-0309-reasoning
  • Grok Build 0.1 model page — https://docs.x.ai/developers/models/grok-build-0.1
  • Grok Imagine Image 2.0 model page — https://docs.x.ai/developers/models/grok-imagine-image-2.0
  • Grok 4.7 launch announcement — https://x.ai/news/grok-4-7
  • Grok 4.6 launch announcement — https://x.ai/news/grok-4-6
  • Grok 4.5 launch announcement — https://x.ai/news/grok-4-5
  • xAI product page — https://x.ai/grok
  • The X Subscription pricing — https://x.ai/pricing
  • The X Subscription pricing, docs — https://docs.x.ai/x-subscription/pricing
  • Grok Bot documentation — https://docs.x.ai/grok-bot/overview
  • Tesla Grok support page — https://tesla.com/support/grok
  • xAI Acceptable Use Policy — https://x.ai/legal/acceptable-use-policy

Share to: , and

Faheem R. Khattak

Senior automation engineer and AI marketing journo covering LLMs, algorithms, analytics, core updates, Bing search, and technical SEO.

0 Comments

Leave a Reply

Optional Take Tag:

🛡️ Spam protection: Honeypot + Time-Trap Active

This site uses Akismet to reduce spam. Learn how your comment data is processed.