AI Mindset · Model Cheatsheets
Grok

The Fast, Real-Time Operator With an Edge

Grok 4.6, released August 12, 2026, is xAI’s current flagship and supersedes Grok 4.5 without moving the numbers around it: the same $2 per 1M input and $6 per 1M output below 200,000 tokens, the same 500k context window and the same February 1, 2026 knowledge cutoff, with a higher tier above the 200,000-token threshold that xAI documents as “$4 / $1 / $12 above.” Since then the story has been distribution rather than capability. Grok 4.6 reached Amazon Bedrock on August 19, the Gemini Enterprise Agent Platform on August 21 and Microsoft Foundry on August 26, so xAI’s flagship now runs inside the clouds its rivals own. Grok Bot outgrew its three-tier beta and now ships with eight subscription plans, and it works with X as of August 29. Grok Build is open source; add-ins put Grok inside Excel, Outlook and Google Workspace; Imagine generates image, video and synchronized audio. Two corrections matter more than any launch: grok-imagine-image-quality retires on November 2, 2026, and the claim in the last edition of this guide that no model card was ever published for Grok 4.5 was simply wrong.

Verified September 3, 2026Grok 4.6 · August 12$2/$6 below 200k tokensModel cards exist for 4.6 and 4.5Grok Bot on eight plansBedrock, Gemini Agent Platform, FoundryImagine quality tier retires November 2
1 / Meet Grok

The colleague closest to the live conversation

Grok is useful when the question changes by the hour. Its differentiator is not merely a more irreverent tone. It is the combination of live X context, frontier reasoning, coding agents and media production.

Personality

Direct, fast and current

Grok is comfortable taking a position and responding to live events, which makes it useful for sense-making and risky when confidence outruns evidence.

  • Real-time social context
  • Strong engineering orientation
  • Less polished corporate default tone
Deploy it for

Live signal and high-velocity execution

Breaking narratives, market reaction, technical work, rapid prototyping, coding agents and integrated image/video experimentation.

  • What is happening now
  • What developers are saying now
  • What can be built or rendered quickly
Choose Grok when

Recency or candor changes the answer

Use Grok when yesterday’s summary is inadequate, the social response is part of the evidence or a blunt stress test improves the work.

  • Live pulse
  • Direct critique
  • Fast action through Build or Imagine

Live signal or verified fact?

Grok’s most important workplace habit is keeping these two products separate.

What are people saying?
Use X as a live signal

Map the dominant narratives, influential accounts, disagreements and velocity of attention.

  • Directional, not representative
  • Useful for emerging language
  • Expect manipulation and selection bias
Narrative map
Map the three dominant narratives about this announcement on X. Show who is driving each one, what evidence they cite and where the narratives disagree. Do not decide which is true yet.
Verification pass
Now verify the factual core against the company release, filing and any other primary source. Separate confirmed facts, disputed interpretation and unsupported social claims.
2 / What’s Current

The flagship stopped moving and the distribution started

Grok 4.6 has been the flagship since August 12, 2026 at unchanged price, context window and knowledge cutoff, and nothing has replaced it. What changed in the three weeks since is where it runs: Amazon Bedrock on August 19, the Gemini Enterprise Agent Platform on August 21 and Microsoft Foundry on August 26. Grok Bot moved from a three-tier early beta to eight subscription plans and gained an X integration. And this edition opens with a correction: the previous edition said no model card was ever published for Grok 4.5. A card exists, it is linked from xAI’s own safety page, and it is dated before the check that missed it.

$2 / $6
API input/output price per million tokens below 200,000 tokens; xAI documents “$4 / $1 / $12 above”
500k
Grok 4.6 context window, unchanged from Grok 4.5
Feb 1, 2026
Grok 4.6 knowledge cutoff, unchanged from Grok 4.5
3 clouds
Third-party platforms carrying Grok 4.6: Amazon Bedrock (August 19), the Gemini Enterprise Agent Platform (August 21) and Microsoft Foundry (August 26)
8 plans
Subscription plans that now include Grok Bot, up from three at the August 11 launch
Nov 2, 2026
Retirement date for grok-imagine-image-quality; requests then move to grok-imagine-image-2.0 at low quality

Grok 4.6 Flagship · August 12

Released August 12, 2026 and still xAI’s current model on September 3. Price, context and knowledge cutoff are unchanged from Grok 4.5: $2 per 1M input and $6 per 1M output below 200,000 tokens, 500k context, February 1, 2026 cutoff. Above the 200,000-token threshold xAI’s price table adds a higher tier it documents as “$4 / $1 / $12 above.” xAI describes 4.6 as “Our flagship model for code and everything else.”

  • Pin grok-4.6 rather than a moving alias
  • Long-context requests cross into a higher price tier at 200,000 tokens
  • Long context pricing bills the long-context rates for all tokens in the request
  • Model card dated August 12, 2026, revision 2026-08-17

Grok 4.5 Prior generation

Superseded on August 12, 2026. xAI’s strongest model until then, trained alongside Cursor on NVIDIA GB300s, and the subject of Elon Musk’s vendor positioning as an Opus-class model that is faster, more token-efficient and lower cost. It does have a model card, dated July 14, 2026 with a July 20 revision, linked from the safety page and hosted on media.x.ai.

  • 500k context, February 1, 2026 knowledge cutoff
  • Vendor claims: about 80 tokens/sec and about 4.2x token efficiency
  • A model card exists, correcting the previous edition of this guide
  • Vendor positioning: benchmark any migration yourself

Three clouds in seven days Distribution

Grok 4.6 reached Amazon Bedrock on August 19, 2026, alongside Grok Build on web and mobile. It reached the Gemini Enterprise Agent Platform on August 21 and Microsoft Foundry on August 26. That is xAI placing its flagship inside the model catalogs its largest rivals operate, which changes the procurement question: many organizations can now buy Grok through an existing cloud agreement without ever contracting with xAI.

  • Bedrock August 19, Gemini Enterprise Agent Platform August 21, Microsoft Foundry August 26
  • Expect Grok to appear in cloud model catalogs your teams already have access to
  • Check which entity holds the data terms on each platform

The Cursor connection Strategic · same owner

The sequence matters more than any single date. Cursor and xAI announced a partnership in April 2026; on July 8, 2026 they shipped together, xAI saying Grok 4.5 “was trained alongside Cursor,” on NVIDIA GB300s, and Cursor putting it as “Today we are releasing Grok 4.5 together with SpaceXAI, our most intelligent model.” On August 14, 2026 Cursor posted that it “has officially been acquired by SpaceX,” completing “the acquisition process that started in April, when we announced our partnership with SpaceXAI to accelerate our model training efforts.” No terms and no valuation were disclosed. Because SpaceX also acquired xAI, on February 2, 2026, Cursor and xAI are now sister companies inside one group, so the training relationship described here is intra-group rather than arm’s-length. Nothing further has been published first-party since August 14: targeted searches on September 3, 2026 found only plan bundling. Record that as not found rather than as nothing having happened, because the xAI news index is known to omit real articles.

  • Cursor has been SpaceX-owned since August 14, 2026, with no terms disclosed
  • Partnership announced April 2026, joint Grok 4.5 release July 8, acquisition completed August 14
  • No further first-party detail found as of September 3, 2026, which is not the same as none existing
  • A buyer evaluating Grok or Cursor should know both products now share an owner

Grok Bot Eight plans · works with X

Launched August 11, 2026 in early beta for three tiers. On August 26 xAI published “Grok Bot is now included with more plans,” extending it to SuperGrok, SuperGrok Plus, SuperGrok Heavy, Cursor Pro, Cursor Pro+, Cursor Ultra, and Cursor Teams Standard and Premium. On August 29 “Grok Bot now works with X” added an X integration that xAI calls “the first version of this integration.” Neither post uses the word beta, but no post says it left beta either, so widen the availability claim without asserting general availability.

  • Eight named plans, up from three
  • X integration since August 29, described by xAI as a first version
  • No first-party statement that it has left beta: do not claim GA

Grok Build Coding agent

A terminal CLI agent for SuperGrok and X Premium Plus subscribers, and available on web and mobile since August 19, 2026. Plan mode lets you approve, annotate or rewrite the plan before execution; larger tasks delegate to parallel subagents in their own worktrees.

  • Approve the plan before it runs
  • Parallel subagents and worktree integration
  • Picks up repository conventions on start

EU availability Resolved

Grok 4.5 was not available in EU products or the API console at launch. That gap closed on July 17, 2026, when the developer release notes recorded Grok 4.5 as available to EU users in the API console.

  • History, not a current limitation
  • Retire any "not available in the EU" language
  • Still separate product and API availability by region

Skills and connectors

Grok can create documents, decks and spreadsheets and connect to external systems through reusable skills and integrations.

  • Persistent expertise
  • Business workflow automation
  • Permission-aware enterprise use

Office and Workspace add-ins

Grok works inside Excel, Outlook, Sheets, Slides and Docs, extending its work story well beyond X and grok.com. See the Workplace Add-Ins section for the July releases.

  • Draft and revise documents
  • Turn outlines into slides
  • Keep tone and source review explicit

How Grok 4.5 rolled out, kept as a worked example

Grok 4.5 is the prior generation now, but its rollout is the clearest first-party case of why a single release date does not describe what shipped. Name the stage that matches the surface you are buying, and expect Grok 4.6 to land on surfaces at its own pace too, as the August cloud releases showed.

API
API availability

The xAI developer release notes date Grok 4.5 availability on the API to July 8, 2026. This is the earliest first-party date on record for that model.

  • Developer surface first
  • Source: docs.x.ai release notes
  • Not yet a consumer release

Choose the Grok surface

The right product depends on whether the job is intelligence, engineering, creation or enterprise context.

Research and live signal
Grok app or grok.com

Use the main assistant for current search, conversation, analysis and rapid creation.

  • Search and X context
  • Model selection
  • Skills and connectors
3 / Grok Build

A coding agent designed for speed and sustained work

Grok Build moves the product from advice into implementation. The /goal command, session dashboard and plugin system make the relevant unit of work an outcome rather than a sequence of terminal prompts. Since July 15, 2026 the agent itself is open source, which changes what a security review can actually inspect.

A reliable /goal workflow

Long-running autonomy needs a short, inspectable contract.

Define
Describe the end state

Name the feature or repair, the affected system, non-negotiable constraints and evidence required for completion.

  • Use repository-specific instructions
  • State forbidden changes
  • Include tests

Open source July 15

xAI open-sourced Grok Build at github.com/xai-org/grok-build: the agent loop, developer tools, the terminal interface and the skills, plugins, hooks, MCP and subagent systems. The announcement page does not state a license, so confirm the terms in the repository before you depend on it.

  • Enables local-first use with your own inference
  • Read the agent loop instead of trusting a description
  • License not stated on the announcement, so verify it

Web and mobile August 19

Grok Build arrived on web and mobile on August 19, 2026, the same day Grok 4.6 reached Amazon Bedrock. A coding agent that is no longer terminal-only is a governance change as much as a convenience: the people who can start a long-running agent are now a wider group than the people with a configured CLI.

  • Not terminal-only any more
  • Decide who may launch agents from a phone
  • Keep repository permissions as the real control

/goal Long running

A command for assigning a sustained outcome rather than manually driving each agent turn. This is the CLI’s long-running execution mode and is distinct from consumer Automations.

  • Good for bounded projects
  • Requires completion criteria
  • Use stop conditions

Agent dashboard

Manage many coding sessions, see progress, respond to questions and dispatch additional work.

  • Parallel engineering
  • Central supervision
  • Avoid duplicate changes

Plugin marketplace

Extend Grok Build with packaged tools and workflows inside the coding environment.

  • Review plugin trust
  • Pin required versions
  • Keep permissions narrow

Composer 2.5

A specialized coding model option within Grok Build for fast software tasks.

  • Use the model menu
  • Benchmark on your repository
  • Do not assume the flagship is always the best route

External coding integrations

Grok subscriptions or models can work through environments such as Warp, Kilo Code and OpenCode.

  • Know which tool owns execution
  • Check data terms
  • Keep configuration documented

Grok Build API

A public-beta API path allows developers to integrate the coding model into their own workflows.

  • Beta stability
  • Programmatic control
  • Requires evaluation and observability
Feature goal
Implement organization-wide audit logging for authentication changes. Follow existing architecture, add migration-safe storage, update API and admin UI, write abuse-case tests, and finish only when all checks pass.
Repository review
Inspect this repository for the highest-risk authentication and authorization flaws. Do not change code yet. Rank findings by exploitability and show the exact evidence behind each one.
Parallel plan
Split this migration into independent workstreams that can run safely in parallel. Identify shared files, sequencing constraints and the final integration test before dispatching sessions.
Completion report
Report only: files changed, user-visible behavior, tests run, security implications, unresolved issues and the exact review I should perform before merge.
4 / Live Intelligence

Use X as a sensor, not a source of universal truth

The X firehose can reveal emerging language, narratives and reaction faster than traditional sources. It can also magnify bots, coordinated campaigns, selection bias and confident misinformation. Since August 29 Grok Bot works with X directly, which puts an agent on the same surface.

Early signal

Narrative emergence

See which explanations and phrases are spreading before they appear in formal analysis.

Market pulse

Real-time sentiment

Track reaction to products, earnings, policy and events as participants discuss them.

Expert network

Practitioner commentary

Find engineers, researchers, operators and witnesses closest to an event.

Distribution

Attention dynamics

See who is amplifying a claim and how quickly it moves through communities.

Bias

Not representative

X users and visible posts are not a census of customers, voters or professionals.

Manipulation

Coordinated activity

Bots, incentives and campaigns can create the appearance of consensus or urgency.

Verification

Primary records still win

Use releases, filings, transcripts, source data and direct evidence for factual claims.

Time

Recency decays quickly

Capture the date and time because a real-time answer can become stale within hours.

How much confidence should the signal earn?

Move the answer only when evidence improves.

One viral post
Very low confidence

Treat it as a lead. Capture the claim and look for the direct source.

  • High manipulation risk
  • No representativeness
  • Do not publish as fact
5 / Imagine

Image, video, speech and audio are becoming one creative stack

Imagine Video 1.5 generates motion, physics, ambience, sound effects and speech in the same pass, and the image side moved to 2.0 on August 7, 2026 before gaining an API update on August 28. That shortens experimentation, but it also joins several quality risks into one output. Two dates belong on a calendar: the voice alias repriced itself on August 5, 2026, and grok-imagine-image-quality retires on November 2, 2026.

Imagine image API, August 28 Parameter change

The August 28, 2026 release notes add an auto quality parameter on grok-imagine-image-2.0 and move the default from medium to auto. Editing now accepts up to 5 source images, previously 3, and two aspect ratios were added: “21:9 (cinematic widescreen) and 5:2 (wide banners).” A changed default is a silent change in output characteristics, so re-run your reference prompts rather than assuming parity.

  • Default moved from medium to auto
  • Up to 5 source images for editing, up from 3
  • New 21:9 and 5:2 aspect ratios

Grok Imagine Image Quality Retires November 2

The dedicated quality endpoint stops being its own model on November 2, 2026. Requests continue to be served, by grok-imagine-image-2.0 with quality set to low, at a lower per-image price. Cheaper output at lower quality with no error is the kind of change that reaches a brand review before it reaches an engineer.

  • Migrate deliberately rather than being migrated
  • Set the quality parameter explicitly on 2.0
  • Add a visual regression check before November

Grok Voice Think Fast 2.0 Live since August 5

Announced July 29, 2026 at $0.08 per minute of audio plus $0.004 per text input, and what grok-voice-latest has served since August 5, 2026. The documentation states the same audio rate as $0.08 per minute, or $4.80 per hour.

  • Priced per minute of audio, not per token
  • The default for anyone still calling the alias
  • Budget the hourly rate for anything conversational

Grok Voice Think Fast 1.0 Deprecated

The previous voice model, still marked [Deprecated] in the xAI docs at $0.05 per minute of audio with the same $0.004 per text input. It is still callable by name, but a deprecated model is a countdown, not a strategy.

  • Naming the version still holds the $0.05 audio rate
  • Deprecated: expect a shutdown date to appear
  • Plan the move to 2.0 rather than defending the old rate

Imagine Video 1.5 GA

Improved video generation with better motion, physics, synchronized audio and clearer speech. The xAI documentation describes creating videos from text or image references with preset voices, up to 1080p.

  • Quality and Fast modes
  • Product and API access
  • Library search for generated media

Documented video resolutions and rates

The xAI documentation lists 480p, 720p and 1080p output, priced at $0.08 per second at 480p, $0.14 per second at 720p and $0.25 per second at 1080p.

  • At the listed rate, six seconds of 1080p is $1.50 of generation
  • Choose the resolution the channel actually needs
  • A living docs page, so re-check the rates before you budget

Video 1.5 Fast

Produces six-second 720p clips in about 25 seconds according to xAI.

  • Rapid iteration
  • Useful for social and concepts
  • Inspect every clip for sync and warping

Audio and speech

Sound effects, ambience and dialogue can be generated with the video rather than added as a separate first pass.

  • Faster concepting
  • Voice and brand risk
  • Still needs final sound review

Imagine Image 2.0 August 7

The image side of Imagine moved to version 2.0 on August 7, 2026 and gained new parameters on August 28. Re-test any prompt library or brand guardrail built against the previous version before assuming the outputs still match.

  • New version, new output characteristics
  • Re-run your reference prompts
  • Re-check text, likeness and product truth

Voice APIs

xAI offers speech-to-text, text-to-speech, voice-agent tools and custom voice capabilities. The Voice Agent Builder is still marked Beta.

  • Multilingual workflows
  • Consent and impersonation controls
  • Secure voice assets

The integrated media workflow

Do not let fast rendering replace creative control.

Concept
Define the communication job

Write the audience, message, visual idea and audio role before generating assets.

  • One central idea
  • Channel and duration
  • Rights and consent constraints
6 / Workplace Add-Ins

Grok moved into Microsoft 365 and Google Workspace

Between July 20 and July 24, 2026, xAI shipped add-ins for Excel, Outlook and Google Workspace. Grok now appears beside the documents your teams already work in, on surfaces owned by Microsoft and Google, and since late August it also appears inside those companies’ model catalogs. Treat this as a distribution decision, not only a feature release.

Grok for Excel July 20

A free Microsoft 365 add-in. Ask questions of a spreadsheet in plain English, generate formulas, and run scenarios and forecasts against the open workbook.

  • Plain-English queries over your data
  • Formula generation and scenario runs
  • Connectors for SharePoint, Drive and email

Grok for Outlook July 21

Thread summarization, reply drafting matched to your writing style, and inbox triage inside Outlook.

  • Available to all paid X and SuperGrok users
  • Style-matched draft replies
  • Review drafts before sending, because tone is imitated rather than verified

Grok in Google Workspace July 24

One free add-on from the Google Workspace Marketplace that places Grok beside the open file in Sheets, Slides and Docs.

  • A single add-on covers all three apps
  • Installed through the Marketplace
  • Confirm your admin allows Marketplace installs

Grok inside the rival catalogs August

Grok 4.6 is available on Amazon Bedrock since August 19, 2026, the Gemini Enterprise Agent Platform since August 21 and Microsoft Foundry since August 26. Add-ins put Grok next to the documents; catalog listings put it inside the platform contract. Both routes can bring Grok into an organization without a decision from whoever owns AI policy.

  • Three third-party catalogs in seven days
  • Procurement may already permit it
  • Decide the policy before a team discovers the listing

Automations July 16

Grok can run recurring autonomous tasks on a schedule or from an email trigger. Each run opens as a full conversation thread you can read afterward.

  • Scheduled automations for all users
  • Email triggers are SuperGrok-only
  • Every run leaves an inspectable thread

Automations are not /goal Clarification

Automations is a consumer feature for recurring tasks. It does not replace /goal, which is the long-running execution mode inside the Grok Build CLI. The two solve different problems.

  • Automations: recurring, consumer, triggered
  • /goal: sustained engineering execution in the CLI
  • Do not substitute one for the other in a plan

Why this matters Strategic

Shipping into Excel, Outlook, Sheets, Slides and Docs puts Grok on the incumbents’ own surfaces, where Microsoft and Google ship their own assistants. Adoption may now happen without a procurement decision.

  • Expect overlap with Copilot and Gemini entitlements
  • Set an add-in approval policy before demand arrives
  • Audit which connectors each add-in requests
7 / Business

Grok’s enterprise story now extends beyond X

Grok Business and Enterprise add company knowledge, administration and security. The decision is no longer simply whether employees may use a consumer assistant, and since late August it is no longer only a decision about buying from xAI directly.

CapabilityBusinessEnterpriseEnterprise Vault
Team management and billingIncludedIncludedIncluded
Google Drive groundingPermission-awarePermission-awarePermission-aware
SSO and SCIMNot includedIncludedIncluded
Advanced audit and securityStandard controlsAdvancedAdvanced
Dedicated data planeNot includedOptional tierIncluded
Customer-managed keysNot includedOptional tierIncluded

Published plan prices

The pricing page, checked September 3, 2026, shows Free at $0 per month, SuperGrok at $30 per month and SuperGrok Plus at $100 per month. Grok Business shows no per-seat price at all, only “Talk to our sales team about custom rate limits, dedicated infrastructure, SSO, compliance requirements, and volume pricing for your organization.”

  • Consumer tiers are published; Business is not
  • Get any Business seat price from xAI in writing
  • Do not reuse the SuperGrok $30 as a business rate

Company knowledge

Grok can search Google Drive while respecting the permissions already defined there.

  • Citation previews
  • Highlighted relevant passages
  • No access to unshared files

Data commitment

xAI states that business data is not used to train its models.

  • Still review product terms
  • Map regional data needs
  • Control connector access

Collections API

Agentic search over large document stores supports data rooms, legal review and financial modeling.

  • Curate the source set
  • Check retrieval coverage
  • Preserve citations

Enterprise controls

SSO, SCIM, administration and audit capabilities support larger managed deployments.

  • Lifecycle users correctly
  • Assign product ownership
  • Monitor usage

Vault

An isolated data plane with application-level encryption and customer-managed keys targets the most sensitive customers.

  • Evaluate operational model
  • Understand logging and recovery
  • Verify contractual scope

Third-party safeguards evaluation September 1

xAI published “Biosecurity at the frontier” on September 1, 2026, reporting a LatchBio evaluation of Grok 4.6 that measured 59.2% refusal on adversarial tasks and 64.8% completion on legitimate ones. Those figures are LatchBio’s, published by xAI. It is a safeguards validation rather than a product change, and the two numbers are in tension by design: refusing more adversarial requests and completing more legitimate ones pull against each other, so read them as a pair.

  • Figures produced by LatchBio, published by xAI
  • A safeguards study, not a capability or product release
  • Useful evidence for a security review, not a compliance certificate

Skills and apps

Reusable work behavior and connected tools turn the product into an organizational workflow layer.

  • Govern creation and sharing
  • Review third-party connectors
  • Retire stale skills
8 / Behavior Playbook

Nine habits for using Grok without confusing speed with truth

Grok’s value is strongest when teams deliberately separate live signal, verified evidence, direct critique and bounded execution, and when nothing you ship depends on a moving alias or an undated absence claim.

01

Use X for signal

Map reaction and language without calling it representative truth.

02

Verify the factual core

Move every important claim to a primary record.

03

Time-stamp live answers

State when the signal was observed because it will change.

04

Ask for direct critique

Use Grok’s tone to stress-test an idea, then return to evidence.

05

Use /goal for outcomes

Give Build a finish line, tests and authority boundaries.

06

Inspect external state

Review the code, file, post or media result rather than the completion claim.

07

Fetch before you say “no card exists”

An absence claim needs a direct fetch of the document, not a glance at an index. This guide got that wrong about the Grok 4.5 model card.

08

Name region, surface and date

API, consumer app, cloud catalog and geography availability are separate claims that landed on different days.

09

Pin the model ID

Never ride an alias. grok-voice-latest repriced itself on August 5, 2026 with no action by the customer, and grok-imagine-image-quality silently drops to low quality on November 2, 2026.

9 / Watch Outs

Fast social intelligence is noisy by construction

Grok’s differentiators also define its risk: live feeds can be manipulated, confident tone can conceal uncertainty and integrated agents can act before a weak claim is corrected. This edition adds one more: a guide can be wrong in the direction of saying something does not exist.

Selection bias

Visible X conversation is not the same as broad public, customer or employee sentiment.

Coordinated manipulation

Bots and campaigns can manufacture salience, repetition and apparent consensus.

Confident tone

Directness can make uncertain interpretation feel settled. Require sources and calibration.

An absence claim is the easiest thing to get wrong

The previous edition of this guide said no model card was ever published for Grok 4.5. One exists, dated July 14, 2026, linked from x.ai/safety and hosted at media.x.ai. Part of the explanation is that data.x.ai/ and data.x.ai/model-cards now return 404, so a check aimed at the old location finds nothing and can be mistaken for proof. Ask for the card by model number and fetch the file itself before reporting that it is missing.

Model cards are per model

Cards exist for Grok 4.6 (August 12, 2026, revision 2026-08-17), Grok 4.5 (July 14, 2026, revision 2026-07-20), Grok 4 Fast, Grok Code Fast 1 and Grok 4. A card for the current model still says nothing about the one you deployed in July, so match the card to the model ID your product actually calls.

Stale pricing page

The x.ai/api page still lists grok-4.3 as flagship at $1.25/$2.50. Do not quote it. Grok 4.6 is $2 per 1M input and $6 per 1M output below 200,000 tokens, with a higher tier above that threshold, and long context pricing bills the long-context rates for all tokens in the request.

Grok Business pricing is unconfirmed

Earlier editions of this guide carried $30 per seat per month for Grok Business. That figure is UNCONFIRMED and may be our own error: x.ai/pricing, checked September 3, 2026, shows SuperGrok at $30 per month and SuperGrok Plus at $100 per month, while Business shows no per-seat price at all, only the invitation to talk to sales. The $30 looks like a conflation with the SuperGrok consumer plan. Do not quote a Business seat price; get it from xAI sales in writing.

A silent quality downgrade has a date

On November 2, 2026 grok-imagine-image-quality retires and its requests are served by grok-imagine-image-2.0 at low quality and a lower price. No error, no outage, cheaper invoice, worse images. Put a visual check and a decision in the calendar before the date, because nothing will alert you on the day.

Voice alias reroute, executed

On August 5, 2026 grok-voice-latest began serving grok-voice-think-fast-2.0: 60 percent more per minute of audio, $0.05 to $0.08, with no action required from anyone. Version 1.0 is marked [Deprecated]. The warning was published in advance and it landed on the day, which is the argument for keeping the watchlist.

The news index is incomplete, not merely lagging

The x.ai/news index has been observed omitting real articles that exist at their own URLs, including xai-joins-spacex and anthropic-compute-partnership. That is why the absence of further Cursor acquisition detail after August 14, 2026 is recorded here as not found rather than as nothing having happened. Never enumerate xAI releases from the index; link the dated article.

xAI pages contradict each other

On September 3, 2026 x.ai/voice still advertised Speech-to-Speech at “$0.05 / min” while the developer docs priced the live voice model at $0.08 per minute. Two first-party pages, two rates. Cite the docs for anything you will be invoiced for, and name the page you took the number from.

Beta language is sticky

Grok Bot expanded from three plans to eight on August 26, 2026 and gained an X integration on August 29, and neither post uses the word beta. No post says it left beta either. Widen the availability claim to the eight named plans and stop short of calling it generally available, because xAI has not.

Regional and tier availability

Grok 4.5 launched without EU access and gained it on July 17, 2026. Availability now also varies by cloud catalog: Bedrock, the Gemini Enterprise Agent Platform and Microsoft Foundry each carry their own regions and terms. Check the current status per surface and per tier rather than a remembered one.

Third-party evaluation is not certification

The LatchBio evaluation xAI published on September 1, 2026 reports 59.2% refusal on adversarial tasks and 64.8% completion on legitimate ones for Grok 4.6. Those are LatchBio’s measurements on chosen task sets, published by the vendor whose model was tested. Cite them with both attributions and do not present them as an approval.

Agent permissions

Build plugins, external integrations and business apps expand the blast radius of mistakes, and Grok Build has been on web and mobile since August 19, 2026.

Voice and likeness

Custom voices and generated speech require explicit consent, impersonation controls and secure asset handling.

Integrated media errors

Video, physics, speech and audio can fail independently inside one polished clip, and changed defaults such as the August 28 move from medium to auto quality shift output without any action by you.

Rapid product churn

The release cadence makes dated claims and direct source links essential.

10 / Sources

First-party evidence behind this guide

These xAI releases anchor the current model, voice and image pricing, Build, add-in, Imagine, distribution and business claims. Each links to the specific announcement rather than the rolling news index, and that is not a stylistic preference: the index has been observed serving a current window while omitting real articles that exist at their own URLs, so it cannot be trusted to enumerate what xAI has shipped. Two model cards are now cited directly, which is the correction at the center of this edition: the previous edition reported that Grok 4.5 had no card, and the card is linked from the safety page and hosted on media.x.ai, while data.x.ai now returns 404. Two entries below are a worked example of why this guide never reverse-engineers a URL from a headline: x.ai/news/biosafety-at-the-frontier serves a post titled “Biosecurity at the frontier,” and x.ai/news/grok-4-6-vertex-ai serves “Grok 4.6 on Gemini Enterprise Agent Platform.” Each title below is the one the page actually carries, and each slug is the one that actually resolves. Several claims here still have no citable stable article behind them and come from release notes and changelog pages that are rewritten in place: the July 8, 2026 API availability of Grok 4.5, the July 17 EU access in the API console, the August 19 Amazon Bedrock listing and Grok Build on web and mobile, the August 28 Imagine image API parameter changes and the September 2 notice retiring grok-imagine-image-quality on November 2, 2026. A reader who needs to rely on any of them should confirm it directly. One further caveat is stated in the guide rather than hidden here: no page published after August 5 restates the grok-voice-latest alias target, so that reroute rests on the July 29 announcement plus the deprecation marking now on version 1.0.

xAI safety and model cardsSpaceXAI · the model card index, which does carry a Grok 4.5 card · checked September 3, 2026Model Card: Grok 4.5SpaceXAI · Model Card: Grok 4.5, July 14, 2026, revision 2026-07-20 · the card the August 19 edition said did not existModel Card: Grok 4.6SpaceXAI · Model Card: Grok 4.6, August 12, 2026, revision 2026-08-17 · the card for the current flagshipBiosecurity at the frontierSpaceXAI · September 1, 2026 · LatchBio third-party safeguards evaluation of Grok 4.6Grok Bot now works with XSpaceXAI · August 29, 2026 · first version of the X integrationGrok Bot is now included with more plansSpaceXAI · August 26, 2026 · expansion from three plans to eightGrok 4.6 on Microsoft FoundrySpaceXAI · August 26, 2026 · Grok 4.6 in Microsoft’s model catalogGrok 4.6 on Gemini Enterprise Agent PlatformSpaceXAI · August 21, 2026 · Grok 4.6 in Google’s model catalogIntroducing Grok BotSpaceXAI · August 11, 2026 · the original early beta for three subscription tiersxAI model referenceSpaceXAI · living model table · checked September 3, 2026 · Grok 4.6 current, grok-voice-think-fast-1.0 marked deprecatedIntroducing Grok 4.5SpaceXAI · post byline July 16, 2026 · the prior generation: model, pricing and Cursor trainingGrok 4.5 EverywhereSpaceXAI · July 22, 2026 · rollout to iOS, Android, web and XIntroducing Grok BuildSpaceXAI · May 25, 2026 · the coding CLI agentGrok Build is now open sourceSpaceXAI · July 15, 2026 · agent loop, tools, TUI and extension systemsGrok Build 0.1 on APISpaceXAI · Build on the API surfaceIntroducing /goalSpaceXAI · June 22, 2026 · long-running task delegationAutomations in GrokSpaceXAI · July 16, 2026 · scheduled and email-triggered runsIntroducing Grok for ExcelSpaceXAI · July 20, 2026 · free Microsoft 365 add-inIntroducing Grok for OutlookSpaceXAI · July 21, 2026 · summarization, drafting and triageIntroducing Grok in Google WorkspaceSpaceXAI · July 24, 2026 · free add-on for Sheets, Slides and DocsGrok Imagine Video 1.5SpaceXAI · June 16, 2026Grok Voice Think Fast 2.0SpaceXAI · July 29, 2026 · $0.08 per minute of audio, and the August 5 grok-voice-latest reroutexAI API pricingSpaceXAI · living price table · checked September 3, 2026 · carries both voice model rates and the 200,000-token tierIntroducing the Voice Agent BuilderSpaceXAI · July 1, 2026 · still marked Beta, not GAIntroducing Grok Business and Grok EnterpriseSpaceXAI · December 30, 2025 · no per-seat price is published; pricing must come from salesxAI joins SpaceXSpaceXAI · February 2, 2026 · why site metadata now reads SpaceXAICursor is now a part of SpaceXCursor · August 14, 2026 · acquired by SpaceX, completing a process begun in April · no terms disclosed

AI Mindset

Explore the model cheatsheets