✓ Reviewed by The Future Signal
✓ Reviewed by The Future Signal
Grok’s models are genuinely competitive, ranking fourth of 168 on independent benchmarking and first on agentic tool use, with real-time X access as a distinct advantage. The difficulty is everything around them. Formal regulatory investigations across several jurisdictions over image generation, bans in some markets, a $530 million litigation provision in SpaceX’s IPO prospectus and roughly 6% enterprise adoption make this a hard vendor to approve. Suitable for individual developers; difficult for regulated or brand-sensitive businesses.
Grok’s practical business impact is strongest in technical work. Its coding-focused flagship was trained on real developer sessions and leads independent rankings on agentic tool use, which translates into genuine productivity for development teams. Real-time access to X data gives it current-events awareness competitors lack, useful for monitoring, research and market sentiment work.
Competitive API pricing makes it viable for cost-sensitive workloads. Against this, the regulatory and reputational exposure creates a countervailing business cost: procurement friction, brand-safety questions, and availability problems in markets where Grok has been restricted.
Compare plans and pricing to find the best option for your needs.
Compare plans and pricing to find the best option for your needs.
Last Updated: 28 August 2026
Grok’s models are genuinely good. Independent benchmarking places the current lineup among the strongest available, and it leads on agentic tool use. That is the easy half of this review. The harder half is that Grok spent 2026 at the centre of a multinational regulatory crisis over image generation, drawing formal investigations across several jurisdictions, national bans in some markets, and a litigation provision in its parent company’s IPO prospectus. Enterprise adoption sits far below rivals. For most businesses this is a capable model attached to a risk profile that is difficult to sign off.
Estimated reading time: 10 minutes
Grok is the AI assistant built by xAI, the company Elon Musk founded in 2023 and positioned as a “maximally truth-seeking” alternative to what he characterised as overly restricted competitors. It runs inside X, as standalone apps, and through a developer API.
Its ownership changed fundamentally this year. On 2 February 2026, SpaceX acquired xAI in an all-stock transaction that valued xAI at $250 billion and the combined company near $1.25 trillion. By May, Musk confirmed xAI would cease to exist as a standalone company, and on 6 July the corporate brand became SpaceXAI, led by former Starlink vice president Michael Nicolls. Reporting indicates nearly all of xAI’s original co-founders departed during the transition.
The product line kept its name. Grok, SuperGrok, the API and the apps are unchanged, and no pricing or endpoint migration accompanied the rebrand.
The current model lineup is unusually fragmented. Grok 4.6 arrived on 12 August 2026 as the interim flagship. Grok 4.5, released 8 July, is a 1.5-trillion-parameter mixture-of-experts model built on the V9 architecture, tuned for coding and trained partly on real developer sessions from Cursor. Older models remain live with larger context windows.
Grok 5 has not shipped. Despite enormous anticipation, the only official confirmation of its existence is a line in xAI’s January 2026 funding announcement stating it is in training. Targets of Q1 and then Q2 2026 both passed. Any Grok 5 benchmark or specification circulating publicly is speculation.
Why this matters for your business: Choosing Grok is no longer just a model decision. It is a decision to depend on a product owned by a rocket company, led by a founder whose public positions are themselves a business variable, and operating under active regulatory investigation in several major markets. Those factors belong in your evaluation alongside benchmark scores.
Grok is a strong fit for:
Grok is probably not the right choice for:
Grok 4.6 and 4.5. The current flagship pairing. Grok 4.5 is built on the V9 architecture as a 1.5-trillion-parameter mixture-of-experts model, tuned specifically for coding and agentic work, with a 500,000 token context window.
Very large context on older models. Grok 4.3 offers 1 million tokens, and Grok 4.20 and 4.1 Fast reach 2 million. Notably, the newest model has the smallest context of the recent lineup, so version choice depends on task rather than recency.
Real-time X integration. Grok can draw on live posts from X, giving it current-events awareness that models relying on periodic training updates cannot match. This is its clearest genuine differentiator.
Native multimodality. Grok 4.3 added native video input alongside image and voice, plus document generation for PDFs, spreadsheets and slides.
Grok Imagine. Image and video generation powered by xAI’s Aurora engine. This is also the component at the centre of the regulatory problems discussed below.
Voice mode and Voice Agent API. Voice has been in the Grok app since February 2025, with a developer-facing Voice Agent API launched in December 2025.
Developer API. Pay-per-token access to the model family, with reported work underway to list Grok models on AWS Bedrock, which would let developers use them without a SuperGrok subscription.
Enterprise tier. xAI launched a business-focused offering with security controls, team management, compliance features and dedicated support.
Grok 4.6 ─► Aug 2026 · current flagship
Grok 4.5 ─► Jul 2026 · 1.5T MoE · 500K ctx · coding/agentic
Grok 4.3 ─► Apr 2026 · 1M ctx · native video
Grok 4.20 ─► Feb 2026 · up to 2M ctx
Grok 5 ─► NOT RELEASED · still in training
└── every "Grok 5 spec" you've seen is rumour[Illustration placeholder: Grok model lineup and context windows]
For end users, Grok is easy. It sits inside the X app most of its users already have, has standalone mobile apps, and requires no prompting expertise. The free tier lets anyone try it without commitment.
For developers, the API is conventional and well documented, with pay-per-token pricing and no subscription requirement.
The genuine confusion is the lineup itself. Several versions run concurrently with different capabilities, prices and context windows, and the newest model is not automatically the best choice — Grok 4.5 has a smaller context window than three older models. Picking correctly requires actually reading the specifications rather than defaulting to the highest number.
Future Signal Tip: Do not assume the newest Grok is the right one. If your work involves long documents or large codebases, an older model with a 1M or 2M token window will serve you better than the current flagship’s 500K. Match the version to the task.
On capability, Grok performs well. Independent testing by Artificial Analysis ranked Grok 4.5 fourth out of 168 models on its Intelligence Index and first on agentic tool use — a genuinely strong result, particularly for coding workflows where the model was trained on real developer sessions.
Real-time access to X is a distinct advantage for anything time-sensitive. If your use case involves breaking events, market sentiment or live public discussion, Grok sees things other models cannot.
Two performance caveats. First, the “maximally truth-seeking” positioning has not translated into measurably lower hallucination than competitors; independent commentary generally finds Grok as prone to error as other frontier models. Second, the model churn means benchmark results age quickly, and a result for one version does not transfer to another.
| Model | Context | Notable | Best for |
|---|---|---|---|
| Grok 4.6 | Varies | Current flagship (Aug 2026) | General use |
| Grok 4.5 | 500K | 4th of 168 on AA index; 1st agentic | Coding, agents |
| Grok 4.3 | 1M | Native video input | Multimodal, long docs |
| Grok 4.20 / 4.1 Fast | Up to 2M | Largest context in lineup | Very large inputs |
| Grok 5 | — | Not released | — |
This section exists because omitting it would make the review useless to a business buyer.
Beginning in late December 2025, Grok’s image generation was widely reported to be producing non-consensual sexualised images of real people, including material that appeared to depict minors. The regulatory response was fast and international.
Ofcom opened a formal investigation into X on 12 January 2026 under the Online Safety Act. The European Commission opened formal proceedings under the Digital Services Act on 26 January, assessing whether X carried out required systemic-risk assessments before deploying Grok. French prosecutors searched X’s Paris offices on 3 February. An Amsterdam court warned of daily fines. The California Attorney General opened a state investigation, Baltimore filed suit, and Indonesia and Malaysia banned Grok outright citing child-protection law.
xAI has since introduced restrictions, and reporting suggests the era of unfiltered generation has ended. But the exposure is quantified in the parent company’s own filings: SpaceX’s June 2026 IPO prospectus set aside $530 million to cover potential litigation losses tied to Grok’s image generation.
⚠️ What this means commercially. Even if your business would never use image generation, vendor due diligence increasingly asks whether a supplier is under active regulatory investigation. For regulated industries, government suppliers and brand-sensitive organisations, that question is often disqualifying on its own — regardless of how good the underlying language models are.
Grok’s integration story is narrower than its competitors’. There is a developer API, native presence inside X, and a Voice Agent API. Reporting indicates AWS Bedrock listing has been in progress, which would give developers access through an existing cloud relationship — commercially significant, since analysis suggests the underlying compute arrangement matters more to Amazon than Grok API revenue does.
What is largely absent is the enterprise connector ecosystem competitors have built: no substantial marketplace, no deep CRM or productivity-suite integrations, no equivalent to Microsoft’s Office embedding or HubSpot’s CRM grounding.
One integration point deserves flagging. xAI uses public X posts to train Grok by default, with an opt-out available in X settings that reviewers describe as not straightforward. Businesses with an active X presence should be aware their public content forms part of the training corpus.
Grok’s consumer pricing is simple. A free tier offers limited messaging without media generation. SuperGrok costs $30 a month with higher limits and premium features. SuperGrok Heavy costs $300 a month for the highest rate limits and priority access. Grok is also bundled into X Premium subscriptions, which is the cheapest route for existing X subscribers.
On the API, Grok 4.5 launched at $2 per million input tokens and $6 per million output — competitive against similarly capable models and considerably cheaper than the top Western flagships. An enterprise tier adds security controls, team management and dedicated support.
The value question is not really price. It is whether the surrounding risk is acceptable. Grok’s enterprise market share sits around 6%, against roughly 55% for OpenAI and 47% for Anthropic, and Grok holds about 2.8% of global AI chatbot web traffic, fifth behind ChatGPT, Gemini, Claude and DeepSeek. Those numbers reflect a market that has largely made its own assessment.
The financial picture is also worth noting: xAI’s revenue was reported at roughly $500 million annually against a burn rate near $1 billion a month. SpaceX’s ownership makes that sustainable in a way it would not be for an independent company — but it also means Grok’s direction now serves a rocket company’s strategy rather than an AI company’s.
Claude or ChatGPT Enterprise — the default alternatives for businesses that need comparable capability with established compliance documentation and enterprise agreements.
Google Gemini — competitive on both capability and price, with deep Workspace integration and no equivalent regulatory overhang.
DeepSeek or other open-weight models — worth considering if low cost is the driver, though with their own jurisdictional trade-offs.
Perplexity — a better fit specifically for real-time information retrieval, if that is the Grok feature you actually want.
The most interesting thing about Grok in 2026 is that it stopped being an AI company’s product. SpaceX absorbed xAI, the founders left, and Grok now sits inside an organisation whose stated ambition is building the cheapest AI compute on Earth and eventually in orbit — SpaceX has filed to operate satellite-based data centres.
That changes what Grok is for. Its purpose is increasingly to justify and consume compute infrastructure rather than to win enterprise software deals, which explains why a product with strong benchmark results has only around 6% enterprise share and appears untroubled by it.
For business buyers the implication is straightforward. Grok is unlikely to compete for your compliance sign-off the way Microsoft or Anthropic will, because that is not where its parent company’s value is created. The models may keep improving; the enterprise wrapper around them probably will not improve at the same rate.
Watch two things: whether Grok 5 ships at all after repeated slippage, and whether the regulatory actions resolve with structural safeguards or with fines that a trillion-dollar parent simply absorbs.
Overall Rating: 6.2 / 10
Grok’s models deserve better than this score, and that gap is the point. On raw capability, agentic performance and real-time data access, this is a genuinely competitive product with pricing to match.
The score reflects it as a business purchase. Active investigations across several jurisdictions, national bans in some markets, a nine-figure litigation provision, roughly 6% enterprise adoption and a thin integration ecosystem add up to a vendor that many procurement processes will reject before the benchmarks are ever discussed.
Future Signal recommends Grok for individual developers and technical teams doing coding or agentic work, particularly those already paying for X Premium. We do not recommend it for regulated industries, government suppliers, or any organisation where brand safety and vendor due diligence are material.
Has Grok 5 been released? No. As of August 2026 it remains in training. The only official confirmation of its existence is a single line in xAI’s January 2026 funding announcement. Targets of Q1 and Q2 2026 both passed, and every published specification or benchmark is unverified speculation.
Who owns Grok now? SpaceX. It acquired xAI on 2 February 2026 in an all-stock deal valuing xAI at $250 billion, and the company rebranded as SpaceXAI on 6 July 2026. The Grok product name, apps, subscriptions and API are unchanged.
How much does Grok cost? There is a limited free tier. SuperGrok costs $30 a month and SuperGrok Heavy $300 a month, with access also bundled into X Premium. On the API, Grok 4.5 launched at $2 per million input tokens and $6 per million output.
Which Grok model should I use? It depends on the task rather than the version number. Grok 4.5 leads on coding and agentic work but has a 500K context window. Grok 4.3 offers 1M tokens, and Grok 4.20 and 4.1 Fast reach 2M. For long documents, an older model is the better choice.
Is Grok safe for business use? For text and coding tasks the models function normally, but the wider risk picture is significant. Grok’s image generation drew formal investigations from Ofcom, the European Commission, French prosecutors and the California Attorney General during 2026, with bans in some markets and a $530 million litigation provision in SpaceX’s IPO prospectus. Regulated and brand-sensitive organisations should treat that as material.
Does Grok train on my data? xAI uses public X posts to train Grok by default. An opt-out exists in X settings, though reviewers describe it as not straightforward. Businesses with an active X presence should check this.
Why is Grok’s enterprise adoption so low? Roughly 6% against 55% for OpenAI and 47% for Anthropic. The likely causes are the regulatory overhang, a thinner enterprise integration ecosystem, and the reputational associations attached to the brand — not model capability, which benchmarks well.
Is Grok financially stable? Yes, but through ownership rather than performance. Reported revenue of around $500 million against a burn near $1 billion a month would be unsustainable independently. SpaceX’s ownership removes that risk while shifting Grok’s strategic purpose toward compute infrastructure.
Grok’s strength is real capability at competitive prices, with a genuine differentiator in live access to X data and a strong showing on agentic and coding benchmarks. For an individual developer, particularly one already paying for X Premium, it is an easy and inexpensive tool to add.
Its weakness is everything around the model. A multinational regulatory crisis over image generation, bans in some markets, a nine-figure litigation provision, departure of the founding team, and enterprise adoption far below rivals combine into a vendor risk profile that most corporate procurement processes are built to reject.
The practical position is therefore split. If you are a technical user choosing a model for personal or small-team work, Grok is worth testing on its merits. If you are a business making a supplier decision — particularly in a regulated sector, or anywhere brand association matters — the capability is not the question you will be asked about, and the answers to the questions you will be asked are difficult.
Our final assessment after evaluating features, performance, value, and business impact:
Grok presents an unusually wide gap between capability and suitability. The models are genuinely strong, independently benchmarked fourth of 168 overall and first on agentic tool use, with competitive API pricing and a real differentiator in live X data. For individual developers and small technical teams, particularly existing X Premium subscribers, it is worth testing. For businesses making a supplier decision, the picture changes sharply.
Active regulatory investigations across multiple jurisdictions over image generation, national bans in some markets, a $530 million litigation provision in the parent company’s IPO prospectus, departure of the founding team, and enterprise adoption around 6% combine into a vendor risk profile most procurement processes will reject regardless of benchmark performance.
We may earn a commission if you purchase through our links, at no extra cost to you. This helps support independent reviews.
Explore other highly rated AI tools we’ve reviewed in this category.
