Not to be confused with Groq (the LPU inference provider, different company, different product). This page is xAI’s Grok chatbot.
xAI’s AI assistant, tightly integrated with X (formerly Twitter) and running on the Colossus 2 supercluster. Two active flagship surfaces as of May 2026:
- Grok 4.20 Beta 2 (the general-availability flagship, 4-agent architecture, ~4.2% hallucination rate vs Grok 4.1’s 12.09%)
- Grok 4.3 (paid consumer access plus API/OpenRouter access as of May 1, 2026; 1T parameters; 16-agent Heavy architecture on Heavy; adds native PDF, PowerPoint, spreadsheet, and video-input output layer)
Three major 2026 capability additions beyond the base models:
- Grok TTS (March 16): 5 voices, 20+ languages with auto-detect, inline speech tags, $4.20 per million characters
- Grok Voice Mode on X (March 19): real-time voice chat bundled into X Premium on Android and web
- Grok Voice Think Fast 1.0 (April 23): enterprise real-time voice-agent model for customer support, sales, and workflows
- Grok Computer (2026): autonomous desktop-agent integration, currently rolling out; Grok 4.3 routes to it
Context is now model-dependent: earlier Fast/4.20 surfaces are tracked at 2M tokens reporting lists 1M context. On February 2, 2026, SpaceX acquired xAI for $1.25 trillion in an all-stock deal (SpaceX valued at $1T, xAI at $250B). Grok now runs under Elon Musk’s consolidated aerospace and technology empire. Grok 5 (6T parameters, MoE) is training on Colossus 2 with a Q2 2026 target.
Recent developments (April-May 2026)
- May 8: GitHub will deprecate Grok Code Fast 1 across Copilot on May 15 following an xAI provider-side deprecation. GitHub points users to GPT-5 mini and Claude Haiku 4.5. Teams treating Grok as a stable coding backend should re-verify the exact surface and model before relying on it.
- May 1: Grok 4.3 moved into the xAI API and OpenRouter while Custom Voices added team-scoped voice cloning. The new buyer question is cost-per-completed-workflow, especially for legal/finance reasoning and voice-agent deployments.
- April 30: Musk testimony said xAI partly used OpenAI models to train Grok. Treat this as a model-provenance and governance caveat, not a feature change; the public record does not quantify the scope.
- April 24: AI Industry Roundup, April 24 kept Grok in the broader frontier-lab watchlist alongside DeepSeek V4, GPT-5.5 in Copilot, Anthropic capital pressure, and Cohere-Aleph Alpha.
- April 23: Grok Voice Think Fast 1.0 launched as xAI’s enterprise real-time voice-agent model with 25+ languages, alongside AI Industry Roundup, April 23 coverage of Grok Voice and Grok outages.
- April 19: Grok 4.3 beta expanded to SuperGrok ($30/mo) and X Premium+ (~$40/mo) 48 hours after Heavy-only launch, pushing the 1T-param model into broader paid tiers.
- April 17: Grok 4.3 beta launched on SuperGrok Heavy at $300/mo with the 16-agent Heavy architecture, priority rate limits, and the new PDF/PowerPoint/spreadsheet/video-input output layer.
System Verdict
Pick Grok if you are on X Premium already or need real-time X data access. The live X firehose is the single non-replicable capability. No other frontier assistant reads live X posts, trending topics, and breaking discussions at query time. For social intelligence, political analysis, journalism, or any workflow tied to live X data, Grok is the only serious option.
Skip it for predictable governance or strongest general reasoning. The SpaceX merger concentrated Grok’s roadmap, content policy, and platform continuity inside a single Musk-controlled conglomerate. That is a concentration risk users of OpenAI, Anthropic, and Alphabet-backed tools do not carry. On raw reasoning, Claude Opus 4.7 and GPT-5.5 beat Grok 4.20 outside X-data tasks. For structured web research without X data, Perplexity produces cleaner outputs than DeepSearch.
Who pays which tier: Free for X account holders (limited daily queries), X Premium ~$8/mo for bundled Grok + Voice Mode, SuperGrok Lite $10/mo, SuperGrok $30/mo or X Premium+ $40/mo for consumer Grok 4.3 access/OpenRouter access for developer workflows. SuperGrok Heavy $300/mo adds the 16-agent Heavy architecture, ~4,000 msg/day, 360 min Deep Search, and max compute priority. Heavy is now justified by multi-agent architecture and priority rate limits, not access to the 1T-param model alone.
Key Facts
| GA flagship | Grok 4.20 Beta 2 (also marketed as Grok 4.20 0309 v2) |
| API / paid flagship | Grok 4.3 · consumer access through X Premium+ ($40/mo), SuperGrok ($30/mo), and SuperGrok Heavy ($300/mo); API and OpenRouter access added May 1, 2026 · 1T parameters |
| 4.20 architecture | 4-agent multi-agent · Grok · Harper · Benjamin · Lucas |
| 4.3 architecture | 16-agent Heavy mode · longer training run |
| Hallucination rate | ~4.2% on 4.20 (down from 12.09% on Grok 4.1) |
| Context window | Model-dependent: 2M on earlier Fast/4.20 surfaces; Grok 4.3 API reporting lists 1M |
| Grok 4.3 new capabilities | PDF creation · PowerPoint slides · spreadsheets · video input · Grok Computer integration |
| Voice capabilities | Grok TTS API (March 16) · Voice Mode on X (March 19) · 5 voices (Ara, Eve, Leo, Rex, Sal) · 20+ languages |
| Voice agent model | Grok Voice Think Fast 1.0 · launched April 23, 2026 · 25+ languages claimed · enterprise voice workflows |
| Grok 4 Heavy benchmarks | 100% AIME 2025 · 88.4% GPQA Diamond · 75% SWE-bench Verified |
| Next model | Grok 5 · 6T parameters · MoE · training on Colossus 2 · Q2 2026 target |
| Ownership | SpaceX (acquired xAI Feb 2, 2026 for $1.25T all-stock) |
| Compute | Colossus 2 supercluster · 1.5GW since April 2026 |
| Subscription tiers | Free (X account) · X Premium ~$8 · SuperGrok Lite $10 · SuperGrok $30 · X Premium+ ~$40 · SuperGrok Heavy $300 |
| TTS API pricing | $4.20 per 1M characters (Beta) |
| API pricing: Grok 4.1 Fast | $0.20/M input · $0.50/M output · 2M context |
| API pricing: Grok 4.20 Reasoning | $3.00 blended (per third-party tracking) |
| API pricing: Grok 4.3 | $1.25/M input · $2.50/M output up to 200K; higher-context tier above 200K |
| Image generation | Aurora / Grok Imagine · restricted and geoblocked after 2026 deepfake controversies |
| Video generation | Grok Imagine video generation/editing API; consumer availability may vary |
| Persistent memory | None (frequent user complaint vs ChatGPT Memory, Gemini Saved Info, Claude Projects) |
Every data point verified on 2026-05-13 against x.ai, docs.x.ai, Artificial Analysis, and AIpedia’s Grok 4.3 API coverage.
What it actually is
A general-purpose AI assistant wired to the X platform’s live data stream. Core modes cover Q&A, writing, analysis, coding, and reasoning. Real-time X access is the defining capability.
DeepSearch combines live web crawl and X content into cited answers. Think mode surfaces step-by-step reasoning for complex queries. Grok 4 Heavy is the high-performance variant, tuned for advanced benchmarks and enterprise reasoning work.
Aurora / Grok Imagine handles image generation and, through xAI’s Imagine API, video generation and editing. Post-2026 deepfake controversies forced restrictions: non-consensual real-person edits are blocked, and geoblocking or tier gating may apply in some jurisdictions.
The real differentiator is data, not architecture. Claude Opus 4.7 and GPT-5.5 match or exceed Grok 4.20 on most general reasoning benchmarks. What none of them can replicate is the live X firehose. For social analysis, journalism, political research, or any workflow where current X discourse is the input, Grok wins by default.
When to pick Grok
- You need live X data in the reasoning loop. No other major assistant reads X posts at query time. This is the single unreplicable capability.
- You’re already on X Premium. Grok access is bundled. Incremental cost is effectively zero.
- You need 2M-token context at commodity API pricing. Grok 4.1 Fast Non-Reasoning handles 2M tokens at $0.20/$0.50 per M. Nothing else hits that ratio.
- You want a less restrictive content policy. Grok declines fewer prompts than ChatGPT or Claude. Useful for certain research and creative workflows; also a liability for compliance-heavy deployments.
- You need cheap long-context enterprise tasks. Grok 4.1 Fast handles long code, long documents, and long transcripts at a per-token rate below Claude or GPT-5.5.
When to pick something else
- Strongest general reasoning: Claude Opus 4.7. Takes leads on agentic coding, financial analysis, and long-form output coherence.
- Largest plugin and custom-GPT ecosystem: ChatGPT. GPT Store has no Grok equivalent.
- Google Workspace integration: Gemini. Native Docs / Sheets / Gmail hooks.
- Structured web research with citations: Perplexity. Cleaner source management than Grok’s DeepSearch for non-X queries.
- Predictable platform governance: any assistant backed by a public company or independent lab. Grok’s SpaceX ownership concentrates decision-making inside one private conglomerate.
- Polished video generation: Grok Imagine now supports video generation/editing through xAI APIs, but Veo via Gemini and specialist tools like Kling remain more proven for production video workflows.
Pricing
Consumer tier pricing for SuperGrok tiers is not uniformly disclosed on xAI’s public pages. Values below reflect third-party-verified rates and published API pricing. Confirm subscription rates at x.ai before signing up.
| Plan | Price | Who’s it for |
|---|---|---|
| Free (X account) | $0 | Limited daily queries via the X app; Grok 4.1 |
| X Premium | ~$8/mo | Existing X subscribers · Grok + Voice Mode bundled · Grok 4.20 |
| SuperGrok Lite | $10/mo | Entry paid tier, basic access |
| SuperGrok | $30/mo | Grok 4.20 · Grok 4 Heavy · Grok 4.3 consumer access · DeepSearch · Think mode · ~1,000 msg/day |
| X Premium+ | ~$40/mo | Everything in X Premium · Grok 4.3 consumer access |
| SuperGrok Heavy | $300/mo | Grok 4.3 with 16-agent Heavy architecture · ~4,000 msg/day · 360 min Deep Search · max compute priority |
API pricing (per 1M tokens):
| Model | Input | Output | Context | Notes |
|---|---|---|---|---|
| Grok 4.1 Fast Non-Reasoning | $0.20 | $0.50 | 2M | Long-context workhorse |
| Grok 4.20 0309 v2 (Reasoning) | $3.00 blended | Not broken out | Standard | Per third-party tracking |
| Grok 4.3 API | $1.25 | $2.50 up to 200K | 1M reported | API and OpenRouter access added May 1, 2026 |
API rates verified 2026-05-13 via xAI docs, Artificial Analysis, and AIpedia’s Grok 4.3 API coverage. Consumer subscription rates for SuperGrok tiers are not uniformly posted on xAI public pages.
Against the alternatives
| Grok 4.20 | ChatGPT (GPT-5.5) | Claude Opus 4.7 | |
|---|---|---|---|
| Context window | 2M tokens (largest) | Undisclosed | 1M tokens |
| Live X / social data | Yes · unreplicable | No | No |
| General reasoning | Strong; 4-agent cuts hallucinations | Broadest feature suite | Strongest on long-form |
| Image generation | Aurora (restricted) | GPT Image 2 | None |
| Video generation | Grok Imagine API | None | None |
| Content policy | Less restrictive | Moderate | More restrictive |
| Ownership / governance | SpaceX (Musk conglomerate) | Microsoft-backed OpenAI | Anthropic (independent) |
| Best viewed as | Social-intelligence specialist | Generalist default | Reasoning specialist |
Failure modes
- Third-party coding distribution can disappear quickly. GitHub will deprecate Grok Code Fast 1 across Copilot on May 15, 2026 because of a provider-side deprecation. Verify the exact Grok model and surface before treating Grok as a stable coding backend.
- Demand-driven outages. On April 23, 2026, Grok experienced intermittent connectivity issues and response delays across web, mobile, and X integration for several hours. Free-tier and lighter-subscription users absorbed most of the friction; SuperGrok and Premium+ subscribers noticed fewer interruptions. Second congestion episode in April 2026 after the 4.3 beta expansion to $30 and $40 tiers grew the traffic pool ~50x while Colossus 2 is still ramping to 1.5 GW.
- Voice-agent rollout is enterprise-first. Grok Voice Think Fast 1.0 matters for xAI’s commercial surface, but public pricing, SLA, latency distributions, and deployment limits still need more documentation.
- No persistent memory, full stop. Every competitor (ChatGPT, Gemini, Claude) has had cross-session memory for 18+ months. Grok still resets every conversation. The sharpest community complaint against the $300/mo Heavy tier.
- SpaceX ownership is a governance concentration risk. The February 2026 $1.25T merger placed Grok’s roadmap, content policy, and platform continuity inside a single Musk-controlled conglomerate. Production dependencies on Grok carry a platform-continuity variable no major competitor does.
- Response latency is slow. User reports of 10-15 second waits for responses that competitors return in 2-3. Noticeable on 4.20 Heavy; not yet measured on 4.3 beta.
- Creative writing is weak. Prose is literal and robotic per user reports. Grok wins on math, logic, and coding; Claude Opus 4.7 wins on long-form writing.
- Heavy’s 16-agent architecture is still a $300/mo paywall. As of April 19, 2026, the Grok 4.3 model itself is available on SuperGrok ($30/mo) and X Premium+ ($40/mo), but the 16-agent Heavy orchestration and priority rate limits remain Heavy-exclusive.
- Content policy is inconsistent. January 2026 tightening surprised users who switched from ChatGPT expecting fewer filters. Less restrictive overall but with no clearly documented boundary.
- DeepSearch is weaker than Perplexity for structured non-X research. Grok’s web search integration lacks Perplexity’s source management, relevance ranking, and clean report formatting.
- Aurora image gen is restricted and geoblocked. Post-2026 deepfake controversies forced restrictions that vary by jurisdiction. Less reliable than Midjourney or Ideogram for image-critical workflows.
- No video generation. Gemini’s Veo 3 and Kling handle that. Grok 4.3 adds video input but not output.
- Grok 5 is unreleased. Current state is a transitional lineup between 4.x and 5. Enterprise adoption decisions should factor that in.
- Advertising pressure. Musk has publicly pressured SpaceX IPO banks to buy Grok subscriptions and advertise on X. That pressure shapes the commercial roadmap in ways external users cannot see.
- Benchmark numbers for Grok 4 Heavy do not apply to SuperGrok. Heavy’s 100% AIME 2025 / 88.4% GPQA Diamond scores are for the top-tier variant. Default SuperGrok subscribers access Grok 4.20, not Grok 4 Heavy.
Methodology
This page was produced by the aipedia.wiki editorial pipeline, an automated system that ingests vendor documentation, verifies pricing and model details against primary sources, and generates the editorial analysis you are reading. No individual human wrote this review. Scoring follows the four-dimension rubric at /about/scoring/ (Utility × Value × Moat × Longevity, unweighted average). Last verified 2026-05-13 against x.ai, docs.x.ai reasoning docs, Artificial Analysis, the May 8 GitHub Copilot Grok Code Fast 1 retirement, the Grok 4.3 API + Custom Voices coverage, and the Grok Voice Think Fast 1.0 launch.
FAQ
Is Grok the same as Groq? No. Grok (this page) is xAI’s chatbot (launched 2023, owned by SpaceX post-merger). Groq is an LPU inference hardware company (founded 2016, Nvidia-acquired 2026). Different companies, different products. Groq has publicly complained about the naming confusion since Grok launched.
Is Grok free? Any X account gets limited free queries per day. X Premium (~$8/mo) bundles meaningful Grok access plus Voice Mode for existing X subscribers. SuperGrok Lite ($10/mo) is the entry paid tier. SuperGrok ($30/mo) and X Premium+ (~$40/mo) both now include Grok 4.3 beta access as of April 19, 2026 (previously Heavy-only). SuperGrok Heavy ($300/mo) adds the 16-agent Heavy architecture, ~4,000 msg/day, 360 min Deep Search, and max compute priority.
What’s new in Grok 4.3 beta? Released April 17, 2026 to SuperGrok Heavy first, then expanded April 19, 2026 to X Premium+ ($40/mo) and SuperGrok ($30/mo). Adds an output layer Grok 4.20 didn’t have: native PDF, PowerPoint, spreadsheet generation, plus video input. 1T parameters (double Grok 4.20’s ~500B). Heavy still gets the exclusive 16-agent architecture and priority rate limits. See the original launch coverage and the expansion coverage.
Does Grok do voice? Yes, as of March 2026. Grok TTS API launched March 16 with 5 voices (Ara, Eve, Leo, Rex, Sal), 20+ languages, and inline speech tags for pauses/laughter/whispers/emphasis, priced at $4.20 per 1M characters. Grok Voice Mode went live on X for Android and web on March 19, bundled into X Premium at no extra cost.
Grok Voice Think Fast 1.0 launched April 23, 2026 as xAI’s enterprise voice-agent model for phone support, sales, and workflow automation. xAI says it supports 25+ languages and is already used by Starlink’s phone sales and support line.
What happened with the SpaceX acquisition? SpaceX acquired xAI on February 2, 2026 in a $1.25T all-stock deal (SpaceX $1T, xAI $250B). Grok now runs under Elon Musk’s consolidated aerospace and technology empire alongside X and Neuralink. Compute runs on Colossus 2 (1.5GW since April 2026). The governance implication is that product decisions and platform continuity sit inside a single private conglomerate.
What is Grok 4.20 vs Grok 4 Heavy? Grok 4.20 Beta 2 (also labeled 0309 v2) is the current production flagship using the 4-agent architecture. Grok 4 Heavy is the high-performance variant tuned for advanced benchmarks (100% AIME 2025, 88.4% GPQA Diamond). Heavy is positioned for enterprise and demanding reasoning tasks, not general consumer access.
How does Grok compare to GPT-5.5 and Claude Opus 4.7? Grok 4 Heavy leads on math and science reasoning benchmarks. On general-purpose reasoning and writing, Claude Opus 4.7 and GPT-5.5 are broadly competitive and often ahead. Grok’s real advantages are the 2M context window (largest available), live X data access, and cheap long-context API pricing. Neither Claude nor GPT-5.5 can replicate the X data firehose.
When does Grok 5 launch? Musk has publicly targeted Q2 2026. Training is in progress on Colossus 2. Architecture is 6 trillion parameters with Mixture-of-Experts. No confirmed release date at the time of writing.
Sources
- xAI: product and company information
- xAI reasoning docs: multi-agent architecture and reasoning parameter
- Artificial Analysis · Grok 4.20 Providers: API rate tracking
- Grok Voice Think Fast 1.0: April 23, 2026 voice-agent launch
- CNBC · SpaceX / xAI merger: February 2026 $1.25T acquisition coverage
Related
- Category: AI Chatbots · AI Search
- Comparisons: ChatGPT vs Grok · Claude vs Grok · Gemini vs Grok
