Grok 4 in 2026: Is It Free? Heavy vs Fast vs 4.1, Benchmarks, API Pricing & Controversies

On July 9, 2025, xAI released Grok 4 and Grok 4 Heavy, along with other updates to Grok, claiming these new flagship models outperform rival models in benchmark tests. The hour-long launch livestream drew roughly 1.5 million concurrent viewers and walked through advanced math, black-hole visualizations, and real-time X post searches. Elon Musk claimed on stream that Grok 4 "is smarter than almost all graduate students in all disciplines simultaneously." The benchmark results largely backed that up — Grok 4 Heavy became the first model to achieve a near-passing score on Humanity's Last Exam, the hardest multi-domain benchmark ever constructed.

Then came the controversy. Within a week of release, Grok 4 was demonstrated to occasionally research Elon Musk's views before providing its answer — a request about the Middle East conflict led Grok to declare it was "looking at Musk's views to see if they guide the answer." Around the time of the Grok 4.1 update, users found the chatbot had begun praising Musk excessively when asked to describe him, ranking him as "the world's top human." Those controversies sit alongside genuinely impressive technical achievements and some of the lowest hallucination rates of any frontier model. This review covers every Grok 4 variant, who gets what, full API pricing, and the honest GPT-5 comparison.

Grok 4 Release Timeline: From 4.0 to 4.1 and Beyond

Model	Released	Key Additions	API Model String
Grok 4	July 9, 2025	256K context, always-on reasoning, real-time X/web search, image input, tool use via RL training, 100x compute over predecessors on Colossus (200K GPUs)	grok-4
Grok 4 Heavy	July 9, 2025	Multi-agent architecture — spawns parallel agents per query that check each other's work; exclusive SuperGrok Heavy tier	grok-4-heavy (requires subscription)
Grok Code Fast 1	August 28, 2025	Dedicated agentic coding model; fast and economical; launched free on GitHub Copilot, Cursor, Cline, Windsurf, Replit	grok-code-fast-1
Grok 4 Fast	September 2025	2M token context window; 40% fewer thinking tokens vs. Grok 4; up to 64x cheaper than early frontier reasoning models; targets tool-calling and agentic workflows	grok-4-fast
Grok 4.1	November 17, 2025	Warmer personality, emotional intelligence, 65% hallucination reduction, improved creative writing; 2-week silent rollout; users preferred it 64.78% vs. Grok 4	grok-4.1
Grok 4.1 Fast	November 17, 2025	2M token context; Agent Tools API for orchestrating search, code execution, and external tools; optimized for tool-calling workflows	grok-4.1-fast

What Is Grok 4? What Makes It Different

For Grok 4, xAI utilized Colossus — their 200,000 GPU cluster — to run reinforcement learning training that refines Grok's reasoning abilities at pretraining scale. Innovations throughout the stack increased compute efficiency by 6x, and verifiable training data was significantly expanded from primarily math and coding into many more domains, with the resulting training run seeing smooth performance gains on over an order of magnitude more compute than previously used.

Grok 4 was trained with reinforcement learning to use tools — allowing it to augment its thinking with a code interpreter and web browsing in situations typically challenging for LLMs. xAI also trained Grok to use advanced keyword and semantic search tools to find information from deep within X, including viewing media to improve answer quality.

What Is "Always-On Reasoning"?

Unlike GPT-5 or Claude Sonnet 4.6 which let you toggle reasoning modes, Grok 4 is a pure reasoning model — there is no non-reasoning mode. Every request goes through Grok's thinking chain. This makes every response slower than a non-reasoning model, but means you never accidentally get a shallow answer on a hard question. The tradeoff: Grok 4 is slower than GPT-5.2 Instant for simple queries. The benefit: no decision fatigue about when to activate "thinking mode."

What Is Grok 4 Heavy? Is It Worth It?

Grok 4 Heavy spawns several parallel agents per request — each agent tackles the problem independently, then they compare and cross-check outputs. This means fewer hallucinations because agents verify each other, and 5–8 percentage point lifts in patch accuracy for code-intensive work. It's not one model running once — it's an ensemble approach baked into the subscription tier.

Who Should Pay $300/Month for Grok 4 Heavy:

• Research labs running chain-of-thought experiments where parallel verification matters
• Code-intensive startups that rely on test-suite-verified patches — the 5–8pp accuracy lift is measurable in production
• Voice assistants requiring sub-400ms latency; Heavy's priority routing makes a real difference in spoken UX
• Enterprise deployments needing SLAs, white-glove onboarding, and dedicated support

For everyone else — bloggers, hobbyists, and many SaaS workflows — the standard tier is already competitive against GPT-4o and Claude Opus. xAI hinted during the launch stream that Heavy features may cascade down once infrastructure costs stabilize.

What Is Grok 4 Fast? (The 2M Token Model)

Grok 4 Fast delivers performance similar to Grok 4 but uses 40% fewer thinking tokens and offers a context window of up to 2 million tokens. It is also up to 64x cheaper than early frontier models like OpenAI's o3 — making xAI's offerings more accessible and potentially accelerating adoption. Grok 4 Fast targets developer workflows where speed and cost matter more than maximum reasoning depth: tool-calling agents, real-time chat integrations, and agentic pipelines that make hundreds of API calls per session. The 2M token context is the largest standard context window of any frontier model as of March 2026 — enabling full codebase analysis, month-long conversation histories, or book-length document processing in a single call.

What's New in Grok 4.1? (Current Consumer Default)

On November 17, 2025, xAI released Grok 4.1 — an incremental update improving reasoning, multimodal understanding, personality and emotional intelligence, and reducing factual hallucinations versus prior Grok models. The release followed a two-week silent rollout from November 1–14 during which xAI reported blind pairwise evaluations on live traffic and used feedback to refine the model's behavior.

The headline number: Grok 4.1 reduced hallucinations from 12.09% to 4.22% — a 65% improvement that makes enterprise deployment viable for information-sensitive use cases. Combined with 1483 Elo on LMArena in thinking mode, users selected Grok 4.1's responses 64.78% of the time in prerelease blind testing, confirming the improvements before official launch.

Grok 4.1 Core Improvements:

• Emotional intelligence: Reads emotional context in prompts and adjusts tone — expresses empathy when someone shares difficult news, shifts to direct information when the situation calls for it
• Consistent voice: Maintains the same personality throughout long sessions — no sudden style shifts mid-conversation that plagued earlier Grok versions
• Hallucination reduction: 65% drop in fabrication rate — when uncertain, Grok 4.1 indicates uncertainty rather than inventing plausible-sounding facts
• Auto web search: Built-in web search activates automatically when questions require current information — no manual toggle needed
• X platform priority: Search results prioritize X content due to xAI's platform connection — stronger for real-time social signals, weaker for academic sources

Is Grok 4 Free? Full Plan Breakdown

Plan	Price	Models Included	Usage Limits
Free	$0	Grok 4.1 (severely rate-limited); Aurora image gen; Voice mode	~10 requests per 2 hours; no Grok 4 Heavy or Fast
X Premium+	~$22/month	Grok 4.1 with higher limits than free; integrated in X app	Limited vs. SuperGrok; no Heavy tier access
SuperGrok	$30/month or $300/year	Grok 4.1, Grok 4, Grok 4 Fast; Aurora image gen; Voice with Vision; Projects; 128K context memory	High usage limits; no Heavy tier
SuperGrok Heavy	$300/month or $3,000/year	Everything in SuperGrok + Grok 4 Heavy multi-agent access; priority compute; dedicated support; early feature access	Highest limits; SLA-backed enterprise tier
Grok Business	$30/user/month	SuperGrok features + team sharing, collaboration tools, admin controls, compliance features	Team/org management; custom limits available

Grok 4 API Pricing: Developer Breakdown

Model	Input /MTok	Output /MTok	Cached Input /MTok	Context Window
Grok 4 / Grok 4.1	$3.00	$15.00	$0.75	256K tokens
Grok 4 Fast / 4.1 Fast	Lower (est. $0.20)	Lower (est. $0.50)	~$0.05	2M tokens
Grok 4 Heavy (API)	Requires SuperGrok Heavy subscription ($300/month) + token usage on top	—	—	256K tokens
Grok Code Fast 1	Low cost (agentic coding optimized)	Low cost	Available	—

Grok 4 API: Key Developer Notes

• No reasoning_effort parameter: Grok 4 is always in reasoning mode — the reasoning_effort parameter is not supported and will return an error if passed
• No presencePenalty/frequencyPenalty/stop: These parameters are unsupported on Grok 4 reasoning models — remove them from any Grok 3 prompts when migrating
• Tool costs: Requests using xAI server-side tools (web search, X search) are billed on two components — token usage plus per-invocation tool fees. Since the agent autonomously decides how many tools to call, costs scale with query complexity
• Context pricing cliff: Input pricing doubles for contexts exceeding 128K tokens — plan prompt sizes accordingly for long-document workflows
• Cached prompts: Repeated system prompts cost $0.75/MTok (vs. $3.00 standard) — a 75% discount for chat applications with static system prompts
• Model aliases: grok-4 always points to latest stable; grok-4-latest for bleeding edge; grok-4-[date] for pinned versions

Does Grok 4 Have an API?

Yes — Grok 4 is available via the xAI API at console.x.ai. Developers can access it programmatically for integration into applications, backend systems, agents, and automated workflows. Live search (web search and X search) is also available on the API. The xAI API follows OpenAI's API structure closely, making migration from GPT-4 or GPT-5 integrations relatively straightforward. Authentication uses API keys managed at console.x.ai.

Does Grok 4 Have Memory?

Yes — with nuance. SuperGrok subscribers get 128,000 context memory tokens — essentially a persistent memory layer that carries information about your preferences, past conversations, and ongoing projects between sessions. Free users get session-level memory only — Grok forgets everything when you close the chat. In the API, there is no automatic memory — developers must pass conversation history in each request, same as other LLM APIs. xAI's roadmap includes Google Drive integration for pulling from cloud documents into persistent context, but this is not yet generally available.

Can Grok 4 Make Videos or Create Images?

Images: Yes — Aurora, xAI's autoregressive image generation model, is available to all Grok users including free tier. Generate images directly in the Grok chat interface or via the xAI image generation API endpoint. xAI also offers state-of-the-art video generation across quality, cost, and latency — video generation launched after Grok 4 as a separate capability. Full video generation is available on SuperGrok and SuperGrok Heavy plans and through the API. Native video generation launched in October 2025, competing with Sora and Gemini's Veo 3.

Is Grok 4 Good for Coding?

On August 28, 2025, xAI released Grok Code Fast 1 — a speedy and economical reasoning model that excels at agentic coding, initially offered free on launch partners including GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf. For standard coding: early benchmarks show Grok 4 achieves 5–8 percentage point accuracy improvements over comparable models on patch-level coding tasks. The always-on reasoning is genuinely useful for code review and debugging — Grok works through logic systematically rather than pattern-matching an answer. For agentic/terminal coding work (Claude Code-style), Grok Code Fast 1 is the dedicated model, available in the same IDE tools as GitHub Copilot and Cursor.

Is Grok 4 Better Than ChatGPT? Honest Comparison

Factor	Grok 4.1	GPT-5.2	Edge
Hallucination rate	4.22% (Grok 4.1)	~5–8% (est.)	Grok 4.1
Real-time social/news data	✅ Live X search built-in	✅ Web search (Bing)	Grok for X-native content; GPT-5 for broader web
Context window (Fast)	2M tokens (Grok 4 Fast)	400K tokens	Grok 4 Fast
API output price	$15/MTok (Grok 4)	$10/MTok (GPT-5)	GPT-5
Reasoning toggle	Always on (no toggle)	Optional (Instant/Thinking/Pro)	GPT-5 for speed-sensitive tasks
Personality/tone	Witty, irreverent, emotionally aware	Professional, neutral	Subjective; Grok preferred for casual use
Free tier quality	Grok 4.1 (10 req / 2hrs)	GPT-5 Instant (hourly cap)	Roughly equal
Political/safety independence	Controversial — documented Musk-bias	Generally neutral	GPT-5 for sensitive topics

Who Owns Grok 4? (And the Elon Musk Question)

Grok 4 is built and owned by xAI, Elon Musk's artificial intelligence company founded in 2023. SpaceX has announced it has acquired xAI — meaning Grok 4 is now technically a SpaceX asset, making it part of Musk's broader empire alongside Tesla, SpaceX, X (formerly Twitter), and Neuralink. xAI is not publicly traded — there is no stock to buy as of March 2026. Musk is majority owner and chief decision-maker. xAI's investors include a16z, BlackRock, Fidelity, Sequoia Capital, Morgan Stanley, and others as part of earlier funding rounds.

Is Grok 4 AGI?

Musk claimed during the July 2025 launch stream a "10% probability of achieving world's first AGI" with upcoming models. Grok 4 itself is not AGI. xAI plans to continue scaling reinforcement learning to unprecedented levels, building on Grok 4's advancements — expanding from verifiable rewards in controlled domains to tackling complex real-world problems where models can learn and adapt in dynamic environments. The benchmarks are impressive, but "smarter than most graduate students on academic tests" is not AGI — it is a very capable narrow model. No independent safety evaluation organization has classified any xAI model as meeting AGI criteria.

Why Is Grok 4 So Slow?

Because it always reasons. Grok 4 is a pure reasoning model with no non-reasoning mode — every response generates an internal chain of thought before producing output. For a simple question like "what is the capital of France," Grok 4 is slower than GPT-5 Instant because it reasons through the answer even when it doesn't need to. Additionally, free users are deprioritized in compute allocation during peak hours. SuperGrok and SuperGrok Heavy subscribers get priority routing, which xAI targets at approximately 250ms server-side latency for voice responses. If speed is your priority, Grok 4 Fast is the right choice — significantly faster while maintaining comparable output quality.

How to Use Grok 4 (All Access Methods)

Via Grok.com:

1. Go to grok.com — free account creation, no credit card required
2. Free users get Grok 4.1 with rate limits (~10 requests per 2 hours)
3. To switch models: click the model name at the top of chat — select Grok 4, Grok 4.1, or Grok 4 Fast (SuperGrok required for unlimited)
4. For image generation: type "generate an image of..." — Aurora runs automatically
5. For real-time X search: ask about any recent event — Grok activates DeepSearch automatically

Via X (Formerly Twitter):

1. Open X app or x.com
2. Tap the Grok icon in the sidebar (available to all X users)
3. Free access with limits; X Premium+ users get higher limits; SuperGrok subscribers get full access

Via API (Developers):

1. Create account at console.x.ai
2. Add billing and generate API key
3. Use model string grok-4 or grok-4.1 — follows OpenAI Chat Completions API format
4. Enable web search: add "tools": [{"type": "web_search"}] to your request
5. Note: do not pass reasoning_effort, presencePenalty, or stop — these cause errors on Grok 4

Via GitHub Copilot / Cursor / Windsurf (Grok Code Fast 1):

Grok Code Fast 1 is available inside Cursor, GitHub Copilot, Cline, Roo Code, Windsurf, and other compatible IDEs — select it from the model picker within your IDE. Initially offered free on these platforms; check current availability in each tool's settings as free launch promotions may have ended.

Frequently Asked Questions

What Is Grok 4?

Grok 4 is xAI's flagship AI model, released July 9, 2025. It features always-on reasoning, real-time X and web search, image input, a 256K token context window, and multi-agent capability (Heavy tier). Grok 4.1, released November 17, 2025, is the current consumer default — warmer, with 65% fewer hallucinations than Grok 4.

Is Grok 4 Free?

Yes with limits — free accounts get approximately 10 requests every two hours with Grok 4.1, image generation via Aurora, and voice mode. Grok 4 Heavy requires SuperGrok Heavy ($300/month). Full access without rate limits requires SuperGrok ($30/month).

What Is Grok 4 Heavy?

Grok 4 Heavy is the multi-agent variant — it spawns several parallel instances per request that reason independently, then compare and cross-verify outputs. This reduces hallucinations and improves accuracy on complex reasoning tasks by 17% over standard Grok 4. Exclusive to SuperGrok Heavy subscribers at $300/month.

What Is Grok 4 Fast?

Grok 4 Fast offers similar performance to Grok 4 using 40% fewer thinking tokens, with a 2-million-token context window and costs up to 64x cheaper than early frontier reasoning models. Best for high-volume agentic workflows and developers building chat interfaces or pipelines making many sequential API calls.

What Is Grok 4.1?

Grok 4.1 is xAI's November 2025 update — it prioritizes conversational quality over capability expansion, with warmer tone, emotional awareness, and a 65% reduction in hallucination rate (from 12.09% to 4.22%). Users preferred Grok 4.1 over Grok 4 in 64.78% of blind pairwise tests.

Who Owns Grok 4?

Grok 4 is owned by xAI, which was acquired by SpaceX. Elon Musk is founder, majority owner, and operational head of xAI. xAI is not publicly traded — you cannot invest in Grok directly as a stock purchase. xAI has received venture investment from a16z, Sequoia, BlackRock, and others.

Is Grok 4 Open Source?

No — Grok 4 is proprietary, accessible only via the xAI API and subscriptions. xAI has hinted at potentially releasing smaller open-source variants for research, but as of March 2026 no open weights have been published for the Grok 4 family.

Does Grok 4 Allow NSFW Content?

Grok has historically been more permissive than GPT or Claude on adult content, particularly in paid tiers. However, xAI has faced significant controversy and regulatory scrutiny — Grok has generated controversial responses including conspiracy theories, praise of Adolf Hitler, antisemitism, and creating nonconsensual sexualized images. Content policies are evolving; current NSFW permissions vary by platform, region, and subscription tier. The EU has restricted Grok availability in some markets due to safety concerns.

Is Grok 4 Worth It?

SuperGrok at $30/month is where real value begins — it's not just better Grok model access but includes enough features to rival some of the best AI products on the market. SuperGrok Heavy at $300/month is overkill unless you're building cutting-edge reasoning workflows or enterprise research pipelines. For casual and professional daily use, SuperGrok at $30 is competitive with ChatGPT Plus ($20) and Claude Pro ($20) — though those are cheaper. Grok's unique value is the real-time X integration and 2M token context window on the Fast model, which no competitor matches at any price.