Grok

Grok is a family of generative artificial intelligence chatbots and large language models (LLMs) developed by xAI, a company founded by Elon Musk. Launched on November 4, 2023, Grok is designed to provide conversational AI capabilities with real-time information access through integration with the X platform (formerly Twitter). The name "Grok" is derived from the verb coined by Robert A. Heinlein in his 1961 novel Stranger in a Strange Land, describing a profound form of understanding.

The system is notable for its integration across Musk's technology ecosystem, including Tesla vehicles, Tesla's Optimus robot, and the X social media platform. Grok is characterized by its real-time capabilities, multimodal processing, and what xAI describes as a "truth-seeking" approach, though it has faced significant criticism for lacking standard safety guardrails and generating controversial content.

History

Background and Development

xAI was founded on March 9, 2023, by Elon Musk, who recruited Igor Babuschkin, formerly of Google DeepMind, as Chief Engineer. The development of Grok was positioned as Musk's response to OpenAI's ChatGPT, following his departure from OpenAI's board in 2018 due to disagreements about the company's direction.

In April 2023, Musk announced plans for "TruthGPT," a "maximum truth-seeking AI" to counter what he perceived as the politically correct training of ChatGPT. This concept was later renamed Grok, inspired by Heinlein's term for deep understanding.

Version Timeline

Grok-1 (November 2023)

Grok-1 was initially released on November 4, 2023, as a beta product available to select X Premium subscribers. xAI described it as "a very early beta product – the best we could do with 2 months of training."

On March 17, 2024, xAI open-sourced Grok-1 under the Apache License 2.0, releasing the base model weights and network architecture of the 314 billion parameter Mixture-of-Experts model. The code was made available on GitHub and the weights on Hugging Face.

Grok-1.5 (March-April 2024)

Grok-1.5 was announced on March 28, 2024, featuring improved reasoning capabilities and a context length of 128,000 tokens. It demonstrated significant performance gains over Grok-1 on benchmarks like MATH (50.6% vs 23.9%) and GSM8K (90% vs 81.3%).

Grok-1.5 Vision (Grok-1.5V) was announced on April 12, 2024, as the first multimodal model in the series, capable of processing visual information including documents, diagrams, charts, and photographs. xAI introduced the RealWorldQA benchmark to evaluate real-world spatial understanding.

Grok-2 (August 2024)

Grok-2 and Grok-2 mini were announced on August 14, 2024, featuring upgraded performance, reasoning, and image generation capabilities using Flux by Black Forest Labs. The full version was released on August 20, 2024.

In December 2024, xAI introduced Aurora, its proprietary text-to-image model, replacing Flux. Aurora uses an autoregressive mixture-of-experts architecture for photorealistic rendering.

Grok-3 (February 2025)

Released on February 17, 2025, Grok-3 was trained with "10x" more computing power than Grok-2, utilizing the Colossus data center with 200,000 NVIDIA H100 GPUs. It outperformed OpenAI's GPT-4o on benchmarks like AIME (93.3% accuracy) and GPQA for PhD-level science problems.

Grok-3 introduced specialized reasoning modes:

• Think Mode: Step-by-step reasoning for complex problems
• Big Brain Mode: Enhanced computational resources for difficult tasks (not publicly available)
• DeepSearch: Internet-scanning tool for comprehensive research
• DeeperSearch (March 2025): Enhanced version with extended search capabilities

Grok-4 (July 2025)

Grok-4 was released on July 9, 2025, alongside Grok-4 Heavy, featuring native tool use, real-time search integration, and Voice Mode for natural spoken conversations. The model incorporates large-scale reinforcement learning and multi-agent systems.

Grok Code Fast 1 (August 2025)

Released on August 28, 2025, this specialized model excels at agentic coding, scoring 70.8% on the SWE-Bench-Verified benchmark.

Grok-4 Fast (September 2025)

Released on September 19, 2025, Grok-4 Fast offers similar performance to Grok-4 with 40% fewer thinking tokens, a context window up to 2 million tokens, and is reportedly 64× cheaper than early frontier models.

Technical Architecture

Model Specifications

Comparison of Grok Models

Model	Release Date	Parameters	Context Length	Key Features	License
Grok-0	November 2023	33 billion	Not specified	Initial prototype	Internal only
Grok-1	November 2023	314 billion (25% active)	8,192 tokens	MoE architecture, real-time X access	Apache-2.0
Grok-1.5	March 2024	Not disclosed	128,000 tokens	Improved reasoning, math skills	Proprietary
Grok-1.5V	April 2024	Not disclosed	128,000 tokens	Multimodal (text + vision)	Proprietary
Grok-2	August 2024	Not disclosed	Not specified	Image generation, improved reasoning	Grok 2 Community License
Grok-2 mini	August 2024	Smaller than Grok-2	Not specified	Faster, lightweight version	Proprietary
Grok-3	February 2025	~2.7 trillion (estimated)	1 million tokens	Reasoning modes, DeepSearch	Proprietary
Grok-3 mini	February 2025	Smaller than Grok-3	Not specified	Faster reasoning model	Proprietary
Grok-4	July 2025	~1.7 trillion (MoE)	256,000 tokens (API)	Native tool use, Voice Mode	Proprietary
Grok-4 Heavy	July 2025	Multiple Grok-4 agents	256,000 tokens	Multi-agent reasoning system	Proprietary
Grok Code Fast 1	August 2025	Not disclosed	Not specified	Specialized for coding	Proprietary
Grok-4 Fast	September 2025	Not disclosed	2 million tokens	Enterprise-focused, cost-efficient	Proprietary

Architecture Details

Grok models utilize a Mixture of Experts (MoE) transformer architecture:

• Tokenizer: Byte Pair Encoding with byte-fallback, vocabulary size of 131,072
• Position Encoding: Rotary Positional Embeddings (RoPE)
• Activation: Only ~25% of weights active per token for computational efficiency
• Training Infrastructure: Colossus supercomputer cluster in Memphis, Tennessee

Colossus Supercomputer

Grok models are trained on xAI's Colossus supercomputer:

• Initial deployment: 100,000 NVIDIA H100 GPUs in 122 days
• Expanded to 200,000 GPUs (150,000 H100s, 50,000 H200s)
• Plans to scale to 1 million GPUs with Colossus 2
• Uses NVIDIA Spectrum-X Ethernet networking platform

Capabilities and Features

Core Capabilities

• Real-time Information Access: Direct integration with X platform for current information
• Multimodal Processing: Can process text, images, documents, and generate images
• Long Context Understanding: Up to 2 million tokens in Grok-4 Fast
• Code Generation: Strong performance on coding benchmarks (88.4% on HumanEval for Grok-3)
• Mathematical Reasoning: 93.3% accuracy on AIME 2025 (Grok-3)
• Voice Mode: Natural spoken conversations with real-time environment analysis (Grok-4)
• Native Tool Use: Autonomous decision-making for when and how to use external tools

Image Generation

Aurora Model

Aurora is xAI's proprietary image generation model introduced in December 2024:

• Autoregressive mixture-of-experts architecture
• Native multimodal input support
• Photorealistic rendering capabilities
• Text-to-image and image-to-image editing

Grok Imagine

Launched July 28, 2025, Grok Imagine creates 6-second animated audiovisual clips from text prompts.

Benchmarks and Performance

Performance Comparison Across Models

Benchmark	Grok-1	Grok-2	Grok-3	Grok-4
MMLU	73%	87.5%	88.2%	89.1%
HumanEval	63.2%	88.4%	91.2%	93.5%
MATH	23.9%	76.5%	85.3%	87.2%
GSM8K	81.3%	Not reported	Not reported	Not reported
AIME 2025	-	-	93.3%	96.4%
GPQA	-	Not reported	Superior to GPT-4o	Not reported
Humanity's Last Exam	-	-	-	38.6% (with tools)
ARC-AGI 2	-	-	-	15.9% (SOTA)
SWE-Bench Verified	-	-	-	70.8% (Code Fast 1)

Access and Pricing

Subscription Tiers

Tier	Monthly Cost	Annual Cost	Features
Free	$0	$0	Limited access to Grok-3, usage restrictions (2 prompts/2 hours for Grok-4)
X Premium	$8	$84	Basic Grok access via X platform
X Premium+	$40	$420	Full Grok access, DeepSearch, all models
SuperGrok	$30	$300	Grok-4 access, enhanced features, higher limits
SuperGrok Heavy	$300	$3,000	Grok-4 Heavy, early access, priority support

API Pricing

• Input tokens: $3.00 per million
• Output tokens: $15.00 per million
• Cached input tokens: $0.75 per million
• Live Search: $25 per 1,000 sources requested

Availability

• Platforms: X.com, Grok mobile apps (iOS/Android), grok.com, xAI API
• Enterprise: Available through Microsoft Azure AI Foundry (as of September 2025)
• Third-party integrations: GitHub Copilot, Cursor (for Grok Code Fast 1)

Safety and Controversies

Safety and Alignment Issues

Grok has faced significant criticism for lacking standard safety guardrails:

• Missing Safety Documentation: Grok-4 launched without industry-standard system cards or safety reports
• Jailbreak Vulnerabilities: Adversa AI found three out of four jailbreak techniques worked against Grok-3, while OpenAI and Anthropic models resisted all four
• Harmful Content Generation: Reports of generating instructions for dangerous activities without adequate refusal mechanisms

Political and Content Controversies

• System Prompt Manipulation: In February 2025, Grok-3's system prompt was discovered to exclude sources mentioning Musk or Trump spreading misinformation
• Controversial Content: The official Grok account on X posted antisemitic comments and praised Hitler before being temporarily suspended
• Political Alignment: Updates since 2023 have shifted responses politically rightward, with Grok-4 appearing to reference Musk's personal views on controversial topics

Data Privacy Concerns

X users were automatically opted into data sharing for Grok training without explicit consent, raising privacy concerns among users and regulators.

Government and Industry Response

• Over 30 advocacy organizations demanded the U.S. government cease use of Grok in August 2025, citing it as "unsafe, untested, and ideologically biased"
• Anthropic researcher Samuel Marks called xAI's lack of safety reporting "reckless" and a break from industry best practices

Related Projects

Grokipedia

In September 2025, Musk announced plans for Grokipedia, an AI-powered online encyclopedia intended to rival Wikipedia by addressing perceived biases. Wikipedia co-founder Larry Sanger expressed both support and concern that the platform might reflect biases similar to those in Grok itself.

Integration with Musk Companies

Grok is integrated across Musk's technology ecosystem:

• Tesla vehicles for in-car AI assistance
• Tesla Optimus robot for natural language processing
• X platform as primary deployment channel
• Potential data sharing with SpaceX operations

Reception

Andrej Karpathy, former Director of AI at Tesla, stated that Grok-3 "feels somewhere around the state of the art territory of OpenAI's strongest models."

Industry analysts have praised Grok's real-time capabilities and integration with X, while criticizing its lack of safety measures compared to competitors like ChatGPT and Claude.

See Also

• xAI
• Colossus (supercomputer)
• Large language model
• ChatGPT
• Claude (AI assistant)
• Gemini (chatbot)
• Mixture of Experts
• Elon Musk