AB
AB
AiBoss

All Categories

project

Grok 4 - A new generation of large-scale models launched by Musk's xAI

Grok 4 is xAI's latest large-scale AI model, boasting a 10-fold improvement in reasoning ability compared to its predecessor. The model possesses exceptional reasoning capabilities, achieving near-perfect scores on challenging exams like the SAT and GRE, and performing well on multiple benchmark tests...

What is Grok 4?

Grok 4 is xAI's latest large-scale AI model, boasting a 10x improvement in inference capabilities compared to its predecessor. The model exhibits exceptional reasoning abilities, achieving near-perfect scores on challenging exams like the SAT and GRE, and outperforming other cutting-edge models in multiple benchmark tests. Grok 4 supports multimodal functionality, understanding subjective concepts, generating code and visualizations, and features significant improvements in voice interaction. Grok 4 is available in two versions: Grok 4 (single agent) and Grok 4 Heavy (multi-agent), supporting up to four agents working simultaneously, with a context window supporting up to 256k tokens.

Main functions of Grok 4

  • Exceptional reasoning abilityHe achieved near-perfect scores on challenging exams such as the SAT and GRE, demonstrating a level of reasoning ability that surpasses that of humans.
  • Multimodal understandingAble to understand subjective concepts and search and analyze images.
  • Information Integration and SummarizationIntegrate information from social media and other channels, extract key events, and sort them by time.
  • Code and Visualization GenerationGenerate complex animations based on scientific prompts, such as simulating black hole collisions.
  • Voice interaction improvementSupports five new voices, making dialogue smoother and emotional expression more natural.
  • Complex task processingThey excel in complex tasks such as simulation management and possess strong strategic planning and execution capabilities.
  • Parallel agent cooperation: SuperGrok Heavy version, which supports multiple agents solving complex problems in parallel.

Grok 4 test performance

  • Official test:
    • Humanity’s Last ExamIt includes 2,500 interdisciplinary expert-level questions. Grok 4 Heavy scores 44.4% with the tools, and can be improved to 50.7% with further optimization.
    • AIME25 (Mathematics Competition)Grok 4 Heavy achieved a perfect score of 100%, crushing other models.
    • GPQA (Graduate Level Q&A)The Grok 4 Heavy scored 88.9%, outperforming the Gemini 2.5 Pro (86.4%) and the Claude 4 Opus (79.6%).
    • HMMT25 (High School Mathematics Competition)The Grok 4 Heavy scored 96.7%, far surpassing the Gemini 2.5 Pro (82.5%).
    • USAMO25 (United States Mathematical Olympiad)The Grok 4 Heavy scored 61.9%, significantly outperforming the Gemini DeepThink (49.4%) and Gemini 2.5 Pro (34.5%).
    • ARC-AGI (Abstract Reasoning)Grok 4 scores 15.9%, nearly double that of the previous commercial SOTA.
    • Vending-Bench (Simulation Management)Grok 4 netted $4,694, far exceeding Claude Opus 4 ($2,077) and human players ($844).
  • Third-party evaluation(Tested using the large model performance evaluation platform Artificial Analysis):
    • Artificial Intelligence IndexGrok 4 scored 73 points, surpassing OpenAI o3 (70 points), Google Gemini 2.5 Pro (70 points), Anthropic Claude 4 Opus (64 points) and DeepSeek R1 0528 (68 points).
    • Coding index and mathematical indexGrok 4 ranked first in all categories.
    • GPQA Diamond ScoreIt hit a record high of 88%, surpassing the Gemini 2.5 Pro's 84%.
    • Humanity’s Last Exam ScoreIt hit a record high of 24%, surpassing the Gemini 2.5 Pro's 21%.
    • speedGrok 4 achieves 75 tokens/second, which is lower than o3 (188 tokens/second) and Gemini 2.5 Pro (142 tokens/second), but better than Claude 4 Opus Thinking (66 tokens/second).

Grok 4 product pricing

  • Paid subscription plan:
    • SuperGrokAnnual fee: $300 USD; Monthly fee: $30 USD.
    • SuperGrok HeavyAnnual fee: $3,000 USD; Monthly fee: $300 USD.
  • API call pricing:
    • enter$3 per million tokens.
    • Output$15 per million tokens.

Grok 4 official website address

  • Official website addressGrok

Application scenarios of Grok 4

  • Educational guidanceIt provides students with personalized learning solutions, answers complex academic questions, and helps students better understand and master knowledge.
  • Scientific researchIt can analyze large amounts of experimental data, predict scientific trends, and help scientists discover new theories and technologies.
  • Business and FinanceTo conduct market analysis and forecasting, provide data support for enterprises to formulate business strategies, and optimize enterprise operational efficiency.
  • Content creationIt assists in creative generation, supports scriptwriting and animation production in fields such as advertising, film and television, and games, and improves creative efficiency.
  • Smart AssistantAs an intelligent voice assistant, it processes multimodal information, helps users complete daily tasks, and improves the convenience of life.
Back