Model Profile: Grok 4 (xAI)

Tagline: xAI's witty, truth-seeking model with real-time X integration, excelling in creative reasoning and factual tasks for engaging AI experiences.

📊 At a Glance

• Primary Strength: Humorous and maximally truthful reasoning, real-time data via X, and strong multimodal support (text, images, code) for dynamic, engaging interactions.

• Performance Profile:

  • Intelligence: ⭐⭐⭐⭐☆ (4.5/5; Excellent in creative and factual reasoning).

  • Speed: ⭐⭐⭐⭐ (4/5; Fast; optimized for real-time X interactions).

  • Cost: ⭐⭐⭐⭐☆ (4.5/5; Accessible; $0.40 input/$1.50 output per 1M tokens, rating 4.5/5 for value).

• Key Differentiator: Unique "truth-maximizing" training with humor, integrated real-time X data access, and partial open-weights for customization, making it ideal for witty, fact-based Mates with social media insights.

• allmates.ai Recommendation: Great for Mates needing engaging, humorous responses with real-time facts from X, such as social analysis or creative brainstorming, where truthfulness and speed are key.

📖 Overview

Grok 4, launched in August 2025 by xAI, is the latest in the Grok series, emphasizing "maximum truth" and humor inspired by the Hitchhiker's Guide. It supports multimodal inputs (text, images, code) with a 128K token context, focusing on agentic reasoning and real-time integration with X (e.g., live tweet analysis). Trained on diverse data up to mid-2025, it features a hybrid architecture for efficient, low-hallucination outputs. Benchmarks show strong results in MMLU (~90%) for reasoning and GSM8K (~92%) for math, with a fun tone. It's designed for engaging, truthful AI on platforms like allmates.ai, blending utility with personality.

🔧 Key Specifications

🎯 Modalities

• Input Supported:

  • Text

  • Images (up to 150 per request, 25MB total; for meme/code analysis)

  • Code (natif support pour génération/débogage)

• Output Generated:

  • Text (primary; with humorous, witty style)

  • Structured outputs (e.g., JSON for tools, tweet summaries)

⭐ Core Capabilities Assessment

• Reasoning & Problem Solving: ⭐⭐⭐⭐☆ (4.5/5; Strong creative reasoning, e.g., 90% on MMLU for fun, logical tasks).

• Writing & Content Creation: ⭐⭐⭐⭐ (4/5; Witty and engaging text, ideal for social or creative content).

• Coding & Development: ⭐⭐⭐⭐ (4/5; Good for scripts, ~85% on HumanEval with humorous comments).

• Math/Sci: ⭐⭐⭐⭐☆ (4.5/5; Excellent in practical math, ~92% on GSM8K).

• Instruct (Instruction Following): ⭐⭐⭐⭐ (4/5; Reliable for dynamic prompts, with truth focus).

• Knowledge: ⭐⭐⭐⭐ (4/5; Strong factual base, enhanced by real-time X data).

🚀 Performance & 💰 Cost

• Speed / Latency: Fast (Throughput: ~65 tokens/sec; Latency: 5s average; Speed Rating: 4/5 – excels in real-time X queries).

• Pricing Tier (on allmates.ai): Accessible

  • Input: $0.40 / 1M tokens
  • Output: $1.50 / 1M tokens
  • (Rating: 4.5/5; Cost-effective for engaging tasks; caching for repeated X data.)

✨ Key Features & Strengths

• Truth-Seeking & Humor: Trained to prioritize facts with witty, sarcastic responses (less censored than GPT).
• Real-Time X Integration: Native access to X data for live analysis (e.g., trend detection in tweets).
• Multimodal Support: Handles images/code for fun tasks like meme generation or code review.
• Agentic Reasoning: Supports tool-use (search on X, code execution) for dynamic workflows.
• Partial Open-Weights: Allows fine-tuning for custom Mates.
• Benchmark Strengths: High in creative reasoning per LMSYS; fun tone boosts engagement.

🎯 Ideal Use Cases on allmates.ai

• Social Media Mates: Analyze live X trends or generate humorous content.
• Creative Brainstorming: Witty writing for marketing or storytelling.
• Real-Time Fact-Checkers: Truthful responses with X-sourced data.
• Code Assistants: Fun, efficient coding with personality.
• Engaging Chatbots: Humorous, fact-based interactions for users.

⚠️ Limitations & Considerations

• Humor Bias: Witty style may not suit formal tasks (tone adjustable via prompts).
• X Dependency: Relies on X for real-time; offline use limited to cutoff.
• Multimodal Limits: Basic image/code support; no full video.
• Hallucination Risk: Low, but verify facts in sensitive areas.
• API Focus: Best via xAI API; integration with allmates.ai for custom Mates.

🏷️ Available Versions & Snapshots (on allmates.ai)

• grok-4 (Alias to the latest stable version).

• grok-4-2025-08 (Specific snapshot for consistent performance).