Grok
lastmod 2025-07-11created 2025-02-194 edits
Grok is a generative AI chatbot developed by xAI. It is based on an LLM and was developed as an initiative of Elon Musk in direct response to the rapid growth of ChatGPT, developed by OpenAI, which Elon Musk co-founded.
Grok 4
- Grok 4 released | GeekNews
- Simon Willison's Grok 4 review | GeekNews
- Grok 4 is xAI's latest large language model, released via API and paid subscription; its key features are image and text input, text output, and support for a 256,000-token context length
- It outperformed competing models (OpenAI o3, Gemini 2.5 Pro, etc.) on major benchmarks and recorded the highest score in independent evaluation with an AAI Index score of 73
- It can generate and describe images, but has limits in fine-grained quality, such as being unable to accurately describe the images it generated
- Recent controversy over Grok 3 system prompt updates (e.g., antisemitism, references to MechaHitler) has raised concerns about the model's safety and reliability
- Pricing is usage-based (input $3/million tokens, output \(15/million tokens), split into a standard subscription (\)30/month, $300/year) and a premium tier (Grok 4 Heavy $300/month, $3,000/year)
Grok 4 is xAI's latest AI model, released after about two years, achieving intelligence and reasoning that surpass graduate students in every field
- Training scale and compute resources increased more than 100-fold; development centered on reinforcement learning (RL), demonstrating problem-solving ability beyond human level
- Achieved an ARC-AGI score of 15.9%, recording outstanding results in abstract reasoning and general intelligence evaluation at the highest level among existing AI
- On various benchmarks such as Humanity’s Last Exam (HLE), it showed groundbreaking results of 26.9% without external tools and 41–50.7% with tools
- Introduced a native voice mode, enabling near-human interaction with real-time conversation, emotional expression, and low-latency responses