Grok 4
by xAI
World's most intelligent model released July 2025, achieving 50% on Humanity's Last Exam and 15.9% on ARC-AGI-2 benchmark. Features native tool use, real-time X integration, and multimodal capabilities with specialized reasoning training using 200,000 GPU cluster.
- World's most intelligent model achieving 50% on Humanity's Last Exam and 15.9% on ARC-AGI-2
- Native tool integration with code interpreter, web browsing, and real-time X platform access
- Advanced multimodal capabilities supporting text and image inputs with voice assistant features
- Reinforcement learning training at pretraining scale using 200,000 GPU Colossus cluster
- Superior benchmark performance with 88% GPQA Diamond, 94% AIME 2024, and 87% MMLU-Pro
Model Information
Supported Formats
Grok 4.3xAI's latest flagship model released April 30, 2026, leading the industry in non-hallucination rate, agentic tool calling, and instruction following. Features 1M token context window with cached input at $0.20/M and supports function calling, structured outputs, image understanding, and live search.
Grok Build 0.1Coding-specialized model optimized for software engineering tasks with 256K context window. Priced at $1.00 input / $2.00 output per 1M tokens - 20% cheaper than Grok 4.3 for code-focused workflows. Ideal for agentic coding, refactors, and large codebase analysis.
Grok 4.20xAI's most advanced flagship model released March 2026 with 2M token context window. Features extended reasoning, multi-agent collaboration, and function calling capabilities. Supports cached input at $0.20/M tokens with 50% batch discount.