Grok 3

Overview

Grok 3 is xAI's flagship large language model, released February 19, 2025 as the third major iteration of [[Elon Musk]]'s "truth-seeking" AI. It is xAI's most capable general-purpose model and serves as the primary model powering the Grok assistant available to X (formerly Twitter) Premium+ subscribers. Unlike most frontier models, Grok has deep real-time integration with the X platform — it can access live posts, news, and trending topics, giving it a current-events awareness advantage in scenarios where recency matters.

At $3.00/$15.00 per million tokens, Grok 3 is priced identically to Anthropic's Claude Sonnet tier and OpenAI's GPT-5.2 base, positioning xAI as a serious API competitor rather than just a consumer product.

Specifications

• Developer: xAI
• Model String: grok-3 / grok-3-beta
• Release Date: February 19, 2025 (Grok 3); April 9, 2025 (Grok 3 Beta)
• Type: Large Language Model, Multimodal (text + vision)
• Context Window: 131,072 tokens (131K)
• Knowledge Cutoff: November 2024 (with real-time X platform access)
• Access: xAI API, X (Twitter) Premium+, enterprise partnerships
• Pricing: $3.00 per million input tokens / $15.00 per million output tokens

Capabilities

Real-Time X Platform Integration: Uniquely among frontier models, Grok 3 can access live content from X — breaking news, trending topics, and recent posts — giving it current-events grounding that static knowledge cutoffs don't provide.

Reasoning: Strong general reasoning with an explicit "truth-seeking" design philosophy — xAI has emphasized building a model less prone to hedging and more willing to engage directly with controversial or complex questions.

Vision: Handles image inputs for analysis, description, and visual reasoning tasks.

Speed: Generates output at approximately 65 tokens per second — above average compared to other non-reasoning models in the same price tier.

Military & Enterprise Use: xAI has secured U.S. military contracts and enterprise partnerships, indicating production-grade deployment at scale.

Limitations

The 131K context window is smaller than competing models from Anthropic, OpenAI, and Google, which now offer 200K–1M+ tokens. For tasks requiring very large context — full codebase analysis, long document processing — Grok 3 is at a disadvantage. Benchmark scores on the Artificial Analysis Intelligence Index (25 for Grok 3 full, 57 for Grok 3 Mini) suggest the Mini variant outperforms on some benchmarks while costing 90% less.

Recent Developments

• February 19, 2025 Launch: Announced as xAI's most powerful model, positioned as a "truth-seeking" alternative to models from Anthropic and OpenAI.
• Grok 3 Beta (April 9, 2025): A refined beta variant with adjustments to safety behavior and response quality.
• xAI $20B Raise (January 2026): xAI closed a $20B Series E round, the largest recent raise in AI, at ~$230B valuation — significantly expanding its ability to scale compute and model development.
• Tesla & SpaceX Integration: Grok 3 is being integrated into Tesla's in-car AI assistant and other Musk-affiliated products.

Last Updated

February 26, 2026