Grok

Grok is xAI's family of large language models, served in xAI's Grok apps, on X and through the xAI API. xAI's timeline dates a first prototype, Grok-0, to August 18, 2023, the announcement of Grok to November 3, 2023, and the public beta of its API to November 4, 2024.¹ On March 17, 2024, xAI released the base-model weights and network architecture of Grok-1, a 314-billion-parameter mixture-of-experts model with 25% of its weights active on a given token, under the Apache 2.0 license.² By the same timeline, Grok 4 followed on July 9, 2025, Grok 4.5 on July 16, 2026, Grok 4.6 on August 12, 2026 and Grok 4.7 on September 21, 2026.¹ xAI calls Grok 4.7 its "most capable model for coding and knowledge work" and says it uses a new, larger base model than Grok 4.6.³

xAI's API documentation lists Grok 4.7 with a 500K-token context window and a May 2026 knowledge cutoff, and Grok 4.3 with a 1M-token window.⁴ Its price list puts Grok 4.7 at $2.00 input and $6.00 output per 1M tokens and Grok 4.3 at $1.25 and $2.50, with every token of a prompt of 200K tokens or more billed at higher long-context rates.⁵ By output price, Grok 4.3 ranks 9th and Grok 4.7 14th of 19 models in Cheapest LLM APIs per 1M Output Tokens (October 2026).⁵

On ARC-AGI-3, ARC Prize measured Grok 4.6 at 2.11% on the Semi-Private set at XHigh effort, at a total cost of $5,612, and Grok 4.7 at 1.82% with its Standard harness; they rank 7th and 8th in Highest ARC-AGI-3 Scores, by Model (Standard Harness).^6,7 With ARC Prize's Provider Adapter harness, which carries reasoning state between requests, Grok 4.7 reaches 10.0% at High effort, a result that table doesn't rank against Standard-harness scores.⁷