xAI released Grok 4 Fast — its "reasoning" version "outperformed" Claude Opus 4.1 in independent tests

It supports a context of 2 million tokens.

Grok 4 Fast results in tests with and without 'reasoning' mode. Source: Artificial Analysis
Grok 4 Fast results in tests with and without 'reasoning' mode. Source: Artificial Analysis
  • The model is available for free on the website and in the Grok mobile apps for iOS and Android. In tests by independent researchers at Artificial Analysis, its 'reasoning' version matched Gemini 2.5 Pro and slightly surpassed Claude Opus 4.1. In search mode, it outperformed o3 and GPT-5 from OpenAI.
  • According to Artificial Analysis, the model from xAI uses tokens more economically than its competitors. It took 61 million tokens to solve all tasks in the test, compared to 93 million for Gemini 2.5 Pro. Thanks to this efficiency, working with Grok 4 Fast in the API is 23 times cheaper than GPT-5-Thinking-high and 25 times cheaper than Gemini 2.5 Pro, the researchers write.
  • The price in the API for developers is $0.2 per 1 million input tokens and $0.5 per 1 million output tokens for requests up to 128 thousand tokens. Longer requests cost $0.4 and $1 per 1 million tokens, respectively.
  • Grok 4 Fast will be available for free testing on the application creation platforms OpenRouter and Vercel AI for a 'limited time'.
A user tested how Grok 4 Fast would write code for an interactive Harry Potter-themed website. Source: Techikansh
For comparison — the result from GPT-5 mini for the same request. Source: Techikansh

#news