Grok 4.20: a Reliable AI Model but Outpaced by Gemini and GPT-5.4

Grok 4.20, the latest artificial intelligence model developed by xAI, struggles to compete with market leaders such as Gemini 3.1 Pro Preview and GPT-5.4. In benchmarks, Grok 4.20 scored 48 on the Intelligence Index, according to the report from Artificial Analysis. This figure is significantly lower than the 57 points achieved by its competitors, although Grok 4.20 improved its score by 6 points compared to its predecessor, Grok 4.

Despite its overall lower performance, Grok 4.20 stands out for its factual reliability. In the AA Omniscience test, which assesses a model's ability to avoid hallucinations and recall accurate facts, Grok 4.20 recorded a non-hallucination rate of 78%, a record according to Artificial Analysis. This means that the model was wrong only once in five times when it did not know the answer, a notable result in the field of AI.

xAI has introduced three API variants for Grok 4.20: one with reasoning, one without reasoning, and one in multi-agent mode. The model supports a context window of 2 million tokens, which is a significant advancement. In terms of cost, Grok 4.20 is offered at a competitive rate of $2 to $6 per million tokens, making it more affordable than its predecessor and competitive with Western models.

In summary, while Grok 4.20 may not be at the forefront in terms of raw performance, its ability to provide reliable factual information could make it a strategic choice for applications requiring high precision.