13:02
12:45
16:21
12:24
16:44
11:38
13:02
12:45
16:21
12:24
16:44
11:38
13:02
12:45
16:21
12:24
16:44
11:38
13:02
12:45
16:21
12:24
16:44
11:38
xAI, the artificial intelligence company founded by Elon Musk, has introduced two new AI models: Grok 4 and Grok 4 Heavy.
Both models have demonstrated impressive performance across key benchmarks and, according to xAI, represent a significant leap forward in understanding, reasoning, and complex content generation.
Grok 4 scored 41% on the Humanity Last Exam — one of the most challenging tests of abstract reasoning and multidisciplinary knowledge. That’s nearly double the performance of models like OpenAI’s ChatGPT o3 (21%) and Google’s Gemini 2.5 Pro (21.6%). The boost is largely attributed to Grok 4’s advanced reasoning mode, which allows the model to dynamically access search tools and other external functions during response generation — a capability previously seen only in GPT-4o.
Grok 4 Heavy takes things further. In a mode where multiple agents within the model collaborate and exchange results in parallel, it achieved a 50.7% score on the same exam. While this approach demands significantly more computational power, it enables deeper and more accurate reasoning. Grok 4 Heavy will be offered via a separate SuperGrok Heavysubscription.
xAI also shared its upcoming roadmap: a developer-focused Grok 4 Coder model is expected in the coming weeks, alongside continued work on multimodal features for Grok 4 — including a planned model for video generation.