AI expert Andrej Karpathy tested Grok 3 and wrote on X: “Insofar as a quick atmosphere control over ~ 2 hours this morning this morning, Grok 3 + feels somewhere in the state of the artificial area of the strongest models of OpenAI (O1-Pro, $ 200 /month), and slightly better than Deepseek-R1 and Gemini 2.0 Flash Thinking.
X premium+ subscribers who pay $ 50 monthly, receive first access to grok 3. Leaks suggest that a new super grock plan per month will be $ 30 or $ 300 annually, making subscribers extra functions, including unlimited image generation.
A family with multiple models
Just like AI models from other companies, the Grok 3 -Family contains different models, including a smaller “mini” version that trades the accuracy for speed. Xai claims that Grok 3 performs better than GPT-4O from OpenAI about certain mathematicians and scientific benchmarks, including Aime and GPQA, who test physics at graduated level, biology and chemistry knowledge.
Two models in the family, grok 3 reasoning and grock 3 mini-reasoning, simulated reasoning functions comparable to OpenAI's O3-Mini and Deepseek's R1 models. Users have access to this via a “Think” assignment or “Big Brain” mode in the Grok app. In addition, the grok -app now contains “DeepSearch”, a research tool that is looking for on the internet and X platform to make summaries of information, similar to Google and OpenAi's deep research functions.
Xai is planning to add speech synthesis to the Grok app within a week and to launch an Enterprise API with DeepSearch capacities in the coming weeks. The company says it will also open the previous grok 2 model as soon as grok 3 stabilizes, which Musk estimates will take a few months.
