xAI has provided an important update for the recently launched Grok-2 and Grok-2 Mini. Thanks to the rewriting of their code, it was stated that the artificial intelligence models experienced a two-fold performance increase. This shows that it can generate much faster responses.
Performance update released for Grok-2 and Grok-2 mini
According to a post by xAI researcher Igor Babuschkin, Grok-2 Mini is twice as fast as it was two days ago. Babuschkin said that all the inferences of the models were rewritten using SGLang, an open-source system.
In addition to the performance increase, the researcher said that the models were able to give more accurate answers thanks to the new code. This allows xAI to host the Grok-2 model on multiple servers and test it with more users.
Performance gains also made a difference in the leaderboard in the Lmsys Chatbot Arena, which tests artificial intelligence agents. Accordingly, Grok-2 ranked second in the overall ranking with 1293 points. While going head-to-head with Google Gemini, it left OpenAI’s ChatGPT 4o model behind.
Grok-2 Mini is in fifth place with 1268 points, leaving models like GPT-4o Mini behind. According to Arena, Grok-2 has a particular lead in mathematical tasks. Both models’ high rankings reflect areas such as coding, instructions and reasoning.
The AI model, which allows you to create visuals with new image creation features and perform daily tasks with Grok, is only available with Premium. With a ChatGPT-like theme, you can switch between Grok-2 versions at the top, and you can also switch to Fun Mode.