Google's New Gemini Model Tops AI Rankings

Google has just released a new AI model called Gemini-Exp-1114, and it's already making waves in the AI community. This model, part of Google DeepMind's Gemini family, is so impressive that it immediately took the top spot on the Imarena Chatbot Arena leaderboard. This platform allows AI labs to compete anonymously, with users voting without knowing which model is which. The new Gemini model matched OpenAI's latest GPT-4o and even outperformed their o1-preview reasoning model. The leaderboard is dominated by models from Google and OpenAI, with xAI's Grok 2 being the first model from a different company. Gemini-Exp-1114 is particularly strong in math and vision tasks, where all Gemini models have traditionally excelled.

The success of this new model coincides with the release of the Gemini app for iPhone, which outperformed ChatGPT in a seven-round face-off. However, the new model is not yet available to the public. You can only access it by signing up for a free Google AI Studio account, which is designed for developers testing new ideas. It's unclear whether Gemini-Exp-1114 is an updated version of Gemini 1. 5 or an early look at Gemini 2, expected next month. While the improvement might not be as dramatic as hoped, it's performing well in technical and creative areas according to benchmarks. This suggests it could be useful for reasoning and managing tasks. The Chatbot Arena is unique because it's based on human perception of performance, not rigid data tests. Whether this is a new version of Gemini 1. 5 Pro or a sneak peek at Gemini 2, the next few months in AI development will be exciting to watch.

Actions