Google has introduced the Gemini-exp-1121 model, a new benchmark in AI with a 2 million token context window. Available in preview mode on Google AI Studio, Gemini-exp-1121 has already achieved top rankings on LM Arena’s leaderboard, demonstrating performance nearing that of OpenAI’s o1-preview model. While impressive, this is not the highly anticipated Gemini 2.0, expected to launch later in December with major advancements.
Google AI Studio has also extended its “Grounding with Google Search” feature to the EU, enabling developers to integrate real-time search results into AI responses. This functionality enhances the factual accuracy and recency of outputs, allowing developers to customize the balance between up-to-date search data and the model’s general knowledge. This feature can be activated in the “Tools” section or through the API by enabling the ‘google_search_retrieval’ tool.
A new experimental tool, Chess Champ Gem, has been added to Gemini’s capabilities. This feature allows users to play chess via text while receiving images of the updated board after each move. Available to all users, including those on free plans, this feature demonstrates Gemini’s ability to generate visuals based on its knowledge. However, it is not yet available on mobile devices.
The Gemini series represents Google’s commitment to advancing multimodal reasoning and flexibility in AI. With the release of Gemini-exp-1121 and the upcoming Gemini 2.0, Google continues to push the boundaries of AI innovation.
