Google debuts top-ranking Gemini-exp-1121 model with 2M context window

· 1 min read
AI Studio

Google has recently announced the Gemini-exp-1121 model, which features an impressive 2 million token context window. This model is now available in preview mode on Google AI Studio. It has been tested extensively on platforms such as LM Arena, where it has climbed to the top of the leaderboard. The performance of Gemini-exp-1121 is approaching that of OpenAI’s o1-preview model. However, this is not the anticipated Gemini 2.0 release. The official Gemini 2.0, expected later in December, is rumored to bring significant performance improvements.

Additionally, Google AI Studio has expanded the availability of its “Grounding with Google Search” feature to EU. Previously available in select regions, this feature now allows developers worldwide to ground their AI responses in real-time search results. Developers can specify the degree to which responses draw from up-to-date search data versus the model’s general knowledge. This capability aims to enhance the factual accuracy and recency of AI outputs.

AI Studio

The grounding feature can be enabled directly in Google AI Studio under the “Tools” section or via the API by activating the ‘google_search_retrieval’ tool. By integrating real-time search data, this feature reduces instances of hallucinations and ensures AI responses are more reliable and contextually relevant, bolstering user trust in AI applications.