Google Unveils Experimental Gemini 1.5 AI Models: Pro & Flash Updates

Google has launched three experimental AI models, including a smaller 8-billion-parameter version of Gemini 1.5 Flash and updated versions of Gemini 1.5 Pro and Flash. These models are now available for testing, and Google is seeking feedback to guide their further development.

Gemini 1.5 Flash 8B: Compact and Versatile

The Google DeepMind team has developed the Gemini 1.5 Flash 8B, a more compact 8-billion-parameter variant designed for a variety of tasks. This model excels in handling multimodal assignments, such as large-scale data processing and summarization over extended contexts. It's available for free through Google AI Studio and the Gemini API, labeled as "gemini-1.5-flash-8b-exp-0827."

Enhanced Gemini 1.5 Pro and Flash Models

Google has also rolled out updated versions of the Gemini 1.5 Pro and Flash models. Logan Kilpatrick, a Google Developer Advocate, reported general enhancements in both models. The updated 1.5 Flash shows significant performance improvements across several internal tests, while the 1.5 Pro model has been optimized for more complex tasks, including advanced coding and handling intricate prompts.

Transitioning to the Latest Gemini 1.5 Pro

The new Gemini 1.5 Pro Exp 0827 model is set to replace its predecessor, released in early August. Starting September 3, all requests for the "gemini-1.5-pro-exp-0801" model will be automatically redirected to the updated "gemini-1.5-pro-exp-0827" model. After this transition, the older version will no longer be available on Google AI Studio or through the API.

Access and Future Developments

These new models can be accessed via the Gemini API and Google AI Studio, under the names "gemini-1.5-pro-exp-0827" and "gemini-1.5-flash-exp-0827." They are also available through the Vertex AI Experimental Endpoint. Google offers a free tier for both models in the Gemini API and plans to release a production-ready version soon.