Google has unveiled three new experimental AI models, including a smaller 8-billion-parameter version of Gemini 1.5 Flash and improved versions of Gemini 1.5 Pro and Flash.
The company is releasing these models to gather feedback and provide developers with the latest updates. Google Developer Advocate Logan Kilpatrick said the findings from these experimental launches will inform the broader release of the models.
The new Gemini 1.5 Flash 8B model, developed by the Google DeepMind team, is a smaller 8-billion-parameter variant of Gemini 1.5 Flash. It's designed for multimodal tasks ranging from high-volume to long-context summarization.
Gemini 1.5 Flash-8B is now available for free through Google AI Studio and the Gemini API under the name "gemini-1.5-flash-8b-exp-0827".
Google improves its most capable AI model
Updated experimental versions of the Gemini 1.5 Pro and Flash models are also now available. Google reports general improvements in both models, with 1.5 Flash showing significant performance gains across many internal benchmarks and 1.5 Pro demonstrating enhanced capabilities in complex prompts and coding, according to Kilpatrick.
Kilpatrick stated that the new Gemini 1.5 Pro Exp 0827 model outperforms the experimental model released in early August in all aspects. Starting September 3, requests for the "gemini-1.5-pro-exp-0801" model will automatically redirect to the new "gemini-1.5-pro-exp-0827" model. The "gemini-1.5-pro-exp-0801" model will be removed from Google AI Studio and the API.
These models are accessible via the Gemini API and Google AI Studio under the names "gemini-1.5-pro-exp-0827" and "gemini-1.5-flash-exp-0827", as well as through the Vertex AI Experimental Endpoint. Google offers a free tier in the Gemini API for both models and plans to release a version for production use in the coming weeks.
