Google announced three new experimental AI models: Gemini 1.5 Flash-8B, a smaller variant of the Gemini 1.5 Flash model; a new Gemini 1.5 Pro model with improved coding and complex prompt performance; and a new Gemini 1.5 Flash model with significant improvements. Google is releasing these experimental models to gather feedback and release them more widely in the future.
Gemini 1.5 Flash-8B
Since its release, the Gemini 1.5 Flash model has quickly gained popularity among developers. Google is now trying to capitalize on that by releasing an 8-billion parameter variant of the Gemini 1.5 Flash model. This new smaller model can be used for high-volume multimodal use cases, long-context summarization tasks, and more. Developers can try out the Gemini 1.5 Flash-8B experimental model for free via Google AI Studio and the Gemini API today using "gemini-1.5-flash-8b-exp-0827."
Gemini 1.5 Flash and Gemini 1.5 Pro updates
Both of these new Gemini 1.5 models include significant performance improvements. The newest Gemini 1.5 Flash model delivers overall improvements on many benchmarks, while the newest Gemini 1.5 Pro model features strong improvements across complex prompts and coding. Both of these models can be accessed via the Gemini API and Google AI Studio using the names "gemini-1.5-pro-exp-0827" and "gemini-1.5-flash-exp-0827," respectively. These models will be available for production use in the coming weeks.
Chatbot Arena results
Chatbot Arena"s results for these new Gemini 1.5 models are now live with over 20,000 community votes. As expected, these models delivered strong results. The new Gemini-1.5-Flash (0827) is now ranked No. 6, up from No. 23. The new Gemini 1.5 Flash-8b outperforms Google"s Gemma-2-9b and matches Llama-3-70b"s performance. The new Gemini 1.5 Pro (0827) delivered strong gains in coding (now ranked No. 4 in the coding arena) and math when compared to previous Gemini 1.5 Pro models.
You can check out these new Gemini 1.5 models at Google AI Studio. Overall, these new models showcase Google"s continued efforts to push the boundaries of AI capabilities and gather valuable user feedback for further development.