Google fulfills its promise to open its most powerful AI model, Gemini 1.5 Pro, to the public after releasing a beta version for developers last month.
Google’s Gemini 1.5 Pro is capable of handling more complex tasks than previous AI models, such as analyzing entire text libraries, long Hollywood movies, or almost a full day of audio data. It has 20 times the data of OpenAI’s GPT-4o and nearly 10 times the information management capability of Anthropic’s Claude 3.5 Sonnet.
In its announcement, Google stated its goal of delivering faster and more cost-effective tools to AI developers, enabling “new use cases, additional production robustness, and higher reliability.”
Image: Google
Earlier in May, Google introduced the model, showcasing a group of selected testers in a video demonstrating how they utilized its capabilities. For example, machine learning engineer Lukas Atkins provided the entire Python library to the model and posed questions to help him solve problems. He said in the video, “It gets it. It can find specific references in the code and specific requests from people.”
Another tester filmed a video of his entire bookshelf, and Gemini created a database of all his books—an almost impossible task for traditional AI chatbots.
Gemma 2 Dominates the Open Source Space
However, Google also made waves in the open-source community. According to LLM Arena rankings, the company released Gemma 2 27B today, an open-source large-scale language model that quickly claimed the throne of open-source models with its top-quality and rapid responses.
Google claims that Gemma 2 offers “best-in-class performance, runs at astonishing speeds on different hardware, and integrates easily with other AI toolsets.” The company states that it aims to compete with models twice its size.
Image: Google
Gemma 2’s license allows free access and redistribution, but it differs from traditional open-source licenses like MIT or Apache. The model is designed for easier access and budget-friendly deployment of AI in the 27B and smaller 9B versions.
This is important for both regular users and enterprise users because, unlike closed models, powerful open models like Gemma are highly customizable. This means users can fine-tune their models to excel at specific tasks and protect their data by running these models locally.
For example, Microsoft’s small-scale language model, Phi-3, has been fine-tuned specifically for math problems and can outperform larger models like Llama-3 and even Gemma 2 in that domain.
Image: Microsoft
Gemma 2 is now available in the Google AI studio, and model weights can be downloaded from Kaggle and Hugging Face Models. The powerful Gemini 1.5 Pro is available for developers to test on Vertex AI.
