On Wednesday, Google unveiled the next generation of its artificial intelligence systems, Gemini 2.0. The first model of the family, Gemini 2.0 Flash, is now available as a chatbot for public testing. Developers have also gained access to an experimental multimodal version, capable of converting text to speech and generating images.
Sundar Pichai, Google’s CEO, highlighted the key distinction between this version and its predecessor. While Gemini 1.0 was designed to gather and analyze information, Gemini 2.0 aims to assist users in applying that knowledge effectively.
The new language model surpasses previous versions in nearly all scenarios. It excels at writing code and provides more accurate answers to questions. However, it still lags behind the Gemini 1.5 Pro model in handling lengthy texts.
Users can interact with the optimized Flash 2.0 version through the Gemini web interface by selecting the model from a dropdown menu on a computer or mobile device. It will soon be integrated into the mobile application as well.
Developers are already testing the multimodal Gemini Flash 2.0 on Google AI Studio and Vertex AI platforms. Google plans to roll it out to all users in January 2024, alongside new variants with different configurations.
By early 2025, Gemini 2.0 technologies are expected to be incorporated into other Google products. The company is vying for leadership not only against tech giants like Microsoft and Meta but also emerging competitors such as OpenAI (creators of ChatGPT), Perplexity, and Anthropic (developers of Claude).
Alongside the Flash model, Google also introduced prototypes of its “AI agents.” These are described as assistants with a deeper understanding of the world, capable of anticipating the consequences of their actions and performing tasks on behalf of users—though strictly under user control.
At the recent DealBook Summit hosted by The New York Times, Sundar Pichai confidently asserted the superiority of Google’s technologies. He even invited Microsoft to an open comparison of their AI models’ capabilities at a time of their choosing.
