Google said it is improving core capabilities in Gemini through research focused on factuality, multilinguality and efficiency, with work done in collaboration with Google DeepMind. The company said these efforts are helping improve Gemini model quality and performance, and expand global access to its products.
Google said its research on LLM factuality dates back to pioneering work on evaluating factual consistency in 2021 and an early benchmark in 2022. It has published FACTS and extended it to support benchmarking of factuality in LLMs, along with techniques to improve factuality in text-to-image, video generation, long-context and expressions of uncertainty.
The company said longer, more complex information journeys are creating new challenges for LLMs, including reasoning over more relevant information in the context window, following constraints set early in a conversation, and using longer reinforcement learning trajectories. Google Research has worked on those issues, and Google said the advances feed into Gemini models.
Google also said it partnered on the new Ask Maps feature to upgrade its evaluation framework and redefine how map helpfulness is measured. The collaboration focused on complex edge cases involving model reasoning and tool execution. Google said it also drove research to improve the quality of Ask YouTube, a new feature that helps users find videos and information.
On multilinguality and localization, Google said it published a benchmark showing how LLMs operate in different languages and in different locations, and that it has been open sourcing data in African languages with the community. Google said these efforts helped expand Gemini to more than 70 languages across more than 230 countries, making it the most widely available AI assistant in the world.
The company also said it has developed new techniques built on speculative decoding, including block verification and tree-structured drafting. Google said the implementation is optimized for its TPU architecture and delivers substantially faster responses with no loss in quality. This work enabled the current speed of Gemini 3.5 Flash, with the same models also powering Antigravity and AI Studio.
Source: research.google.
Companies can share verified announcements through Newz9’s international press release submission page.
