According to DeepLearning.AI, Google's launch of its Gemma 3 family of large language models, including multimodal versions that process text, images, and video, could significantly enhance AI-driven trading tools. Available for free with open weights, these models range from 1B to 27B parameters, and their ability to run on consumer hardware can democratize access to advanced trading algorithms, potentially influencing high-frequency trading strategies and market analysis tools.

According to Jeff Dean, the release of Gemma 3, an open source model with 27B parameters, marks a significant advancement. It is multimodal, efficiently fits in a single H100 GPU, and performs exceptionally well on TPUs, indicating potential for broad application in AI-driven trading algorithms.

According to Sundar Pichai, Gemma 3 introduces groundbreaking efficiency in AI models, with its largest 27B model operating on just one H100 GPU. This represents a significant leap, requiring at least 10 times less computational power for similar performance compared to other models, potentially reducing operational costs and energy consumption for traders and developers.