BigQuery Launches SQL-Native Inference for Open Models, Simplifying AI Deployments

Google’s BigQuery has introduced a new feature that enhances accessibility to various large language models (LLMs) for text and embedding generation, including its own Gemini models and those managed in collaboration with partners like Anthropic and Mistral. This capability, which facilitates the use of these models directly within SQL queries, aims to simplify the deployment and management of generative AI models for users, regardless of their technical expertise. Alongside this, BigQuery is extending its support to models available on platforms such as Hugging Face and Vertex AI Model Garden, marking a significant advancement in database management and AI integration.

With the launch of managed third-party generative AI inference in BigQuery (currently in Preview), users can execute open models with just two SQL statements. This streamlined approach offers four primary advantages: simplified deployment, automated resource management, granular resource control, and a unified SQL interface. The deployment process is designed to be straightforward; users can create an open model by issuing a single CREATE MODEL statement that includes the model ID string, such as google/gemma-3-1b-it. BigQuery handles the provisioning of compute resources automatically, making it accessible even for those less familiar with AI model management.

One of the standout features is automated resource management. BigQuery actively releases idle compute resources, which helps prevent unexpected costs for users. This functionality can be customized through the endpoint_idle_ttl configuration, allowing users to define how long resources should remain active without use. Additionally, users have the option to customize backend computing resources, adjusting parameters like machine types and minimum or maximum replicas within the CREATE MODEL statement. This flexibility ensures that users can tailor the performance and expenses of their models to suit their needs.

To illustrate how the process works, one can create a BigQuery managed open model by first executing a CREATE MODEL statement with the appropriate open model ID. Depending on the size of the model and the chosen machine type, the query typically completes within a few minutes. For models sourced from Hugging Face, users must specify the hugging_face_model_id in a format that includes the provider name and model name—an example being sentence-transformers/all-MiniLM-L6-v2.

This initiative reflects a broader commitment by Google to democratize access to advanced AI capabilities while ensuring that organizations can leverage these technologies without extensive resources or expertise. As generative AI continues to evolve and become more integrated within various sectors, the implications of such developments are significant. The ability to utilize powerful models through a familiar SQL interface could redefine workflows across industries, enabling more users to harness the potential of AI-driven analytics and insights.

Grok: Donovan’s AI Strategy Challenges Shell’s Crisis Management Amid Digital Warfare

AI Reshapes Canadian Hiring: 25% of Employers Reduce Staff Amid Automation Surge

Character.ai Reveals Slonk: A New System Integrating SLURM with Kubernetes for Efficient ML Research

BlueMatrix and Perplexity Launch AI-Driven Research for Regulated Institutional Investors

Invest $100 in the Dan Ives Wedbush AI Revolution ETF for Top AI Stock Exposure Now