Your guide to Provisioned Throughput (PT) on Vertex AI

Vertex AI's Provisioned Throughput (PT) is updated to guarantee consistent AI agent performance, crucial for business needs. The updates offer increased model diversity, allowing users to select the optimal model from the Vertex AI Model Garden. Multimodal innovation is supported with enhanced PT for processing text, images, and video. Operational flexibility is improved through short-term contracts and proactive capacity planning. PT now supports Anthropic models through direct purchasing and management from the Vertex AI console, streamlining workflows. Open-source models like Llama 4, Qwen3, and others receive PT support under a unified governance framework. The Gemini Live API benefits from PT, offering guaranteed throughput for demanding multimodal streams. Flexible terms and proactive scheduling options make scaling more dynamic and efficient. PT integrates with caching for cost-effective handling of long, repetitive contexts. Several customers like Reve AI, Knowunity, and others are already benefiting from PT.