Google introduces Veo 3 Fast, an optimized model for speed and price, along with new image-to-video capabilities for both Veo 3 and Veo 3 Fast, enabling developers to efficiently create high-quality video content from text or still images, with varying pricing based on the model and audio inclusion, now available in the Gemini API.

LangExtract is a new open-source Python library powered by Gemini models for extracting structured information from unstructured text, offering precise source grounding, reliable structured outputs using controlled generation, optimized long-context extraction, interactive visualization, and flexible LLM backend support.

The Gemini Embedding model enhances AI applications, particularly through context engineering, which is being successfully adopted by various organizations across industries to power context-aware systems, leading to significant improvements in performance, accuracy, and efficiency.

Gemini 2.5 Flash-Lite 之前处于预览版阶段，现已稳定并面向公众发布。这款具有成本效益的模型比 2.0 Flash-Lite 和 2.0 Flash 约快 1.5 倍，质量上乘，并具备 2.5 系列的功能，如 100 万个令牌的上下文窗口和多模态。

Vertex AI 现已为 Gemini API 正式推出 logprobs 功能。该功能通过展示模型所选 token 及其替代 token 的概率分数，深入揭示了模型的决策过程。本分步指南将详细介绍如何启用和解读此功能，并将其应用于强大的用例，例如：置信度分类、动态自动补全以及 RAG 的定量评估。