Google 发布 Gemini 3 模型，强化代理式编码与多模态能力
原标题：Start building with Gemini 3
AI 摘要
Google DeepMind 发布了 Gemini 3 系列模型，其中 Gemini 3 Pro 在推理、编码和多模态理解上超越前代，并在 Terminal-Bench 2.0 等基准上取得领先成绩。该模型通过 Gemini API 提供预览，定价为每百万输入 token 2 美元、输出 token 12 美元，并集成于 Google Antigravity 等开发工具。同时，Gemini 3 增强了代理式编码、视觉推理和空间推理能力，支持更广泛的企业应用场景。
正文节选
Start building with Gemini 3 Today we are introducing Gemini 3, our most intelligent model that can help bring any idea to life. Built on a foundation of state-of-the-art reasoning, Gemini 3 Pro delivers unparalleled results across every major AI benchmark compared to previous versions. It also surpasses 2.5 Pro at coding, mastering both agentic workflows and complex zero-shot tasks. Gemini 3 Pro fits right into existing production agent and coding workflows, while also enabling new use cases no