谷歌发布 Gemini 3：基准测试全面超越前代，编程能力再度突破
Start building with Gemini 3
谷歌发布 Gemini 3 系列，其中 Gemini 3 Pro 在所有主流 AI 基准测试中均超越前代，编程能力更胜 2.5 Pro，熟练掌握智能体工作流和复杂零样本任务。该模型支持多模态理解、视觉推理和空间推理，可通过 Google AI Studio 和 Vertex AI 中的 Gemini API 调用，也可使用 Google Antigravity 平台进行智能体开发。Gemini 3 还在 AI Studio 中解锁了"vibe coding"能力，用户可通过自然语言提示直接创建应用。
Google is launching Gemini 3, its most intelligent model that helps bring any idea to life. Gemini 3 Pro delivers unparalleled results across every major AI benchmark compared to previous versions and surpasses 2.5 Pro at coding, mastering both agentic workflows and complex zero-shot tasks. It integrates into existing production agent and coding workflows while enabling new use cases in multimodal understanding, visual reasoning, and spatial reasoning. Developers can access Gemini 3 Pro through the Gemini API in Google AI Studio and Vertex AI, or use the Google Antigravity platform for agentic development. The model also unlocks ‘vibe coding’ in Google AI Studio, letting users create apps from natural language prompts.
对 AI 行业的影响
Gemini 3 的发布标志着多模态 AI 竞争进入新阶段。编程能力再进化意味着 AI 辅助开发的可靠性进一步提升，智能体工作流的产业落地有望加速。
原文参考
来源：Google DeepMind Blog · 2025-11-18