Google DeepMind发布Gemini 3.8 Flash模型

Google DeepMind在2026年9月发布了Gemini 3.8 Flash模型，这是Gemini 3系列的最新迭代版本，基于Gemini 3.7 Flash构建1。该模型在软件工程和智能体知识工作流程方面实现了性能提升1。

Gemini 3.8 Flash支持文本、图像、音频和视频文件输入，具备100万token的上下文窗口容量和64K token的文本输出限制1。模型的知识截止日期为2026年3月，部分领域可能仅限于2025年1月1。该模型已通过Gemini app、Gemini Enterprise Agent Platform、Google AI Studio、Gemini API等多个渠道向用户提供1。

在安全性方面，Gemini 3.8 Flash与Gemini 3.7 Flash相比在安全性和语气表现相似，具有低拒绝率，并通过了自动化红队测试和儿童安全评估1。

Google DeepMind has unveiled Gemini 3.8 Flash, the latest iteration in its Gemini 3 series built upon the Gemini 3.7 Flash foundation 1. The new model demonstrates performance improvements in software engineering and agentic knowledge workflows, while maintaining a context window capacity of up to 1 million tokens and supporting text outputs of up to 64,000 tokens 1.

The model accepts multiple input modalities including text, images, audio, and video files 1. Gemini 3.8 Flash has a knowledge cutoff date of March 2026, with certain domains limited to January 2025 1. The model will be distributed across multiple platforms, including the Gemini app, Gemini Enterprise Agent Platform, Google AI Studio, Gemini API, Google AI Mode, and Google Antigravity 1.

Regarding safety, Gemini 3.8 Flash demonstrates performance comparable to its predecessor in security measures and tone, featuring a low refusal rate and having undergone red team testing and child safety assessments 1. The model was released in September 2026 1.