Gemma 4 12B 正式开源: 16GB 显存可跑的多模态模型

Gemma 4 12B 正式开源: 16GB 显存可跑的多模态模型

Source: <https://x.com/demishassabis/status/2062241713398149524&gt;

Author: @demishassabis

Original Date: 2026-06-04

Quality Score: 5

Fetched: 2026-06-05 20:24:48

Summary / 摘要

EN: Google releases Gemma 4 12B, a unified encoder-free multimodal model under Apache 2.0 that runs on laptops with 16GB VRAM. Hassabis notes Gemma line has crossed 150M downloads. The release targets on-device reasoning and offline deployment scenarios.

中文: Google 推出 Gemma 4 12B 多模态模型, Apache 2.0 协议发布, 声称对一台 16GB 显存的笔电即可本地运行. Demis Hassabis 同步宣布 Gemma 系列累计下载量突破 1.5 亿. 模型采用统一无编码器多模态架构, 定位是把高强度推理能力带到边缘设备, 而不是只能跑在云端. 对个人开发者研究者小团队来说, Apache 2.0 加上 16GB 显存门槛意味着本地试验离线部署私有化场景的可行性大幅提高.

Original Tweet / 原始推文

@demishassabis (Wed Jun 03 18:35:13 +0000 2026):

Celebrating the milestone of a massive 150+ million downloads of Gemma 4 with the release of the new Gemma 4 12B model! It's incredibly powerful for such a small model and it’s tiny enough to run locally on a laptop with just 16GB VRAM. Apache 2.0 license - happy building!

*Engagement: 3032 likes, 285 retweets*

中文翻译 / Chinese Translation

Google 推出 Gemma 4 12B 多模态模型, Apache 2.0 协议发布, 声称对一台 16GB 显存的笔电即可本地运行. Demis Hassabis 同步宣布 Gemma 系列累计下载量突破 1.5 亿. 模型采用统一无编码器多模态架构, 定位是把高强度推理能力带到边缘设备, 而不是只能跑在云端. 对个人开发者研究者小团队来说, Apache 2.0 加上 16GB 显存门槛意味着本地试验离线部署私有化场景的可行性大幅提高.

Notable Replies / 热门回复

@Gauri_the_great (❤️ 4):

@demishassabis So now we are calling 12 B params tiny

@boredmotivation (❤️ 1):

@demishassabis https://t.co/SMgmaCOEqu

@S_Fadaeimanesh (❤️ 2):

@demishassabis 12B at fp16 needs ~24gb just for weights, 16 means quantized

@ValeriusLabs (❤️ 2):

@demishassabis Meta shipped llama 4 scout with 17b params and 100k context on a single 4090 two weeks ago.

This is catch-up, not leadership.

@ShahzaibAli3029 (❤️ 1):

@demishassabis Wish there wasn't 30 seconds cap on audio input 😔😔

*Content fetched via opencli twitter thread command. Entry ID: af94b63c*