Google celebrates Gemma 4 31B inference speed with hackathons for small-model coding
English summary
Google is now running hackathons featuring its small model Gemma 4 31B to celebrate record inference speeds of 1500 tokens per second, which is 50–100× faster than what can be achieved locally. This initiative highlights the company's continued belief in the value of small models for AI-assisted software engineering. The event aims to foster coding innovation using efficient, open-source models, aligning with the community’s interest in vibe-coded projects.
Chinese summary
谷歌正在举办以 Gemma 4 31B 小模型为特色的黑客马拉松，庆祝其创纪录的 1500 tokens/秒推理速度，比本地部署快 50–100 倍。这一举措凸显了该公司对小型模型在 AI 辅助软件工程中价值的持续看好。活动旨在促进利用高效开源模型进行编码创新，符合社区对“氛围编码”项目的兴趣。
Key points
Google is hosting hackathons for its small model Gemma 4 31B.
谷歌正为其小模型 Gemma 4 31B 举办黑客马拉松。
The model achieves an inference speed of 1500 tokens/second, 50–100× faster than local setups.
该模型实现 1500 tokens/秒的推理速度，比本地部署快 50–100 倍。
The event signals big tech's support for small-model AI-assisted coding.
这一活动表明大型科技公司对小型模型辅助编码的支持。