Google runs hackathons for small models like Gemma 4 31B

Google is organizing hackathons focused on small language models, specifically the Gemma 4 31B, to demonstrate their value in AI-assisted software engineering. This initiative highlights the company's continued belief in the utility of smaller models despite the industry trend toward larger ones.

• Google is hosting hackathons for the Gemma 4 31B model.
• The event celebrates inference speeds of 1500 tokens per second, which are 50–100 times faster than local capabilities.
• The move signals that major tech players see real value in small-model AI-assisted software engineering.