Gemma 2 9B → Gemma 4 E4B: Local LLM in a Commercial Indie Game

Solo indie developer from Japan upgraded the local LLM powering the commercial game Hex Judge from Gemma 2 9B to Gemma 4 E4B, cutting AI conversation startup time on an RTX 2060 from about 45 seconds to less than 1 second. The game, a medieval witch trial investigation title supporting 12 languages, now features smoother NPC conversations and improved multilingual dialogue quality, all running locally on the player's PC without an internet connection or external AI API.

I’m a solo indie developer from Japan, and I recently upgraded the local LLM powering my game Hex Judge from Gemma 2 9B to Gemma 4 E4B . Hex Judge is a medieval witch trial investigation game where players interrogate villagers through AI-generated dialogue and uncover contradictions, relationships, and secrets. One of the biggest improvements has been conversation startup time . On an RTX 2060 , starting an AI conversation went from around 45 seconds with Gemma 2 9B to less than 1 second with Gemma 4 E4B . Gemma 2 9B → Gemma 4 E4B comparison: Gemma2-9BとGemma4-E4BをHex Judgeで比較してみた【ローカルLLM】 https://www.youtube.com/watch?v=IbQ97PQwicQ We also experienced: • Smoother NPC conversations • Improved multilingual dialogue quality • Faster AI interaction The LLM runs locally on the player’s PC , without requiring an internet connection or an external AI API. Hex Judge currently supports 12 languages. This is my real-world experience using Gemma 4 E4B as the local AI powering NPC conversations in a commercial game . I’m excited to see what Gemma 4 E4B can enable for local AI in games. Hex Judge is available on Steam. Gemma4 Gemma LocalAI GameDev IndieGame