Gemma 4 lands Apache 2.0, Ollama

The license change is the real unlock here: Apache 2.0 removes a lot of friction for teams that want to fine-tune, ship commercially, or keep inference fully local.

• –Google’s launch positions Gemma 4 as its most capable open model family, with day-one support across Ollama, llama.cpp, LM Studio, Docker, Hugging Face, and more
• –The benchmark story matters, but the useful question is whether those gains hold up in real developer workflows like coding, retrieval, and tool use, not just leaderboard slices
• –Local runners care less about hype than model size, quantization, and hardware fit; Gemma 4’s edge-to-server spread makes it easier to pick a version that matches the machine you actually have
• –Apache 2.0 plus local-first deployment pushes Gemma closer to a true open-model alternative to proprietary stacks