Google Gemma 4 Ships Under Apache 2.0

This is a meaningful licensing shift, not just a model bump: Apache 2.0 makes Gemma 4 far easier to adopt in commercial products and internal stacks. The size spread also looks intentionally practical, with small on-device variants for local agents and larger models for heavier reasoning workloads.

• –The Apache 2.0 license is the headline: it lowers legal friction for teams that want to fine-tune, bundle, and ship the models commercially
• –Four variants cover different deployment tiers, which makes the family easier to slot into real products instead of forcing one model to do everything
• –Native multimodal input, long context, and function-calling make Gemma 4 more interesting for agentic apps than for simple chat benchmarks
• –Google is clearly leaning into a two-track strategy: open weights for flexibility, Gemini for flagship proprietary capability
• –The benchmark claims are strong, but the real test will be latency, safety tuning, and how well these models hold up outside curated demos