Keftek

Google Releases Gemma 4: Open Trimodal Models With 256K Context

Apache 2.0 trimodal (text, vision, audio) family with 128K–256K context, 140+ languages, and agentic-workflow tuning. Self-hostable foundation for SME automation stacks.

LLM InfraMulti-ModelAI Agents
Read original on Google DeepMind