Gemma 4 models now available with quantization-aware training (QAT), improving inference efficiency. Multiple weight configurations being tested (Q4_k_M, QAT variants).
Community testing shows DeepSeek V4 Flash delivering strong performance with llama.cpp optimization (PR #24162). Active development in quantization optimization.
GitHub Copilot now supports custom model endpoints, allowing developers to use alternative LLMs or self-hosted models for code completion.
Google committed to a $920 million monthly cloud compute partnership with SpaceX, representing massive infrastructure investment for AI workloads.