Developer prunes Gemma 4 to 15% of size, keeps task accuracy

A developer pruned Gemma 4 26B-A4B down to 15% of its size, retaining 92% agreement with the original on one classification task. The resulting dense model has 3.81B parameters, runs at 79 tokens/s on a Mac and uses 3.8GB of memory.

• Google's first data center satellite goes online in orbit2026-10-30
• Community releases Gemma-4-31B-Humanlike-Chat fine-tune2026-10-30
• Google launches local-first AI meeting note-taker Foresight2026-10-08
• Google Releases Open Multimodal Embedding Model EmbeddingGemma 22026-10-06
• Google backs TeraWulf's Kentucky data center for Anthropic2026-11-04