Ollama Gemma4:26b

A high-performance 26B Mixture of Experts model that delivers flagship intelligence with only 3.8B active parameters per token.

Gemma 4:26b is a workstation-class Mixture of Experts (MoE) model released by Google DeepMind in April 2026. By activating only 3.8 billion of its 26 billion total parameters during inference, it achieves near-flagship performance (88.3% on AIME 2026) while maintaining the speed and memory efficiency of a much smaller model. It features a massive 256K context window, native multimodal support for text and images, and a dedicated thinking mode for complex reasoning. Optimized for local execution via Ollama, this Apache 2.0 licensed model fits comfortably on a single 24GB consumer GPU like the RTX 4090, making it the ideal choice for private, agentic workflows and advanced coding assistance.

https://ollama.com/library/gemma4:26b