Local models went from mostly useless to actually useful really fast. What changed?

Local models have significantly improved in usability over the past year, transitioning from primarily privacy-focused tools to effective solutions for coding, document management, and local workflows. Key advancements include better base models like Gemma, Qwen, GLM, and Kimi, improved quantization techniques, and enhanced tools such as llama.cpp and Ollama, which have collectively increased performance and functionality. These developments are crucial for practitioners as they enable more robust local deployment of AI models, allowing for greater flexibility and reduced reliance on external APIs.