Google's TurboQuant offers LLMs up to 6x compression(arstechnica.com)

https://arstechnica.com/ai/2026/03/google-says-new-turboquant-compression-can-lower-ai-memory-usage-without-sacrificing-quality/