Google's Gemma 4 is turning heads in the local AI community for something less flashy than raw benchmark scores: it's significantly better at European languages than its predecessors. Danish, Dutch, French, and Italian are all getting specific praise, with users reporting noticeably improved grammar, vocabulary, and overall fluency.
This matters because most open-source AI models - models you can download and run on your own hardware, without sending data to a cloud service - have historically been trained on predominantly English text. Non-English performance tends to degrade the further a language gets from English in structure or dataset size. Danish, with roughly 5.5 million speakers, is exactly the kind of language that smaller models tend to mangle.
Google hasn't published a dedicated multilingual benchmark for Gemma 4 versus Gemma 3, so these assessments come from hands-on testing by practitioners rather than controlled evaluations. But the consistency of feedback across multiple languages suggests this isn't anecdotal noise.
For businesses and freelancers building AI workflows for European markets, better local model quality has a direct dollar value. Running inference locally - meaning the model processes your request on your own machine rather than a remote server - eliminates per-query API costs and keeps sensitive data off third-party servers. If Gemma 4's multilingual performance holds up under broader testing, it becomes a more credible option for document processing, customer support, and content drafting in languages that were previously underserved by open models.
Gemma 4 is available through Google's model hub and can be run locally via tools like Ollama.