A Reddit user shared a comparison between the Qwen3.5B language model and similarly sized models available about two years ago. The image generated with Gemini shows a significant improvement in performance.

Progress in the field of LLMs

The post highlights how the ~9 billion parameter models from two years ago were barely usable, while Qwen3.5B offers significantly superior performance. This progress is significant because it allows more powerful models to be run even on less powerful hardware or in resource-constrained contexts.

Implications for local use

The increased efficiency of models like Qwen3.5B makes it more feasible to use LLMs locally, paving the way for new applications and use cases that require low latency, data sovereignty, or simply the absence of a reliable internet connection.