Gemma 4 deserves a place in task comparisons; making it the default just because the weights are open would be lazy on my part. The April 2026 launch brought four sizes, open weights, and an Apache 2.0 license, including the prospect of local execution. I like having that option. Now it needs a job that justifies the choice.
Looking back at the announcement in October 2026, two later requests explain my interest. In June, I asked for the best available model. In September, I asked for selection to consider the best model for each task. I still want capability, but the word best needs the job I expect it to do attached to it.
The scene is fairly unglamorous: there is work to assign, and I want to know why a particular model should handle it. A familiar name in the menu does not answer that. Neither does a larger size. My September request tied the choice to the role. That is the question I bring to a family with several configurations.
For a local evaluation, I prefer starting with a bounded request, such as finding information in files or examining a small change. That is the comparison I propose for choosing a configuration. I want to define a sufficient answer beforehand. Setting up a demonstration that flatters the candidate and then feeling impressed is far too easy.
Open weights change what I can control. They also introduce decisions about the environment and execution that a ready interface has already made for me. The license makes a deployment worth considering; it does not prepare the machine or maintain the configuration. That effort belongs alongside answer quality because somebody will have to handle it.
Technical curiosity is a legitimate reason to experiment. I just prefer calling curiosity what it is. When another tool already serves the work, a new configuration needs to justify the time I give it. If the purpose is learning to operate a local model, fine: that learning becomes the intended result.
My shortlist starts with the task and records why a particular size deserves a try. I then want to keep the request and result so I can reconsider the choice. Gemma 4 joins as a candidate that expands my operating options. I will compare a bounded task before selecting a size. The license opens the option; the result needs to justify the choice.