Some people use AI / LLMs to completely vibe code something whereas some use it simply as a tool to augment their own capabilities.
Regardless of the use case, running models locally can be confusing with all the technical requirements, token jargon, and metrics.
In this article, we dive into what it takes to actually running various models locally and what those context windows actually mean in terms of capability and requirements. **Note**, lots of jargon inside, but we tried to explain to the best of our minimal understanding of the tech, our understanding of WTF the numbers meant.
Also included in the article, a simple bash script to run your own performance tests on your own hardware!
We looked at a few models, including DeepSeek, Qwen, Codestral, and Gemma.
https://bitsnbytes.dev/posts/2026-10/comparison-and-tuning-local-ai.html