Phi4-mini has done well on the test. Foundational-Sec8B failed. Qwen3 is good except that the memory consumption is very tight for my machine.
In that regard the only local AI models will be Phi4-Mini from Microsoft for high frequency low latency task such as coding. Dumped CISCO’s Foundational-Sec-8B. Will retain Qwen3 for for complex task. Will still use frontier models while with caution on how prompts are made so as to not make it easy for adversaries to infer my designs on the cybersec product.
In the meantime I need to prepare for a 21km night run under bad weather.
This site will hold short thoughts, daily observations, technical discoveries, project progress, and records that do not require a full article.