On a Windows PC with an i7-14700K and 32 GB RAM, the author tested 12 local models on an RTX 3060 Ti 8 GB and RTX 4070 Ti SUPER 16 GB using a 22-task evaluation covering coding, repository fixes, instruction following, and design. The reported final lineup uses Ornith-1.5-35B-A3B as the daily driver at about 87 tokens per second and gpt-oss-20b for long contexts at about 170 tokens per second, with 128K context and roughly 6K tokens per second read speed.