You can already get Opus 4.6 level of performance on subtasks with some local models. So you need to pick a proper code writer, plan writer, code tester etc. mo

You can already get Opus 4.6 level of performance on subtasks with some local models. So you need to pick a proper code writer, plan writer, code tester etc. model that matches your target expectations and use a coding tool that allows calling different LLMs for different subtasks. For example, people use StepFun 3.x or DeepSeek4-Flash for planning, Qwen3.6-27B for coding.