Ollama makes 64K Mac coding viable

The practical answer is yes, but only if you stop chasing giant models and treat context length as a memory budget, not a badge of honor.

• –Ollama’s own docs say coding tools should generally be set to at least 64k context, but larger windows increase memory use fast.
• –On 32GB unified memory, quantized 7B-14B models are the sweet spot; 32B-class models get uncomfortable quickly.
• –Qwen2.5-Coder is a strong fit here because it officially supports up to 128K context and is tuned for coding tasks.
• –MLX may squeeze better Apple-silicon performance, but Ollama and llama.cpp remain the most practical defaults for broad local model support.