How is Codex as a harness for local models?

<!-- SC_OFF --><div class="md"><p>I read the latest Anthropic blog post with great interest. How can I setup a similar harness?</p> <p><a

<a href="https://news.ycombinator.com/item?id=50151390">Comments</a>

<!-- SC_OFF --><div class="md"><p>Been using ClaudeCode CLI with Opus 4.6 and many MCP's and honestly its addicting. </p> <p>Just tell it what to build and it does everything —

<!-- SC_OFF --><div class="md"><p>Which one do you prefer for working with GPT: Codex or OpenCode?</p> <p>I'm mainly interested in code quality, context management, token

<!-- SC_OFF --><div class="md"><p>I hear Qwen code unlocks the model better. I also think it has more power user features than open code?</p> <p>It’s nice open code can work with

<!-- SC_OFF --><div class="md"><p>A noticeable trend is happening: more users are moving away from heavily aligned cloud models (like ChatGPT, Claude, Gemini) toward local or

<!-- SC_OFF --><div class="md"><p>I have experience using llama.cpp on Windows/Linux with 8GB NVIDIA card (384 GB/s bandwidth) and offloading to CPU to run MoE models. I typically

<!-- SC_OFF --><div class="md"><p>Hey all, I’ve been using Claude Code with Opus and Sonnet but as you all know the rate limits as well model capabilities have degraded