For the first time, I'm using the #opencode #neovim plugin as a sparring partner for writing a piece of text. I am keeping it in read-only mode, and discussing formulations, weaknesses, consistency errors, paragraph-by-paragraph, as I like to work this way ever since. Using qwen3.5:122b with ollama, locally hosted does not seem any worse than #gemini felt a year ago.
Everything is running open source, except for the model, which is only open-weight, and no data leakage to external parties. It is a bit slower than the cloud models on my very machine (a frame.work desktop with AMD Strix Halo), but sufficiently fast for this application, as I am a slow thinker, so usually I ponder some aspect of the text longer than the model does anyway.
I did not try to use it for critiquing code yet, only text, but I suspect for code it might be too slow.
#qwen #llm #foss