Follow up on "Is it time to jump ship from OpenCode? "
retained text
I have to say there are major downsides from going to V1 to V2. I have now been working on one my projects ca 16hrs straight, that I started with V1, every change had been smooth sailing (using strata with official qwen flash settings). When moving to V2, major or minor changes has become too hard for the model to implement in one go. OpenCode V1 was the better platform for medium sized models it seems. I do all my planning with Qwen 3.8 MAX so that hasn't changed, which means performance is even worse considering the planning was kept separate. I know this is anecdotal evidence, but engine has been kept the same and it's settings. Face always falls flat on V2, leading to lots of follow up work on even easy website design updates. Honestly this is telling that OpenCode only cares now for bigger paid models performance and not medium to small models anymore, even though they claimed to make it "lighter and faster". Has anyone found a better harness than OpenCode V1? DSH disappointed too much even after a major update. Opinions appreciated from people who work with projects where even SOTA models fall flat on planning and reasoning, I do not make basic websites etc.
document fetched 2026-11-03 00:42 · canonical https://reddit.com/r/LocalLLaMA/comments/1xmd9aw/follow_up_on_is_it_time_to_jump_ship_from_opencode