Ask HN: Models Comparable to Opus 4.6?
I use Opus 4.6 a lot across many different python coding projects and it has a pretty good first shot rate with good success at fixing issues and bugs that pop up along the way.
Sonnet on the other hand… isn't great. It seems to hallucinate quite a bit, use strange functions and code blocks and then bandaid issues rather than fixing them.
With GitHub Copilot getting rid of it for individual plans and Claude code trying to silently hide it away I fear the end is near and the “discounted” 7.5x usage for Copilot and 2-2.5x usage for Claude Code has me worried about the cost implications of blindly switching over to Opus 4.7. So I am wondering if anyone has had success with some other models that are near the performance of Opus 4.6 and at least better than Sonnet. OpenCode Zen is a great way to try different models for code generation. Pay $20 one time and switch between models effortlessly. For my use case this revealed Kimi 2.5 to be excellent and insanely affordable. I believe it’s also the foundation for Cursor’s own model, so they’re doing something right. You can see all the models they provide here: https://opencode.ai/docs/zen/#how-it-works (Not affiliated) Opus 4.7 has gotten better over the last few days. When it came out was honestly worse than Opus 4.6 IMHO. I've heard great things about Gemma 4 as well - planning to try it out soon Codex 5.4, Opus 4.7, Codex 5.5, Cursor's Composer 2.