I need to know. Haven’t found any benchmarks about code-supernova yet. I was using gpt-5-high, because it was reliable and today I changed to code-supernova MAX mode, it seems to be working well and solving issues like gpt-5, but I need your thoughts. Can’t afford it screw my whole codebase.
supernova will break your codebase. so will gpt5. but with gpt5 the codebase might still be functional after a couple of prompts, cant say the same about supernova
gpt-5 >> grok code fast > supernova
Really? You guys think this? I though code-supernova is pretty good, though. And GPT 5 did not break my codebase. They seems to be working pretty nice for me.
At the moment, I am using code-supernova for more junior tasks and gpt-5-high for most senior tasks.
keep in mind that some models might do better working on specific languages/tasks,
i use svelte / typescript in pretty big project, gpt for me has not been so great while supernova so far have been able to produce pretty good output, with bigger tasks i usually need to correct it in few places but overall it works well
Code-supernova is probably some kind of Chinese LLM. Or a smaller companies LLM. It’s tremendously overfit, and has not seen enough training data. It’s not xAI, not anthropic, 100%–this is someone with limited compute.
(it is xAI—trying to be offensive, so they fix it)
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.