Reasoning model from genimi is immensely good, i used tons of it from the playground in ai studio, it would be a solid add to the windsurf as it will have even more access to the codebase to reason through.

If it's the 6-20-2024 version of Claude then leave it alone, please :)

Otherwise, yes, please delete.

However, Gemini 1.5flash is seriously an out-performer and 2.0 is there as well, so should be considered based on that alone. This is coming from someone that despises Google...

I don't even care if it's better than Claude. I just want something close to Claude that doesn't constantly delete unrelated part of my code ;) Hopefully Gemini doesn't have this tendency....

I think the codeium base model is pretty good, i would love to see it on the benchmark. The problem is that a multishot query works well, we just need to have good guardrails. that's probably what they did to train the codeium base model.

all these benchmark are 1 -shot what we need to look at is being able to do multi-shot using agents, and I think codeium are doing that well. as our prompts of 'fix this' doesn't have enough context, so codeium or any LLM will have to work harder to find the problem.

WE'RE getting lazy.