Comment by TeMPOraL
9 hours ago
Thanks. I've started digging into Aider's features I've been ignoring so far, we'll see how this goes.
And yes, I've been using gemini for the past month or two - ever since gemini-2.5-pro came out and topped the Aider benchmark. It's good, but it sure does comment excessively, including in places like quoted scripts, where those comments are a syntax error...
I've tried the current top combo from Aider's benchmarks last night - that is, o3 (high) architect + GPT-4.1 editor. It's refreshingly concise, generates much smaller diffs, but man does it burn through money - a factor 16x relative to gemini-2.5-pro-preview-05-06. Not sure if it's worth it.
I will occasionally switch to O3 briefly when gemini gets stuck, it's not a good daily driver but it can sometimes crack tough problems and let you get back to smooth progress. If you want to save some money you can tell gemini to make a concise summary of the problem it's stuck on to provide to another LLM so the other model can provide instructions on how to solve it, then you can paste that into ChatGPT and use O3 for free (if you have a sub). You just then paste O3's reply back into Aider and you're usually good to go again.