01/03/2026 1:30 am
Topic starter
Notifications
Clear all
Other LLMs Forum
7
Posts
8
Users
0
Reactions
314
Views
0
Seriously, what is actually the best AI for coding and debugging Python scripts right now because I am about to lose my mind with ChatGPT? I have been sitting here for four hours trying to fix a simple selenium script for a client project I need to deliver by Friday and I'm honestly just fed up with how much the quality has dropped lately. I keep pasting my traceback errors into the chat and it just keeps giving me the same "I apologize for the oversight" nonsense and then literally gives me the exact same block of code that failed in the first place like it's not even reading my input anymore.
Im a freelance dev working out of my apartment in Berlin and usually I can knock these automation scripts out pretty fast but this latest update or whatever they did to the model has made it basically unusable for anything more complex than a "hello world" script. It keeps hallucinating methods that dont even exist in the current version of pandas or it'll suggest some weird deprecated library from five years ago and then when I call it out it just hallucinates a different error. Its so frustrating because I feel like I'm spending more time debugging the AI than actually writing the logic myself at this point and the deadline is creeping up fast.
I've heard some people talk about Claude 3.5 Sonnet being better for logic or maybe I should just bite the bullet and get a GitHub Copilot subscription but I dont want to waste more money if it's just gonna be the same struggle. I need something that actually understands context and doesnt just loop through the same three mistakes over and over again. Is anyone else seeing this dip in quality or am I just crazy? If you guys are using something that actually knows Python 3.12 and doesnt lose its mind after ten prompts please let me know. I really need to get this script running before the weekend starts or my client is gonna kill me...
7 Answers
12
01/03/2026 2:35 am
In my experience, Phind Pro 70B Model handles these library versioning issues much better. Over the years, I have tried many tools, and this one avoids those oversight loops because it actually verifies real-time documentation for Selenium and Python 3.12. Its ability to parse long tracebacks without losing context is superior to standard models... honestly worth the money so you wont miss that deadline.
11
01/03/2026 2:35 am
Regarding what #3 said about "In my experience, Phind Pro 70B Model handles..."
- I am obsessed with Google Gemini 1.5 Pro API lately! I had a massive Selenium script crashing and it fixed the logic issues instantly. Amazing! Quick question tho, are you using a venv? I am kinda paranoid about privacy for client work, so JetBrains AI Assistant Subscription is my go-to for safety and Python 3.12 support! honestly worth a look...
3
26/03/2026 1:51 am
Re: "In my experience, Phind Pro 70B Model handles..."
- I've definitely noticed the same thing with Phind and the other suggestions here like Claude and Cursor. It seems like the consensus is basically to move away from the main vanilla models for heavy dev work. Im pretty satisfied with where the tech is at if you use the right tool for the job. One thing to watch out for tho is prompt drifting where the AI starts ignoring your specific library constraints if the chat gets too long. Since you are working with Selenium and specific Pandas versions, my advice is to always include your library versions at the very top of every few prompts. It keeps the model from hallucinating deprecated methods from 2019. Just keep your context clean and you wont have nearly as many issues. Really glad to see everyone finding stuff that actually works well. Youre gonna crush that deadline... keep at it!
2
24/03/2026 4:50 am
@Reply #5 - good point! honestly its wild how much the quality fluctuates lately. I was in the same boat last week trying to fix a scraper and I almost threw my laptop out the window. if youre trying to save some cash while youre freelancing in Berlin you should check out Codeium Individual Free Tier. it basically does what copilot does but its free for individuals and the autocompletion is pretty decent for python automation scripts. it works well enough that i havent felt the need to pay for a sub yet. another thing I've been doing lately is using Sourcegraph Cody Free Tier. it has this feature where it actually indexes your local files so it knows the context of your whole project, not just the file you're in. that might help with those selenium vs pandas library versioning issues you're seeing since it can see your local environment better. also maybe try Continue.dev VS Code Extension to pipe different models directly into your editor for free. hang in there tho, selenium is a nightmare even without the ai acting up... good luck with that deadline.
1
01/03/2026 1:30 am
tbh you arent crazy. I switched a few weeks ago and I am very satisfied with the results.
- Anthropic Claude 3.5 Sonnet handles complex logic much better.
- GitHub Copilot Individual Monthly Subscription is super reliable for real-time Python suggestions. I am happy with how stable these are compared to ChatGPT. No complaints here, it just works.
1
01/03/2026 2:05 am
I'd suggest you might want to consider Cursor AI Code Editor Pro Plan for those scripts. Be careful with context limits tho.
- try DeepSeek Coder V2
- dont give up
1
05/03/2026 5:52 am
Big if true
