I ran into that exact Pydantic v2 migration issue last week. Copilot kept hallucinating the old .dict() method instead of model_dump() which killed my flow. I shifted to these and they handle Python 3.12 way better:

• Cursor Code Editor Pro Claude 3.5 Sonnet
• Its 200k context window is vital for indexing your whole codebase.
• Supermaven Pro AI Coding Assistant
• High speed and a 1-million token context window.

Saw this thread late but figured Id chime in. Ive been writing Python for over a decade and honestly, Im always a bit wary of these tools making creative choices in my codebase. I had a nightmare scenario last year where an auto-suggestion introduced a subtle bug in some async code that took me way too long to hunt down. Now I stick to tools that prioritize my actual repo context over just being flashy. For my current 3.12 projects, Ive had way better luck with these:

• Sourcegraph Cody Pro Context-Aware AI
• Tabnine Pro Private AI Coding Assistant Cody is great because it actually scans my local files before it suggests anything. Tabnine is my go-to for sensitive client stuff since I can keep things local. It might not feel as fast as Copilot, but its reliable and wont hallucinate old Pydantic syntax if you point it at your current env.

Caught this thread late but i totally agree with the first point about Copilot failing on migrations. Over the years, ive found that dedicated IDE brands usually handle the deep logic better than the general cloud tools.

• Just stick with JetBrains for Python.
• Or check out Tabnine. They focus way more on syntax accuracy than the massive models do lately. Pick either and you'll be fine.

Wait really?? Thats actually super helpful. I always thought it was the other way around.

@Reply #4 - good point! Quick reply while I have a sec. Im honestly a bit scared of how fast these tools change things. I used a newer setup recently and it almost deleted my config file because i didnt read the diff carefully. Ive learned a couple things:

• Always double check the syntax manually
• Never trust the AI with critical migrations Id suggest just doing the boring manual work for that Tuesday deadline to stay safe.

bump