I’ve been testing GPT-6 Sol in ChatGPT Work across real development projects, and I keep seeing more execution mistakes than I do with Astra.

Missed instructions. Incomplete gates. Context that gets read but does not always survive the full execution.

I went looking to see if it was just me. It isn’t. Other users are reporting similar problems with context handling and long-running Work/Codex sessions.

That does not mean Sol is universally worse. It does mean I’m changing how I use it.

For technical project work, I’m back to Astra first, usually at the lowest reasoning level that can reliably finish the job.

Sol still has a place.

It just is not my default for technical Work anymore.