GPT 5.6 Luna Max vs. Sol Medium
I spent the last 2 days running several sessions with both, and I think the official benchmarks are misleading in suggesting they're equivalent.
Luna is obviously much cheaper. Despite using many more tokens, it cost an average of $1.20/session, while Sol averaged $29. That's a big difference, and it also affects subscription rate limits! So if you're prioritizing cost over other factors, Luna is probably the way to go.
However, I had to redo most of Luna's work. In fact, Luna wasn't able to produce any shippable code for my use cases, so the huge cost reduction was just an illusion.
The bigger issue with Luna was that it often ignored initial instructions, especially as the conversation grew longer. Toward the end of tasks, it got lost on very basic stuff, which made me switch to Sol for most of them.
I had Opus 5 review all their work, and Luna generally introduced more P1 regressions.
After reviewing the sessions, both Opus 5 and Sol xhigh concluded that Sol Medium produced better results despite the price, which matches my own assessment.