Early user: gpt-6 astra follows instructions far better than 5.6-sol, no more tangents

haider1 · x · 2026-09-20

An early user reports that gpt-6 astra (md) follows instructions significantly better than 5.6-sol-md and no longer gets lost in endless second-guessing or random tangents.

The author is now eagerly awaiting gpt-6 sol and luna variants, noting that if they land at 5.6 sol/luna pricing, the result would be both faster and cheaper — a major win for heavy users.

Original post →

More from Models

Models channel →