GPT-6.2 Sol coding score drops from 52 to 49, but costs 19% less per task

lxfater · x · 2026-11-18

Artificial Analysis benchmarked two generations of GPT Sol on Codex, both at max: the score fell from 52 to 49, roughly a 5.8% regression.

However, per-task API cost dropped from $1.51 to $1.22, about 19% cheaper.

A 19% price cut in exchange for a 5.8% capability drop — is that a trade worth taking?

Original post →

More from Models

Models channel →