gpt-6IMAGE
Screenshot claims GPT-6 Sol makes ~50% fewer factual errors than 5.6 Sol
A screenshot claims GPT-6 Sol cuts factual errors by about half, nearing Astra-level reliability at lower cost, while higher-effort GPT-6 Luna matches 5.6 Sol factuality at roughly 1/100th the cost.
Ananth
@Ananth7eArchived 2026.09.23Originally posted 2026.09.23

Ananth @Ananth7eOriginally posted 2026.09.23
GPT-6 Sol makes ~50% fewer factual mistakes than gpt-5.6 sol. openai says it's approaching astra-level factual reliability at a fraction of the cost. even crazier, higher effort GPT-6 Luna matches 5.6 Sol factuality at around 1/100th the cost.
x.com/ananth7e/status/2102467178536788402Archived 2026.09.23 10:17