Key differences from GPT-4 and o3

Developers note that GPT-5 is significantly different from its predecessors. Let's consider the differences by comparing it with GPT-4 and o3:

Self-determination mode. The model decides when to activate deep reasoning mode (GPT-5 Thinking) and when to provide a brief response.

Increased accuracy. The probability of errors in normal mode is reduced by 45%, and in Thinking mode — up to 80% compared to GPT-4o and o3.

Less flattery. The tendency for excessive agreement has decreased from 14.5% to less than 6%.

Honesty. GPT-5 is less likely to provide false answers — 2.1% compared to 4.8% for o3.

$ERA $NOT $LAYER