GPT-5.4 is also more reliable, producing 18% fewer errors and 33% fewer false claims than GPT-5.2, according to OpenAI.
Mainstream chatbots presented varying levels of resistance to deliberate requests for fabrication, study finds.