OPENAI HAS released a ‘research preview’ of its flagship language model, which it describes as its ‘largest and best for chat yet’. Anyone willing to pay $200 per month for a ‘Chat Pro’ subscription can try out version 4.5 today.
This ‘non-reasoning’ model, known internally as ‘Orion’, is designed to better interpret user intent and have a higher virtual ‘EQ’ when it comes to interactions. Nevertheless, the company has emphasised that GPT-4.5 is not one of its cutting-edge ‘frontier’ models. Even so, it claims that the LLM hallucinates less than GPT-4o, and is better at managing conversations.
OpenAI also tested GPT-4.5 using its ‘SimpleQA’ knowledge quiz. The model scored 62.5 percent. This compares well with GPT-4o (38.6 percent) and o3-mini (15 percent).