OpenAI has released GPT-4.5, a model focused on usability rather than advanced capabilities like coding or reasoning.
GPT-4.5 showed a 40% reduction in hallucination rate compared to its predecessor GPT-4o.
On the SimpleQA benchmark, GPT-4.5 scored higher accuracy rates than competing models like o3-mini, o1, and GPT-4o.
Early testers praised GPT-4.5 for its verbal and emotional intelligence, describing it as the highest verbal intelligence model they've used.
CEO Sam Altman noted that GPT-4.5 feels like talking to a thoughtful person, offering good advice and a different kind of intelligence.
OpenAI plans to release the GPT-5 model soon, incorporating general purpose and reasoning capabilities in a single model.
However, GPT-4.5 comes at an exponential cost, being the most expensive model offered by OpenAI, with pricing significantly higher than GPT-4o and o3-mini.
The GPT-4.5 Preview costs $75 and $150 per 1 million input and output tokens, respectively, compared to GPT-4o's $2.5 and $10 pricing.
Some experts suggest that open-sourcing GPT 4.5 could have a more significant impact than distributing it through an API, citing potential benefits in user engagement and goodwill.
OpenAI faces competition from upcoming models like DeepSeek-R2 and Meta's Llama 4, expected to launch soon.