Before the public launch of GPT-4, we had law firms using our AI without knowing it was powered by GPT-4, and they were amazed by its capabilities, completing tasks that previously took a whole day in just minutes.

When we launched GPT-4.5, we were proud of the model, but users liked it much more than we anticipated, expressing that it was significantly better than GPT-4 in various ways.

GPT-4.5 is potentially more than 10 times the size of GPT-4, representing a step forward in both pre-training and post-training processes.

The main advantages of GPT-4 include faster response times, sharper reasoning that reduces misunderstandings, and more natural language that feels relatable rather than robotic.

In testing GPT-4.5, we found it has nuanced abilities that were unexpected, indicating it is more intelligent in ways that are hard to characterize ahead of time.

GPT-4.5 achieves around 61.9% accuracy on benchmarks like simple QA, which evaluates its effectiveness in answering questions, and answers fact-based questions with a 38.4% accuracy rate, significantly reducing hallucination rates to roughly 37%.

The transition from GPT-3 to GPT-4 marked a significant moment where users began to see real value in the technology.

OpenAI revealed GPT-4.5 as a model capable of reasoning and systematically thinking through complex problems, clearly surpassing GPT-4 in mathematics, coding, and logical tasks.

Reactions to GPT-4.5 have been muted because its improvements over GPT-4 are seen as relatively incremental, despite this launch being an important milestone for future models.