OpenAI Unveils GPT-4.5 With Fewer Hallucinations and Higher Emotional Intelligence
OpenAI releases a research preview of GPT-4.5, a new general-purpose language model that produces fewer inaccuracies and feels more natural to interact with. The rollout begins for Pro subscribers and developers, with broader access coming in the following weeks.
OpenAI announces the research preview of GPT-4.5, a general-purpose large language model that currently remains available exclusively to software developers and ChatGPT Pro subscribers. The company reports that this new model produces inaccurate information less frequently than its predecessors, demonstrating a lower hallucination rate than both the GPT-4o and o1 models in internal testing.
CEO Sam Altman describes GPT-4.5 as a "giant, expensive model" that possesses a unique magic, noting that it does not claim the top spot on standard benchmarks despite its improvements. However, OpenAI highlights that the system features a broader knowledge base, better intent tracking, and higher emotional intelligence, making it highly effective for writing, programming, and solving practical problems.
The company plans a staggered rollout to its massive user base of over 400 million weekly active users, with ChatGPT Plus and Team subscribers gaining access next week. Altman explains that simultaneous access for all paid tiers is impossible due to GPU shortages caused by rapid growth, but OpenAI is adding tens of thousands of GPUs to accommodate the demand before expanding access to Edu and Enterprise users the following week.