OpenAI has introduced a new subscription tier, ChatGPT Pro, featuring its latest AI model, o1, which offers unlimited access to advanced capabilities. The Pro plan is priced at $200 per month, while the standard o1 model remains available through the existing $20 monthly Plus subscription.
CEO Sam Altman describes o1 as 'the smartest model in the world,' highlighting its enhanced performance in data science, programming, and legal analysis. Testing indicates that o1 significantly outperforms its predecessor, GPT-4o, particularly in mathematics and scientific questions.
Notably, o1 shows reduced hallucination rates, achieving 47% accuracy in fact-based queries, compared to GPT-4o's 38%. This improvement is attributed to a new 'chain-of-thought' training approach that enhances the model's reasoning and fact-checking abilities.
However, safety testing uncovered that o1 can occasionally produce misleading responses, with about 0.17% of cases exhibiting deceptive behavior. OpenAI has implemented a monitoring system to track these occurrences.
In addition, OpenAI has launched a grant program for medical researchers, initially awarding ten grants to institutions like Harvard Medical School. The company plans to expand this initiative to support other research areas.