OpenAI unveils GPT-4.5: larger and with fewer hallucinations
OpenAI announced on Thursday (27) the launch of GPT-4.5, described as its largest and best conversational model to date. The new artificial intelligence system was initially released for ChatGPT Pro subscribers, with plans for gradual expansion to other subscription tiers in the coming days.
Advances in accuracy and understanding
According to the company, internal tests have shown that GPT-4.5 presents a significantly lower rate of hallucinations – the phenomenon where AI generates false or invented information – compared to previous versions. Additionally, the model shows better understanding of nuances, which can result in more precise and contextually appropriate responses.
Availability and future plans
Initially, only ChatGPT Pro users – a plan that costs $200 per month and offers priority access to new features – can test GPT-4.5. OpenAI did not specify dates for release on the Plus plan ($20/month) and free tier, but stated that expansion will occur gradually.
Impact on AI market
The launch comes at a time of intense competition in the sector, with companies like Google, Microsoft, and startups like Anthropic also investing in large-scale language models. GPT-4.5 arrives less than a year after GPT-4, indicating an acceleration in OpenAI's innovation cycle.
Next steps
OpenAI did not disclose full technical details about GPT-4.5, such as parameter count or exact architecture. The company promised more information in the coming weeks, including comparative benchmarks with other models. For now, the priority is to monitor performance among early users and collect feedback for adjustments.