OpenAI is Training GPT-5: Three Major Upgrades to Anticipate for the Successor to GPT-4

OpenAI's recent launch of the GPT-4o model represents a significant leap in the field of large language models (LLMs), yet the company is already progressing on its next flagship model, GPT-5. Anticipation for GPT-5 grew even prior to the GPT-4o release. To address speculation, CEO Sam Altman confirmed on X that it's “not GPT-5, nor a search engine.” Shortly after, OpenAI announced the formation of a new safety and security advisory committee to aid its board. In a blog post, OpenAI confirmed that training for their next flagship model—likely GPT-5—has commenced, stating, “OpenAI recently started training its next cutting-edge model, which we expect will elevate our capabilities on the pathway to artificial general intelligence (AGI).”

While GPT-5's release may take several months due to the lengthy training process for LLMs, we can anticipate several advancements in this next-generation model, ranked from the least to the most exciting:

1. Improved Accuracy: Following historical trends, GPT-5 is expected to demonstrate enhanced accuracy in its responses, thanks to training on a larger dataset. Generative AI models like ChatGPT benefit from extensive training data, leading to more coherent content generation. Previous model upgrades have seen significant increases in data usage, with GPT-3.5 trained on 175 billion parameters and GPT-4 utilizing 1 trillion. GPT-5 will likely exceed these numbers.

2. Enhanced Multimodal Capabilities: By analyzing the differences between flagship models like GPT-3.5, GPT-4, and GPT-4o, we can predict GPT-5’s function. Each iteration has introduced smarter features, improving pricing, speed, contextual length, and modality. GPT-3.5 operated solely on text, while GPT-4Turbo added text and image processing. The GPT-4o model further integrated text, audio, images, and video inputs with various output formats. Following this progression, GPT-5 is expected to include video output capabilities, particularly with recent advances from OpenAI’s Sora model for text-to-video integration.

3. Achieving Autonomous General Intelligence (AGI): Current chatbots are highly capable, assisting in tasks like generating code, creating Excel formulas, and drafting articles. However, the growing expectation is for AI to understand our intent and perform tasks with minimal input—this embodies the vision of AGI. If GPT-5 achieves AGI, users could simply say, “Help me order a burger from McDonald’s,” and the AI would manage everything necessary, from navigating the website to processing payment, leaving users to enjoy their meal. Advancements in AGI would fundamentally change how we interact with AI assistants, empowering them to offer comprehensive support rather than basic information.

In conclusion, the transition to advanced AI models like GPT-5 is set to redefine our expectations and experiences with digital assistants, potentially transforming everyday interactions into seamless, intelligent experiences.