From day one, OpenAI’s ChatGPT has been a game-changer in generative AI, continuously raising the bar with new and improved models. As Google and Meta close the gap, the race to dominate generative AI rages on—at least until we reach artificial general intelligence (AGI)!

OpenAI, a leading force in AI innovation, is working on a new large language model called Orion—totally unrelated to Meta’s Oryon AR smart glasses. This highly anticipated model is also known as GPT-5.

Buzz has been swirling that OpenAI’s next AI model could drop as soon as December 2024, but a spokesperson recently shot down those claims to The Verge. This speculation might be linked to the exodus of top leaders, including former CTO Mira Murati.

The Verge reports that Orion might skip a public release like GPT-4o. Instead, this model is set to be exclusive for OpenAI partners, such as Microsoft, powering services like Copilot—at least for now.

Since the debut of GPT-4 in March 2024, OpenAI has rolled out several revamped models like GPT-4o and OpenAI o1. Now, it looks like we’ll have to hold tight until early 2025 for the official scoop on the mysterious Orion, possibly branded as GPT-5.

OpenAI’s GPT-5 is shaping up to be the most powerful language model yet, boasting exceptional reasoning abilities. Reports suggest that pre-training could wrap up by December 2024, followed by a few months of fine-tuning before its grand reveal.

While it won’t quite reach AGI status, it’s a significant step towards the pinnacle of artificial intelligence. Expect GPT-5 to outshine GPT-4o across the board, featuring a larger context window, an expanded knowledge base, and enhanced reasoning skills.

Orion could bring us one step closer to achieving AGI

OpenAI’s former CTO, Mira Murati, recently shared that “systems like GPT-4 are more akin to smart high schoolers,” while hinting that in the next few years, we might see models with “PhD-level intelligence” for specific tasks. She noted that advancements are happening quickly, possibly suggesting just how capable GPT-5 will be. OpenAI co-founder Sam Altman has also mentioned that the next model will represent a “significant leap forward,” although he emphasized that there’s still much work to be done.

GPT-5 is expected to be a multimodal AI model, capable of processing and generating text, audio, and images, or a combination of these formats. Similar to models like Llama 3.2 and Gemini, it could be available in different sizes tailored for various applications.

The performance of an AI model is often gauged by the number of parameters it has been trained on, and GPT-5 is rumored to be one of the largest language models ever, boasting over 1.5 trillion parameters.

Human-level intelligence is nearly within reach

At the recent Build with AI summit in India, Meta’s chief AI scientist Yann LeCun noted that even the most advanced AI models today are not as intelligent as a four-year-old. He mentioned that many AI companies are striving to develop models with human-level intelligence, and GPT-5 could be among them.

Conclusion

Safety remains a top concern for AI models, with many implementing measures like watermarking AI-generated content and avoiding the use of personal data for training. Recent reports highlight alarming incidents, including a mother claiming that AI contributed to her son’s death. As a result, OpenAI is expected to tackle these critical issues head-on with GPT-5.