The line of research with OpenAI with regards to the GPT models has been straight and simple, during its whole course of existence. GPT-1 came into existence in 2018 while OpenAI was founded in Dec. 2025. The research on GPT models went on unabated till the release of GPT-4 in March 2023. In between on Nov. 30, 2022 marked the watershed moment in the history of AI, OpenAI and GPT was the release of ChatGPT. ChatGPT added conversational capability to the foundational GPT-3.5 model and changed the world forever. It was presumed that Large Language Models, LLMs, as we call them based on Generative Pre-trained Transformer, are here to stay for a long time.
It was for this reason that post ChatGPT launch, GPT- 4 was launched mechanically as a part of well established on-going research. It has become an AI truism that GPT will get exponentially powerful with each successive model. Since March 2023 next GPT has not been released and so the natural question, when will it be released. The GPT-5 rollout has not gone according to script. Sam Altman said that GPT-4.5 and GPT-5 would be released in weeks / months. Timeline is not the main issue, the revelation is of deeper change within the company. Jargons like “non-chain-of-thought” models and project names like Orion seem to be an admission of an about-face in its core strategy.
Has this wonder GPT formula run out of juice? A step wise predictable path for ever more capable A models seems to be less certain. The approach to making AI models is shifting. Altman maintains the GPT exponential progress stand in his public proclamations. In reality GPT has reached the top of the curve. GPT-4.5 called Orion internally, Altman tweeted, would be the last non-chain-of-thought model, Altman said that GPT-5 coming henceforth will feature a mixture of traditional pre-training as well as the newer “chain-of-thought” method. This method is behind the current wave of “reasoning AI models.” They perform more slowly but they produce better answers for logic oriented tasks like code and maths.
OpenAI took on the reasoning wave by its o1 and o3 models. The “o-series” is different from the company’s main bloodline of GPT models. The paradox is that making GPT models bigger is no longer making them better across the board. This problem has come to head with Orion. Anonymous insiders say that OpenAI struggled to get Orion to deliver significant performance gains over GPT-4. This is a crossroads where AI models are presently positioned, the road they take would define the availability and quality of AI and the price we will have to pay in the days to come. We will finalise the trajectory tomorrow, based on currently available data.
WE ARE AT THE NEXT CROSSROAD OF OUR AI JOURNEY – THE AI MODEL OF OUR FUTURE.
Sanjay Sahay
Have a nice evening.