GPT-5, OpenAI’s next big language model (LLM), is in the works and is expected to be launched within a matter of months, according to those close to the matter who told Business Insider.
Recommended Videos
Two anonymous sources informed the publication that GPT-5 might be set to debut during the summer, and that OpenAI is currently sending demos of the technology, along with updates to the accompanying ChatGPT chatbot, to its enterprise customers for testing.
One CEO who had the opportunity to experience a GPT-5 demo that offered use cases specific to his company was highly impressed by what OpenAI has showcased so far. “It’s truly excellent, like significantly better,” he told Business Insider.
The brand’s internal presentations also focus on the unreleased features of GPT-5. One such function is an AI agent that can carry out tasks independently without human assistance.
While enterprise partners are testing GPT-5 internally, sources claim that OpenAI is still training the upcoming LLM. This timeline will ultimately determine the model’s release date, as it must still undergo safety testing, including red teaming. This is a cybersecurity process where OpenAI employees and other third parties attempt to penetrate the technology under the guise of a bad actor to uncover vulnerabilities before it is made available to the public.
This process could take several months, so OpenAI has not set a definite release date for GPT-5, and current predictions may change.
The last official update on GPT-5 provided by OpenAI was in April 2023, stating that there were “no plans” for immediate training. Just a month after the release of GPT-4, CEO and co-founder Sam Altman quashed rumors about GPT-5, saying at the time that the rumors were “silly.” There were also early rumors of an incremental GPT-4.5 that persisted through late 2023.
But since then, there have been reports that training was already completed in 2023 and it will be launched sometime in 2024.
I’ve been told that GPT-5 is scheduled to complete training this December and that OpenAI expects it to achieve AGI.
Which means we’ll all be engaged in intense debates about whether it actually achieves AGI.
And which means it will.
— Siqi Chen (@blader) March 27, 2023
GPT-5 is likely to be targeted at OpenAI’s enterprise customers, who generate the majority of the company’s revenue. Potentially, with the launch of the new model, the company could establish a tier system similar to Google Gemini LLM tiers, with different model versions serving different purposes and customers. Currently, the GPT-4 and GPT-4 Turbo models are well-known for powering the ChatGPT Plus paid consumer tier product, while the GPT-3.5 model runs the original and still freely available ChatGPT chatbot.
Throughout the past year, users have reported experiencing “laziness” and a “dumbing down” of GPT-4 as they encountered hallucinations, sassy backtalk, or query failures from the language model. There have been various potential explanations for these occurrences, including the possibility that GPT-4 is becoming smarter and more efficient as it is better trained, and that OpenAI is working with limited GPU resources. Some have also speculated that OpenAI may have been training new, unreleased LLMs alongside the current ones, which overwhelmed its systems.
In December 2023, observers referred to the hypothetical LLMs as GPT-4.5 and GPT-4 Turbo, having discovered pricing tiers and information about advanced multimodal capabilities. When asked about the validity of the leak on X (formerly Twitter), Altman replied: “Nah.”