ChatGPT-5: Everything We Know So Far About OpenAI's Next Game-Changing AI – Medium
Sign in
Sign in
Member-only story
Susan Boyche
Follow
—
1
Share
OpenAI’s ChatGPT chatbot took the world by storm when it was released in November 2022. But the AI research company is already hard at work on the next version, ChatGPT-5, which promises to be even more advanced.
While details are scarce, there are some clues about what we can expect from ChatGPT-5 based on OpenAI’s typical development process and comments from insiders. In this story, I explore everything we know so far about the potential capabilities, architecture, training process and release timeline for ChatGPT-5.
Before launching into full-scale model training, OpenAI typically conducts smaller preparatory training runs. For example, before developing GPT-3, they trained a 1.5 billion parameter “sibling” model called GPT-2. Similarly, before GPT-4, they trained a 300 million parameter “sibling” called InstructGPT.
These smaller models allow OpenAI to gather insights and test architectures on a reduced scale before committing the massive computational resources required for full training. It’s likely they are already conducting similar preparatory training for ChatGPT-5.
Sam Altman, CEO of OpenAI, stated that ChatGPT-5 is utilizing “maximal compute” during its training, indicating they are throwing enormous resources at the project.
The GPT-3 model was trained on approximately 3,640 petaflop days of compute power. In comparison, GPT-4 took about 6,000 petaflop days. ChatGPT-5 will likely require over 10,000 petaflop days, making it OpenAI’s most ambitious training run yet in terms of compute requirements.
Accessing cutting-edge supercomputers like the Frontier system at Oak Ridge National Lab, which can hit over 1.5 exaflops, will be key to achieving OpenAI’s…
—
—
1
Writer, Marketer, Creative.
Help
Status
About
Careers
Blog
Privacy
Terms
Text to speech
Teams
