ChatGPT 5: A New Model From OpenAI

Forbes

2 days ago

It was a big day over at OpenAI. First, Sam Altman came onstage to kick off a livestream at 1:00 EST (10:00 PST) in which the company unveiled the new ChatGPT5 model that's going to boost the capability of the LLM.
Calling the new model 'a significant step along our path to AGI,' Altman described GPT5 as like having a team of Ph.Ds in any given subject to help you through a project or question. The model, he said, is also very versatile.
'It can do stuff for you,' he said.
Next, a team of employees showed off how GPT5 scores on benchmarks including SWE, Aider Polyglot, MMMU and AIME 2025. In terms of reliability, presenters showed how transparency will help, and how more painstaking work can address any hallucinations that might occur.
GPT5, they explain, will also be better at mitigating deception, which can happen, for example, when a model is under pressure and wants to hide something.
Multimodal Design
In addition, teams showed off the new model's voice capabilities, which are going to change how we interact with the technology. In going over some translation tasks, they showed how GPT5 can imitate various accents, speak at different speeds, and generally seem highly authentic in a range of languages.
New Features and Engineering
A tech also went over many of the new features in ChatGPT 5. There will be three different options in this model: a regular GPT5, a GPT5 mini, and a nano version able to run well on end devices, to help customers handle the 'cost-latency curve' of the tech.
In addition, GPT5 will allow for a minimal reasoning effort, or allow users to trade off between time and power of response. A custom tools feature will help developers to go beyond the routine use of JSON as a standard. A verbosity parameter will help users to control how wordy the model's answers are, and tool call preambles build robustness into responses.
Along the way, OpenAI people provided a range of demos including 'mini-games,' a mouse chasing cheese with French words as targets, a visual explanation of the Bernoulli effect, and what appeared to be a castle game with intrusive meteors.
Notes on Development
OpenAI president Greg Brockman took the stage later to show how power users are building creatively with GPT5.
'Software engineering is already fundamentally changing, and GPT5 will turbo-charge that revolution,' Brockman said. 'GPT5 sets a whole new standard. It is the best model at agentic coding tasks. You can ask it to go and accomplish something very complicated, and it'll go off and it'll work on it. It'll call many tools. It'll work for many minutes at a time, sometimes even longer, to accomplish your goal, your instruction, your tasks, whatever it is that you're trying to build. It's incredible at the front end. It makes very beautiful visualizations and interactive games.'
One team member after the other unveiled relevant use cases: evaluating an API system, troubleshooting with an SDK, fixing a bug.
'Benchmarks, they're exciting numbers, but we're starting to saturate them,' Brockman said. 'Like when you're moving between 98% and 99% of a benchmark, it means you need something else to really capture how great the model is. One thing we've done very differently with this model is really focus on not just these numbers, but really on real-world applications, being really useful to you in your daily workflow. So that hearing about it is as exciting as seeing it.'
Team members estimated that 5 million businesses worldwide are already dependent on GPT. The consensus seemed to be that this number is about to grow precipitously.
That's big news for today, partly because this new model is rolling out to all OpenAI user tiers. Even free users will get some form of use of GPT5. So if you have any kind of task that prior versions tried to handle, the new one will do it faster, more thoroughly, and with more agentic capabilities.
Stay tuned.