OpenAI CEO Sam Altman has confirmed that the makers of ChatGPT are delaying the launch of its latest base model, GPT-5, after facing ‘biblical’ demand for its recently launched native image generation feature. Notably, ChatGPT’s underlying model GPT-4o received a massive update last week that allowed it to directly generate images instead of relying on an external model such as DALL-E 3. The new features allowed ChatGPT to better understand the user’s context while generating more nuanced and accurate images, resulting in a global Studio Ghibli-style image trend by social media users.
The viral trend led to a shortage of GPUs within the company, leading to image generation limits being imposed even on ChatGPT Plus and Pro members. However, OpenAI seems to have learnt the lessons of the recent supply shortages, saying that it wants to “make sure we have enough capacity to support what we expect to be unprecedented demand.”
Altman explained that the GPT-5 release has been delayed for “a bunch of reasons,” but the most exciting one being that the Microsoft-backed company now believes it can make the language model much better than it originally imagined.
What do we know about GPT-5?
Not a lot is known about GPT-5 other than the details that Altman shared via a social media post earlier this year where he outlined that GPT-4.5 – which was released last month – will be the last non-reasoning model for the company.
“A top goal for us is to unify o-series models and GPT-series models by creating systems that can use all our tools, know when to think for a long time or not, and generally be useful for a very wide range of tasks.” Altman explained the roadmap for GPT-5 in his post.
Notably, OpenAI’s newer ‘o’ series models like o3 Mini and o1 are reasoning first models that are aimed at mimicking human level thinking capabilities in order to solve more complex problems like coding, maths and reasoning related tasks. The reasoning models while offering nuance do take a bit more time than their standard counterparts and may not be ideal for simple queries.
Altman had confirmed that OpenAI will do away with the model picker in GPT-5 and integrate a lot of companies existing models. OpenAI plans to give ‘unlimited access’ to GPT-5 at standard intelligence settings (subject to ‘abuse thresholds’) while Plus and Pro subscribers are planned to get an ever higher level of intelligence.
OpenAI’s new language models:
Altman also confirmed that OpenAI is rolling out 3 new models o3 Pro, o3 and o4 Mini – the latter two scheduled for release in the coming few weeks.
Notably, OpenAI had first announced its o3 and o3 Mini language model during a 12 day event in December. Amid the rise of China’s DeepSeek AI that provides free reasoning abilities, OpenAI also rushed to bring out its o3 Mini model in late January and also made it available to free users with certain restrictions. However, this is the first time since then that the company has given a tenative deadline for the rollout of o3.
Apart from these models, OpenAI also plans to launch its first open weights model since GPT-2 in the coming months which Altman has previously confirmed will be subject to additionall safety evaluations.
