OpenAI has officially shelved the rollout of o3, which was anticipated to be the company’s next significant AI model, as CEO Sam Altman opts for a more “streamlined” product strategy.
In a message on X on Wednesday, Altman revealed that OpenAI plans to introduce a model named GPT-5 in the near future. This new model will “incorporate a great deal of [OpenAI’s] technology,” including o3, within its AI-driven ChatGPT platform and API. Consequently, OpenAI has abandoned its plans to release o3 as a separate entity.
Originally announced in December, the o3 model was projected for launch in the early part of this year. Just weeks ago, Kevin Weil, OpenAI’s chief product officer, indicated in an interview that o3 was set for a “February-March” debut.
“We aspire to improve our communication regarding our intended roadmap and to significantly streamline our product offerings,” Altman noted in his post. “We want AI to operate seamlessly for you; we understand how convoluted our model and product offerings have become. We share in your frustration about the model picker [in ChatGPT] and aim to restore a more intuitive experience.”
Altman further revealed that OpenAI intends to provide unlimited access to GPT-5 at a “standard intelligence level,” which will be subject to “abuse thresholds” once the model is generally accessible. He mentioned that ChatGPT Plus subscribers would have the ability to use GPT-5 at an “enhanced intelligence level,” while those with ChatGPT Pro subscriptions could operate GPT-5 at an “even higher intelligence level.”
Altman clarified that o3 will be integrated into GPT-5.
“[GPT-5] will feature voice capability, canvas, search, deep research, and various other functionalities,” he elaborated, referencing a suite of features OpenAI has introduced in ChatGPT in recent months. “[A] primary objective for us is to unify [our] models by developing systems that utilize all our tools, discern when to engage in deep reasoning, and ultimately be beneficial across a myriad of tasks.”
Before the GPT-5 launch, OpenAI plans to unveil the GPT-4.5 model, codenamed “Orion,” within the next few weeks, as per Altman’s message on X. He indicated that this will be the company’s final “non-chain-of-thought model.” These non-chain-of-thought models, unlike o3 and others, typically struggle with reliability in areas such as mathematics and physics.
It appears that OpenAI is wholeheartedly embracing the trend of reasoning models, a concept it essentially initiated with its inaugural reasoning model, o1, last year. Reasoning models effectively validate their outputs, helping them navigate the challenges that often hinder other models. This validation process results in slight delays—meaning reasoning models may take a few seconds to minutes longer to produce answers—but they generally demonstrate greater reliability and capability.
Recently, the Chinese AI lab DeepSeek captured global attention with its R1 model, which matched o1 on several benchmarks. Unlike o1, the R1 model is “open” under a permissive license, enabling developers to download and utilize it as they choose.
In his recent social media updates, Altman acknowledged that DeepSeek has narrowed OpenAI’s technological advantage in the AI sector and mentioned that OpenAI would expedite some releases to stay competitive.
GPT-4.5, or Orion, reportedly faced numerous performance challenges and technical obstacles. Independent sources, including Bloomberg, The Information, and The Wall Street Journal have reported that Orion’s improvements compared to its predecessor, GPT-4o, are not as significant as those between GPT-4 and GPT-3.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
