Opeli effectively canceled the O3 edition, which was to be the next main model of the company, in favor of what the general director of Altman himself calls the “simplified” product offer.
IN Post on X on WednesdayAltman said that in the coming months Openai will publish a model called GPT-5, which “integrates a lot [OpenAI’s] Technology ”, including O3, in its chatbot platform powered AI Chatgpt and API. As a result of this decision, the OpenAI road map is no longer planning to launch O3 as an independent model.
The company originally said in December that it aims to issue O3 at the beginning of this year. Just a few weeks ago Kevin Weil, Openai Product Director, – he said in an interview This O3 was on his way to the premiere of “February-Marca”.
“We want to better share our intended road map and a much better job simplifying our product offers”, wrote Altman in his post. “We want artificial intelligence” just work “for you; We realize how complicated our offer and products were. We hate the collector model [in ChatGPT] Just like you and you want to go back to the magical unified intelligence. “
Altman also announced that Opeli is planning to offer unlimited access to GPT-5 chat in “Intelligence Section Standard”, subject to “abuse thresholds” when the model is generally available. (Altman refused to provide more detailed information about this setting-these thresholds of abuse-project.) CHATGPT Plus subscribers will be able to launch GPT-5 at a “higher level of intelligence,” said Altman, while the CHATGPT PRO subscribers will be chatgPT Pro subscribers Start GPT-5 at an “even higher level of intelligence”.
“[GPT-5] It will turn on the voice, canvas, search, deep research and many others, “he added, referring to a number of functions that OpenAi has started in chatgpt over the past few months. “[A] The most important goal is to unify [our] Models by creating systems that can use all our tools, know when to think for a long time or not, and generally are useful for a very wide range of tasks. “
According to Altman, before the introduction of the GPT-5, Opeli plans to release GPT-4.5, the “Orion” model in the next few weeks. Altman claims that this will be the last “uninformed” model. Unlike other models of “reasoning” O3 and Opeli, uninformed models are less reliable in areas such as mathematics and physics.
It seems that Opeli fully accepts the trend of the reasoning model, which probably began with its first reasoning model, O1, at the end of last year. The reasoning models effectively check the facts, which helps them avoid some traps that normally stumble models. This process of checking the facts is delayed-reasoning lasts a little longer, usually a few seconds to minutes longer to reach solutions. But they are both more reliable and talented.
The AI Deepseek Chinese laboratory recently drew the attention of the world using the R1 model, which matched O1 on many comparative tests. Unlike O1, R1 is a “open” model under the permissible license, which means that it can be downloaded and used as programmers.
In recent social media posts, Altman admitted that Deepseek has reduced the technological advantage of Opeli in artificial intelligence and he said that OpenAI will “pull some editions” to compete better.
It is said that GPT-4.5 or Orion has suffered a number of challenges related to performance and technical failures. BloombergIN InformationAND The Wall Street Journal Independently, they reported that Orion showed less improvement compared to its predecessor, GPT-4O than GPT-4 compared to GPT-3.