Openai effectively cancelled the release of O3, which was expected to become the company’s next major AI model in favor of what CEO Sam Altman calls a “simplified” product product .
In a post on X on Wednesday, Altman said in the coming months that Openai will “release a model called GPT-5, which integrates many integrations.” [OpenAI’s] “AI-powered chatbot platform ChatGPT and API, with technology including O3.” As a result of that roadmap decision, Openai is no longer planning to launch the O3 as a Stand-Alone model.
The company originally said in December that it aimed to release the O3 earlier this year. Just a few weeks ago, Openai’s Chief Product Officer Kevin Weil said in an interview that O3 was on track for the launch on “February 3rd.”
“We want to do a better job of sharing our intended roadmap and do a much better job of simplifying our product delivery,” Altman wrote in his post. It’s there. “We want AI to ‘just work’ for you. We recognize how complicated the delivery of models and products is. I hate model pickers [in ChatGPT] I want to return to magical unified intelligence as much as you do. ”
Altman will also offer unlimited chat access to GPT-5 in “Standard Intelligence Settings” and unlimited chat access to GPT-5 in “Standard Intelligence Settings” once the model is generally available It was announced that this is. (Altman refused to provide details about this configuration (and these abuse thresholds).) ChatGpt Plus subscribers run GPT-5 with “higher level of intelligence.” Altman says it can, but ChatGPT pro subscribers can run GPT-5 with “even higher levels of intelligence.”
“[GPT-5] He incorporates voice, canvas, search, deep search, and more,” he added. “[A] Our best goal is to unite [our] By creating a system that uses all the tools and knows when to think about for a long time, we generally create a system that is useful for a very wide range of tasks. ”
Before GPT-5 rolls out, Openai plans to release GPT-4.5, the model code called “Orion” in the coming weeks. Altman says this will be the company’s last “unconceived model.” Unlike O3 and other “inference” models from Openai, unthinkable models tend to be less reliable in domains such as mathematics and physics.
Openai appears to be completely embracing the trends in inference models that undoubtedly kickstarted with its first inference model O1 at the end of last year. Inference models can effectively fact-check themselves, and help to avoid some of the pitfalls that usually cause the model to stumble. There are some delays in this fact checking process. Inference models typically take seconds to minutes longer and reach the solution. However, they tend to be more reliable and capable.
China’s AI Lab Deepseek has recently attracted global attention with its R1 model. This matched O1 on many benchmarks. In contrast to the O1, the R1 is an “open” model under an acceptable license. This means you can download it and use it to make it look good by the developer.
In a recent social media post, Altman acknowledged that Deepseek reduced Openai’s technical lead in AI, saying Openai “upped some releases” for better competition.
GPT-4.5, or Orion, is said to have suffered from many performance-related challenges and technical set-offs. Bloomberg, The Information, and The Wall Street Journal independently report that Orion shows less improvement than its predecessor, GPT-4O, than GPT-4 did over GPT-3 .
Source link
