OpenAI CEO Sam Altman revealed a significant shift in the company's release plans on Friday, announcing that two intermediate models will arrive before its highly anticipated GPT-5.
"Change of plans: We are going to release o3 and o4-mini after all, probably in a couple of weeks, and then do GPT-5 in a few months," Altman wrote on X Friday.
change of plans: we are going to release o3 and o4-mini after all, probably in a couple of weeks, and then do GPT-5 in a few months.
there are a bunch of reasons for this, but the most exciting one is that we are going to be able to make GPT-5 much better than we originally…
The surprise announcement comes as OpenAI grapples with technical complexities in its flagship model development. Altman admitted the company "found it harder than we thought it was going to be to smoothly integrate everything" into GPT-5, suggesting the staggered release will help ensure sufficient capacity "to support what we expect to be unprecedented demand."
The move places OpenAI in an increasingly crowded field of AI heavyweights rolling out advanced models. Google recently launched Gemini 2.5 Pro, which boasts 1 million tokens of context and has been widely regarded as the best reasoning and coding model available—and is free to use.
Meanwhile, DeepSeek R2, Grok-3, and Claude 3.7 Sonnet with extended thinking capabilities are all slated for imminent release—each undercutting OpenAI's reasoning model on price.
The best reasoning models available. Image: Artificial Analysis
Altman teased a silver lining: releasing the intermediate models will give OpenAI more time to supercharge GPT-5.
"The most exciting [reason] is that we are going to be able to make GPT-5 much better than we originally thought," he wrote. GPT-5 is expected to be fully multimodal, merging all of OpenAI's specialized models into a single system. That would eliminate the current need for ChatGPT to switch between reasoning models, standard language models, and image generation models based on the prompt. Instead, all these functions would be handled by a unified model.
Technical specifications for o3 and o4-Mini remain under wraps, but they're expected to bridge the capabilities gap between GPT-4 and the forthcoming GPT-5, which industry watchers believe will feature substantial improvements in reasoning, planning, and memory functions.
OpenAI's latest release, the reasoning-focused o1 Pro, came with eyebrow-raising pricing: $150 per million tokens (~750,000 words) for input, and $600 per million tokens generated. That's double the input cost of GPT-4.5 and 10 times the price of regular o1. And for reference, DeepSeek R1 costs less than $1 per million tokens.
The revised roadmap arrives just days after OpenAI closed a historic $40 billion funding round—the largest single fundraising event by any private tech company.
Generally Intelligent Newsletter
A weekly AI journey narrated by Gen, a generative AI model.
Elon Musk's AI chatbot Grok had a strange fixation last week—it couldn't stop talking about "white genocide" in South Africa, no matter what users asked it about.
On May 14, users started posting instances of Grok inserting claims about South African farm attacks and racial violence into completely unrelated queries. Whether asked about sports, Medicaid cuts, or even a cute pig video, Grok somehow steered conversations toward alleged persecution of white South Africans.
The timing raised concern...
Amateur musicians now have a digital band of AI assistants ready to help at every stage of music creation—from inspiration to final mastering. Whether you're struggling with writer's block, can't quite nail that perfect melody, or just want to experiment with new sounds, there's an AI tool that fits your needs.
In fact, there’s an overwhelming amount of tools for musicians and would-be musicians.
After reviewing dozens of AI music tools from our recent research, here are the best options for eac...
Right this moment, thousands of AI agents trained on billions of Solana transactions are acting as meme coin traders, competing to rack up as many tokens as they can while fighting through simulated economic disasters and pump-and-dump schemes—just like real traders.
And so far, they really seem to be into hot dogs.
It’s all part of a crypto game called DX Terminal, with AI agents that were minted as NFTs on Coinbase’s Ethereum layer-2 network Base. They’re all competing as part of a one-week ex...