Orion, OpenAI's new upcoming model, won't launch on ChatGPT

It’s been almost two years since OpenAI first launched ChatGPT. The AI chatbot has now seen models come and go, like the current GPT-4o and o1. So, to commemorate the second anniversary, the Microsoft-backed company is set to launch a new flagship model called Orion in December.

Exclusively revealed by The Verge, it was reported that OpenAI will allow select partner companies to develop their own AIs using Orion instead of a broad release on ChatGPT like it was with the o1 model that excels in reasoning.

Although Orion is positioned as a successor to GPT-4 with reportedly up to 100 times more capability than the previous model, it’s uncertain if it will be named GPT-5 publicly. Microsoft engineers are also preparing to host Orion on Azure, potentially by November.

Earlier in August, The Information reported that OpenAI was using the o1 model—which excels in reasoning—to train Orion and provide synthetic data. Though more costly and slower than GPT-4o, the o1 also excels at coding and complex problem-solving, with developer access priced at %15 per million input tokens and $60 for output tokens. It has less tendency to hallucinate, too.

Orion comes at a crucial time as OpenAI transitions to for-profit status. The AI company is also reportedly seeking more access to Microsoft’s advanced Nvidia H100 GPUs to host Orion, which are better suited for training sophisticated AI models, but Microsoft is balancing this request against its own internal demands and paid AI services.