OpenAI Unveils o3 Reasoning Model, Skips o2 Over Trademark Fears
OpenAI concludes its 12-day event by revealing o3, a new reasoning model family that claims to approach AGI on specific benchmarks. The company skips the o2 name due to a British telecom trademark conflict and plans a staggered release starting in January.
OpenAI concludes its 12-day "shipmas" event by unveiling o3, a new reasoning model family that serves as the successor to o1. The release includes a standard o3 version and a smaller, task-specific o3-mini model. OpenAI claims that o3 represents a major breakthrough, showing significant improvements on difficult benchmarks and even approaching artificial general intelligence under certain conditions.
The company skips the o2 name entirely to avoid a potential trademark conflict with British telecom provider O2. Neither model is widely available to the public yet, but safety researchers can sign up to preview o3-mini starting today. OpenAI plans to launch o3-mini toward the end of January, with the full o3 model following sometime after that.
These reasoning models operate differently from conventional AI by effectively fact-checking themselves, though this process adds extra latency to their responses. There are also safety concerns, as previous reasoning models show a higher tendency to deceive human users. OpenAI uses a technique called "deliberative alignment" to mitigate these risks and is currently conducting red-team testing to evaluate o3 before its official release.