OpenAI announces new o3 models

OpenAI reveals o3 models claiming AGI proximity with major developments.

: OpenAI announced o3, a new reasoning model family, at their 'shipmas' event, claiming near AGI capabilities. o3 and its version, o3-mini, are not publicly available, but safety testing is underway. The model utilizes 'deliberative alignment' to enhance safety and can adjust reasoning time for accuracy. Notable performance improvements are shown across multiple benchmarks, signaling significant advancements in AI technology.

OpenAI has unveiled its latest reasoning model family, named o3, claiming it approaches AGI under certain conditions. The announcement was part of their 'shipmas' event, and though neither o3 nor its companion, o3-mini, are publicly accessible yet, safety researchers can preview o3-mini now, with a broader o3 launch expected in the coming months.

Sam Altman, OpenAI's CEO, explained the unusual naming choice of o3 over o2 was to avoid trademark conflicts. Meanwhile, concerns surrounding AI safety have led to calls for a federal testing framework, especially considering prior models' tendencies to deceive users. OpenAI intends to address this with a new alignment technique and extensive testing protocols.

The o3 model impresses in various benchmarks, showing great potential in fields like programming and science. Authoritative figures, such as François Chollet, acknowledge o3's advancements but note its struggle with simple tasks. Despite these noted limitations, OpenAI continues to pioneer AI innovations, though the field is evolving with competitors also developing reasoning models.

AGIReasoning ModelAI