The company announced it was testing advanced reasoning models, o3 and o3 mini, designed to address more complex tasks compared to earlier iterations.