I doubt o3s costs will be lower for that performance. They juice their benchmark results by letting it spend $100k in thinking tokens.