Mark my words, by the end of 2027, there will be an open weights model that is better than anything OpenAI and Anthropic are capable of making. They will lose at inference scaling too.