China's open source restrictions on large models are gradually tightening, while the United States' large model reverse distillation+closed source MiniMax 2.7 is open-source but can be deployed on its own without any issues. Commercialization for profit requires authorization. Qwen 3.6 Plus， Not yet open sourced, serving our own products, with decent modeling capabilities, ranking second domestically only to GLM5.1. Several domestic top models have shown a clear shift towards semi open source and a pursuit of commercialization A year ago, this was unimaginable in the Chinese AI industry. At that time, the game rules for domestic large models were only one word, volume. If you open source, I will also open source. If you are free, I am even more free than you. ByteBean Pack charges the Token to 0.0008 yuan/thousand Token, and Alibaba Qianwen directly sends it. DeepSeek, not to mention, believes in open source. At that time, the entire industry was filled with an almost crazy sense of "open source justice". Whoever dared to shut down the source was at odds with the developers and stood on the opposite side of history. But look at it now. In just a few months, the wind direction changed. It's not a change in one family, it's a collective shift. This is the new script for domestic large models in 2026, with flagship closed source and small model open source, or 'open source but you can't afford it'. In the words of industry insiders, open sourcing immediately is equivalent to tearing down one's own moat. Giants like Alibaba, ByteDance, and Tencent, due to their abundant computing resources, are likely to shut down their flagship models, which is in their best interest. The three little dragons of domestic models (GLM, Kimi, and minimax) will continue to maintain open source. Due to limited computing power, they cannot meet all the demands, but they need influence and external cooperation. Therefore, the commercialization proposed by minimax requires authorization, and similar cooperation between Kimi and cursor is the best solution when the product scale and revenue reach a certain scale. Why did it become like this? Let's talk about external reasons first. In February of this year, Anthropic directly named and reported three Chinese companies, DeepSeek、 The Dark Side of the Moon MiniMax， They said they set up about 24000 fraudulent accounts and had over 16 million interactions with Claude, conducting an 'industrial scale distillation attack'. How to divide them specifically? MiniMax has over 13 million views, Moonshot has over 3.4 million views, and DeepSeek has about 150000 views. Anthropic has also raised this matter to the level of national security, stating that the distilled model may be used for cyber attacks, spreading false information, and even military purposes. On April 7th, something even more explosive happened. OpenAI, Anthropic, and Google have officially launched an intelligence sharing mechanism specifically targeting the distillation behavior of Chinese AI companies through the Frontier Model Forum (FMF) established in 2023 in collaboration with Microsoft. Bloomberg quoted insiders as saying that the three giants are trying to curb Chinese competitors from extracting capabilities from advanced AI models in the United States. OpenAI and Google usually wish they could strangle each other, but now they are holding hands and saying, 'We need to defend against China together.'. In this environment, did you immediately open source all the strongest models? Are you afraid you're crazy? Let's talk about internal reasons. To be frank, Chinese big model manufacturers have finally figured out one thing: tokens can be sold at a certain price. Zhipu's API pricing increased by 83% in the first quarter of this year. What was the result? demand exceeds supply. Recently, the price has increased by another 10%, but it is still almost impossible to grab at 10 pm every day, even harder than TM Maotai. The daily average token consumption of Chinese AI has skyrocketed from 0.12 trillion in May 2024 to 140 trillion in March 2026. In less than two years, it has increased by over 1000 times. ByteDance alone accounts for 100 trillion yuan. In the past, the purpose of open source was to grab the niche, and it was the Internet game of losing money for growth. But now the industry has fully entered the "commercialization realization period" from "technological breakthrough", and the research and development investment of single round base models often amounts to billions. You can't burn billions of dollars in training while giving people away for free. This logic is actually the same as that of the United States. The United States has always been dominated by closed source business models, including OpenAI's GPT series, Anthropic's Claude, and Google's Gemini. China's shift from open-source advantages to semi closed source is actually converging in the same direction to protect core assets. From this perspective, reverse distillation and closed source may seem like two extremes, but their underlying logic is highly homologous, both aimed at protecting core technological assets and responding to commercialization pressures. The United States says' you are not allowed to copy ', while China says' I will no longer copy for you'. Of course, I also understand the other side. For many small and medium-sized developers, the past two years have been a true golden age. You can freely access the world's top open-source models, build applications, create products, and run businesses at almost zero cost. That feeling is like someone has opened the door of an entire arsenal and told you to take whatever you want. Now, this door is slowly closing. Musk responded to Anthropic's report on social media and said a classic sentence, 'How dare they steal what Anthropic stole from human programmers?'? This sentence directly punctured the window paper, and American AI companies are also distilling each other. Why did it turn into a 'national security incident' when it was the turn of Chinese companies? To be honest, distillation itself is not a taboo. Some forms of distillation in the AI industry have been widely accepted, but the legal boundaries of model distillation are not yet clear, and there are also widespread cases of mutual distillation among American companies. But the rules are like this, when you are ahead, you make the rules. When you chase, you follow the rules. Alternatively, you can go your own way. The open-source dividend period is coming to an end. The shortcut of learning through open source and catching up through distillation is being blocked. The United States is blocking exports, China is tightening its openness, and the top models that global developers can obtain for free will become increasingly scarce. But this is not necessarily a bad thing. If an industry always relies on free user acquisition, it only means that it has not yet found a true business model. The Chinese AI industry has completed this cycle in two years, and the next challenge is how to find a balance between closed source protection and ecological openness. Now that you've seen this, if you think it's good, just like it, watch it, and share it three times. Thank you for reading my article. See you next time. Will the gap between reverse distillation and closed source Chinese American models continue to widen?