After Musk liked it, he hugged his face and paid out of pocket: GLM-5.2 6 hours globally free, open source will surely win!

Last night, the world's largest AI open source community, Hugging Face, announced an unprecedented decision: it will provide global free computing power support for up to 6 hours for the flagship model GLM-5.2 of KNOWLEDGE ATLAS (02513) AI NKY Medical at its own expense. This is the first time that Hugging Face has financially supported a domestic model in this "exclusive VIP channel", and overseas netizens have praised this "investment" wave! The popularity of GLM-5.2 continues to rise. On June 18th, Tesla, Inc. (TSLA.US) founder Musk replied to a netizen question on social media about "what is your current expected timeline for China to reach the Fable level? GLM-5.2 undoubtedly shortens the gap," by saying, "perhaps Q1 2027." KNOWLEDGE ATLAS founder and chief scientist Tang Jie then responded on X, saying, "It won't take that long." Marc Andreessen, co-founder of top Silicon Valley venture capital firm a16z, also joined the conversation, as did Perplexity CEO Aravind Srinivas on X. Many foreign netizens sided with them, indicating that Musk's prediction was "too conservative" and it immediately trended with #won't take that long. With Fable5 being taken offline, KNOWLEDGE ATLAS GLM-5.2 steps into the spotlight Musk's judgment came as the new model from Anthropic sparked global discussions. On June 9, the American model giant Anthropic released its strongest AI models to date - Claude Fable 5 and Claude Mythos 5. On June 13, the US Department of Commerce issued an export control order prohibiting providing these models to "non-US users", leading Anthropic to ultimately choose to completely take down the two models. Along with the export control, there was a tightening of access to top closed-source models in the US. The access to cutting-edge closed-source models shifted from account subscriptions in the past to a more identity, regional, and qualification-based access mechanism. On the same day as the US imposed export controls on Anthropic, KNOWLEDGE ATLAS announced that its strongest model GLM-5.2 was about to be launched, leaving a hint in the article: "At the moment when some cutting-edge models suddenly become unavailable, we choose to believe in another way: cutting-edge intelligence should not only belong to a few, nor should it be taken back by a few rules at any time. It should be open, available, and buildable, and serve every developer." On June 17, KNOWLEDGE ATLAS officially announced the launch of GLM-5.2, with the most lenient MIT license for open-source use, allowing for free commercial use. It achieved a stable long-range programming capability level of 1M. On the architectural level, GLM-5.2 introduced the IndexShare mechanism, where every four layers of sparse attention shared the same indexer, reducing the computational load of each token by about 2.9 times in a context of millions of tokens. In the Artificial Analysis comprehensive rankings, GLM-5.2 scored 51 points, entered the top three global models, and ranked as the SOTA open-source model; in authoritative benchmarks such as FrontierSWE, Terminal-Bench, etc., the gap between GLM-5.2 and the top international model Claude Opus 4.8 narrowed to 1%-4%. Through multiple benchmark evaluations, it is evident that GLM-5.2 is on par with the Claude Opus and Fable 5 models. This also means that the code capability of open-source models has reached the industry-recognized level of top closed-source models for the first time. Chinese large models join the ranks of model "big three" Since the emergence of DeepSeek, Chinese large model enterprises have embraced open source, leading to a rapid rise of the open-source model legion and narrowing the gap between Chinese models and top American models. On the popular AI model access platform OpenRouter, the usage of Chinese models has increased from 1.2% by the end of 2024 to over 50%, surpassing American models in terms of overall usage. However, surpassing in usage does not necessarily mean leading in cutting-edge capabilities. For a long time, Chinese models were seen more as cost-effective alternatives, open-source supplements, or catch-up players for specific capabilities. What sets GLM-5.2 apart is its attempt to enter the same competitive range as closed-source models like Claude and GPT in smart programming and long-range complex tasks. During the Chinese New Year period, the KNOWLEDGE ATLAS GLM-5 model debuted on OpenRouter under the alias of Pony Alpha, offering free API calls and experiences to developers worldwide. This led to widespread speculation in the global tech community, and its paradigm shift from "single-turn code completion" to "smart body engineering" established the leading position of Chinese open-source models globally. The subsequent release of GLM-5.1 further enhanced code performance, and the ability to work for 8 hours on long-range tasks truly pushed Chinese large models beyond the basic usability line set by Anthropic. The arrival of GLM-5.2 has directly advanced the code capability to the level of world-class models. Renowned American open-source model researcher Nathan Lambert commented, "KNOWLEDGE ATLAS GLM-5.2 surpasses Alphabet Inc. Class C Gemini in smart body capability at the moment, which is a highly significant achievement," indicating that for the first time, a model enterprise from China has challenged the "big three" of Alphabet Inc. Class C, OpenAI, and Anthropic, who previously monopolized global model capabilities. Chinese open-source ecology offers equal access tickets to global models The release of KNOWLEDGE ATLAS GLM-5.2 is a microcosm of the developmental progress of Chinese large models in the AI field. Musk's prediction that Chinese large models may reach the level of Fable in the first quarter is not just about the specific timing accuracy, but reflects a reevaluation of the global market on the speed at which China is catching up with cutting-edge models. A report by Fortune stated, "The US ban on Fable and Mythos proves that China's broader efforts towards technological self-sufficiency are correct. China's acceleration towards technology self-sufficiency is evident." On the first day of its release, GLM-5.2 completed full compatibility with domestic computing platforms such as Huawei Ascend, Pingtouge, Moore Thread, Cambricon, Kunlunxin, Muxi, Haiguang, and Biren. The previously released Chinese large model DeepSeek V4 has also completed full-stack adaptation on such domestic computing platforms, indicating that the comprehensive open ecosystem from China is empowering the global AI industry for open, inclusive, mutually beneficial ecological prosperity. As closed-source cutting-edge models in the US increasingly resemble a pass that requires qualification checks, GLM-5.2 offers a different path: cutting-edge capabilities do not necessarily have to be closed off in a few platforms, but can be open-source and handed over to developers worldwide to rebuild. In the landscape of global large model competition, a Chinese variable has emerged for the first time as both close to top closed-source experience and choosing an open-source route. Musk's much-discussed prediction of "Q1 2027". From generationally lagging behind, to breaking through in a single point, and now stepping into the same arena with the toughest smart bodies and long-range complex tasks, Chinese large models have shortened the unit of measurement for catching up from "years" to "months". In this sense, GLM-5.2 is not just the flagship model release from KNOWLEDGE ATLAS, but a landmark point for Chinese large models moving from a narrative of catching up to competing at the same table.