Zhipu AI founder and Tsinghua University professor Tang Jie engaged in a social media exchange with Tesla founder Elon Musk on June 18 regarding when Chinese AI models would reach the level of Anthropic's Fable. A netizen asked Musk when Chinese large models would achieve Fable-level performance, noting that Zhipu's GLM-5.2 had narrowed the gap, to which Musk replied 'Probably Q1' (referring to 2027 first quarter in his response). Tang Jie directly responded to Musk stating 'Won't take that long.' The exchange occurred against the backdrop of Anthropic receiving a US government order on June 12 (local time) requiring immediate shutdown of Claude Fable 5 and Mythos 5 access to all non-US nationals, including the company's own foreign employees.
Tang Jie Challenges Musk's Timeline Prediction
On June 18, a social media user asked Musk 'When do you estimate Chinese large models will reach Anthropic's Fable level? Zhipu GLM-5.2 has undoubtedly narrowed the gap.' Musk responded 'Probably Q1' (indicating 2027 first quarter). Tang Jie, Tsinghua University professor and Zhipu AI founder, replied to Musk: 'Won't take that long.'
Musk followed up stating: 'Looking at benchmark scores alone, perhaps so; but judging by actual practical value, catching up by Q1 next year would already be quite impressive. Anthropic's approach is correct—they focus entirely on polishing intelligence capabilities with real practical value, capabilities that don't show up on leaderboards but genuinely convert to revenue.'
Tang Jie responded: 'Focus is the only thing we need, especially focusing on what the essence of intelligence really is...'
Zhipu's official account joined the exchange, posting three 'taking notes' emoji images. Silicon Valley investor and venture capitalist Marc Andreessen retweeted the exchange with the comment 'Interesting.'
In Musk's comment section, multiple users questioned his position, with one asking: 'Elon, are you on Anthropic's (Fable's parent company) side, or humanity's side?'
Zhipu Stock Hits Record High as GLM-5.2 Launches
As of June 18 closing, Zhipu (02513.HK) rose 26.15% to 2,094 Hong Kong dollars, reaching an all-time high with total market capitalization exceeding 900 billion Hong Kong dollars. Data shows Zhipu is currently China's highest-valued AI company.
Zhipu announced the official launch and open-sourcing of its new-generation flagship model GLM-5.2. The model is built for long-range task capabilities, with core features including 1M lossless context, enhanced coding ability, Day 0 adaptation to domestic computing platforms, and MIT open-source licensing.
According to Zhipu's official introduction, on CodeArena—a front-end development evaluation system with global blind testing by millions of users—GLM-5.2 achieved first place among globally available models. In authoritative evaluations including FrontierSWE and Terminal-Bench, GLM-5.2 narrowed the gap with overseas leading model Claude Opus 4.8 to 1% to 4%, ranking as the highest open-source model.
In FrontierSWE testing, GLM-5.2 scored only 1% lower than Opus 4.8, surpassing GPT-5.5. In Terminal-Bench 2.1 evaluation, it scored 81.0, a substantial 17.5 percentage point increase from the previous generation GLM-5.1's 63.5.
The model was first made available to all GLM Coding Plan users. Online inference completed Day 0 adaptation with domestic computing platforms including Huawei Ascend, Pingtouge, Moore Threads, Cambricon, Kunlun Core, Muxi, Hygon, and Biren. GLM-5.2 weights were uploaded to HuggingFace and ModelScope, open-sourced under the most permissive MIT license, freely downloadable, deployable, and commercially usable. API simultaneously launched on BigModel open platform and Z.ai.
Anthropic Shuts Down Fable Models Under US Government Order
Zhipu announced its opening at a time when overseas 'strongest models' were required by the US government to urgently go offline. On local time June 12, overseas AI giant Anthropic issued a statement: the company received an emergency letter from the US government at 5:21 PM that afternoon. The US government, citing national security department directives, issued an export control order requiring Anthropic to immediately cut off access to its most advanced AI models Claude Fable 5 and Mythos 5 for all foreign nationals—whether within the US or abroad, including the company's own non-US citizen employees.
This meant Anthropic was forced to completely disable these two flagship industry models—released less than 72 hours prior—for all non-US users globally. Anthropic stated the US government's decision was based on 'serious misunderstanding.'
On June 13, Zhipu announced GLM-5.2 opened to all GLM Coding Plan users at 17:21, covering Lite, Pro, Max, and team versions. In the announcement, Zhipu explicitly stated: 'At a moment when some frontier models suddenly become unavailable, we choose to believe in another path: frontier intelligence should not belong only to a few, nor should it be arbitrarily withdrawn by a few rules. It should be open, available, buildable, and serve every developer.'
FAQ
What did Tang Jie say to Elon Musk on June 18 about Chinese AI models?
On June 18, Zhipu AI founder Tang Jie responded to Elon Musk's prediction that Chinese models would reach Anthropic Fable level by Q1 (2027 first quarter) by stating 'Won't take that long.' This was in response to a netizen's question about when Chinese large models like Zhipu's GLM-5.2 would achieve Fable-level performance.
What happened to Anthropic's Fable models on June 12?
On local time June 12 at 5:21 PM, Anthropic received an emergency US government letter citing national security directives, ordering immediate shutdown of access to Claude Fable 5 and Mythos 5 models for all foreign nationals globally, including the company's own non-US citizen employees. The models had been released less than 72 hours before the shutdown order.
How did Zhipu's GLM-5.2 perform in benchmark tests?
GLM-5.2 achieved first place among globally available models on CodeArena's front-end development evaluation system with millions of users. In FrontierSWE testing, it scored only 1% lower than Claude Opus 4.8 and surpassed GPT-5.5. In Terminal-Bench 2.1 evaluation, it scored 81.0, a 17.5 percentage point increase from GLM-5.1's 63.5.
