A netizen asked, "When will Chinese large models reach the Fable level?" Musk replied, "Possibly Q1 next year," and Zhipu CEO Tang Jie said, "It won't take that long."
```
Centered around Anthropic’s frontier models being removed due to export controls, a public debate about the timetable for China’s large models to catch up is unfolding on the X platform. The clash of multiple viewpoints reflects the accelerated reshaping of the AI industry landscape.
After Zhipu AI released GLM-5.2, on June 18, an X user asked independent researcher and AI development blogger Teortaxes, “When will China’s large models reach Fable-level?”
Teortaxes gave a judgment of a 7-month gap, Musk responded immediately with “possibly (Q1) 2027,” while Zhipu AI CEO Tang Jie directly stated “it won’t take that long.”
Musk subsequently added that catching up in benchmarks is relatively easy, but judging by “real practicality,” Q1 next year is already quite impressive.
The context for this conversation is that GLM-5.2 achieved a score of 74.4 on the key programming benchmark FrontierSWE, lagging only about 1 percentage point behind Anthropic’s top closed-source model Opus 4.8, and surpassing GPT-5.5.
Wallstreetcn previously mentioned that the U.S. Department of Commerce imposed export controls on Anthropic’s Fable 5 and Mythos 5, requiring government permits before granting access to any foreigners. Anthropic immediately closed global access to both models.
The dual forces of technological catch-up and AI costs are reshaping the global AI competitive landscape. As Zerohedge pointed out, when a “10% intelligence gap” may correspond to a “90% cost advantage,” whether there is a mismatch in capital allocation exceeding $5 trillion will become the core issue the market continues to ask.
The Timetable Debate: From 7 Months to “It Won’t Take That Long”
The starting point of the debate comes from Teortaxes’ evaluation of GLM-5.2’s performance.
He thinks GLM-5.2’s overall capability is in the Opus 4.7–4.8 range. He claims Opus’ visual capability is weak, and after considering visual ability, the gap for Chinese large models is around 7 months.
His calculation logic is that the Mythos model reaches Preview status (i.e., ≥Opus 4.8 functional level) in early February 2026. By analogy, the time window for Chinese models to reach full “Fable” level is about November to December 2026.
Musk’s assessment is more conservative, replying only “Probably Q1,” corresponding to the first quarter of 2027.
But he then added an important distinction: At the benchmark testing level, catch-up progress may indeed be impressive; but judged by “real practicality,” even reaching Q1 is quite difficult.
He pointed out that Anthropic’s advantage lies in focusing on improving real intelligence, which doesn’t show up in benchmark scores but is directly reflected in revenue.
Regarding Musk’s remarks, some AI industry insiders think his prediction is somewhat conservative, and the time gap between Chinese and American model levels may be less than 7 months.
Google DeepMind CEO Demis Hassabis also previously said that Chinese AI models may be “only a few months behind” overseas models in capabilities.
Tang Jie’s response was even more brief and direct: “won’t take that long.” This statement implies Zhipu’s confidence in its own iteration speed, but no specific timeline was offered.
AI research institute Proximal commented that GLM-5.2 is “the first model that truly closed the huge technological gap between Anthropic/OpenAI and other model providers.”
GLM-5.2’s Technical Position: Approaching Closed-Source Frontiers, Gap Remains
GLM-5.2’s technical indicators are the main basis for this timetable discussion.
On June 15, Zhipu announced the official launch and open-sourcing of the new flagship large model GLM-5.2. On Code Arena, a frontend development evaluation system with millions of global users participating in blind tests, the model ranks first among available models worldwide.
Unlike previous models focused on instant Q&A capabilities, GLM-5.2 mainly tackles “long-range tasks”—enabling AI not just to handle instant Q&A, but to work continuously for hours like humans, independently completing complex large projects.
According to release data, GLM-5.2 has 753B parameters, a stable 1M token context window, and is fully open-source under the MIT license.
On the long-range programming benchmark FrontierSWE, GLM-5.2 scored 74.4, Opus 4.8 scored 75.1—a difference of about 1 percentage point. It also exceeded GPT-5.5’s 72.6 and Opus 4.7.
On PostTrainBench (testing Agent training on small models), GLM-5.2 scored 34.3, ranking second, just behind Opus 4.8’s 37.2 and ahead of GPT-5.5’s 28.4.
The gap remains. On the hardest SWE-Marathon benchmark, GLM-5.2 scored 13.0, while Opus 4.8 scored 26.0—a clear gap.
Combined, the three benchmarks convey that: For medium-complexity long-range tasks, GLM-5.2 has entered the same competitive arena as top closed-source models; for the most extreme complex tasks, it still lags but leads among open-source models.
Anthropic Models Pulled Down, Open-Source Substitution Logic Accelerates
Another key background for this discussion is the sudden withdrawal of Anthropic’s Fable 5 and Mythos 5.
Wallstreetcn mentioned that the Anthropic event revealed the fragility of closed-source commercial models in availability and brought strategic value beyond technical aspects for the open-source camp.
At the release of GLM-5.2, Zhipu AI CEO Tang Jie made a statement on the X platform:
At a time when frontier models are arbitrarily cut off from access, we are more convinced of one thing: science should be global. The road to AGI must never be surrounded by high walls.
This timing elevates open source from a technical route to the narrative level of technological sovereignty. The tweet received over 880,000 views and 252 replies within 24 hours.
Orient Securities believes open-source models, with features like open weights, autonomy, and local deployment, are a superior choice for avoiding geopolitical risks and ensuring business continuity.
As domestic models lead in performance, most are open-source and offer lower API calling costs, Chinese models already hold a leading position on token distribution platforms like OpenRouter. With Anthropic’s two models removed, API call volumes for domestic models are expected to increase further.
But from a medium- to long-term perspective, lower costs and access thresholds may drive simultaneous growth in token consumption and computing power demand. For investors, rising open-source model market share and high computing power demand are becoming the core variables for the AI industry chain’s reassessment.
Risk Warning and DisclaimerThe market has risks, investment requires caution. This article does not constitute personal investment advice and does not take into account individual users' special investment objectives, financial conditions, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article suit their particular situation. If investing based on this, responsibility is at your own risk. ```