The U.S. Export Control Order That Started It All

On June 12, 2026, at 5:21 PM Eastern time, U.S. Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei. The letter, citing national security authorities, ordered Anthropic to suspend all access to its two most advanced AI models — Fable 5 and Mythos 5 — for any foreign national, anywhere in the world.

The stated concern was that the models' built-in safeguards could be bypassed through jailbreaking, potentially allowing military intelligence users in China, Russia, or other countries of concern to exploit the technology. Anthropic publicly disagreed, saying the government had only provided "verbal evidence of a potential narrow, non-universal jailbreak."

Because Anthropic had no way to verify the citizenship of every user in real time, the company concluded that the only path to compliance was to disable both models for all customers globally. The move was described in multiple reports as the first export control action targeting access to an advanced AI model itself, rather than the hardware or infrastructure around it.

Z.ai's GLM-5.2: A Launch Hours After the Ban

Within 24 hours of the Anthropic announcement, Z.ai (formerly Zhipu AI, a Beijing company that grew out of Tsinghua University) made its own move. On June 13, 2026, Z.ai announced GLM-5.2, its new flagship model, available immediately to all subscribers of its GLM Coding Plan.

Key specifications of GLM-5.2, documented in Z.ai's official release notes from June 16, 2026:

• 1-million-token lossless context window, which Z.ai says improves long-horizon task capability and reduces context drift and goal forgetting.
• Two thinking-effort levels: High mode for standard tasks, Max mode for complex coding.
• 131,072 max output tokens per response.
• MIT-licensed open weights, with full open-source release scheduled for the following week.
• Z.ai claims the model achieves "open-source SOTA performance on coding and long-horizon task benchmarks."

Notably, Z.ai shipped GLM-5.2 without publishing benchmark scores at launch. Third-party evaluations were pending, and developers immediately noted the absence of numbers on standard tests like SWE-bench Verified or HumanEval.

The timing was unmistakable. Z.ai's founder Tang Jie explicitly referenced the Anthropic ban in GLM-5.2's announcement, stating: "We deeply regret the sudden restrictions on specific frontier models. The sudden blocking of access to frontier models for non-technical reasons has strengthened our conviction that science should be global."

The Musk–Tang Exchange: A Timeline Measured in Months

On June 18, 2026, AI commentator @teortaxesTex posted an estimate on X: that Z.ai's GLM-5.2 was performing at roughly the level of Claude Opus 4.7–4.8, and that a Chinese "Mythos-class" or Fable-equivalent model could arrive by November or December 2026.

Elon Musk replied: "Probably Q1" of 2027.

Z.ai co-founder Tang Jie directly countered Musk, posting that China "won't take that long." Musk then responded: "In terms of benchmarks, that would be impressive. But if measured by true practicality, even by Q1, it would be very impressive."

The significance lies in what the two sides agreed on. Both Musk and Tang were arguing over a timeline measured in months, not years. The debate was not about whether China would close the gap, but exactly when.

Why This Sequence Matters

Four takeaways:

1. Export controls have moved to model access. The U.S. directive targeted foreign access to Anthropic's most advanced AI models, not just hardware. The legal mechanism was a "deemed export" rule, treating display of controlled technology to a foreign national within the U.S. as an export.

2. China's response was immediate and public. Z.ai's GLM-5.2 launch, with its framing as an open-weight, MIT-licensed alternative to the now-restricted Anthropic models, was explicitly positioned as a countermove.

3. The gap is narrowing. The Musk–Tang exchange centered on whether China would reach Fable-class capability by late 2026 or Q1 2027. This is a debate over months, suggesting parity is on the horizon.

4. Open-source is becoming a geopolitical fault line. Z.ai emphasized GLM-5.2's open-source and open-weight availability at the same time U.S. restrictions were limiting foreign access to Anthropic's leading models. The contrast could not be starker.

The chain of events — an export ban, a rapid Chinese model launch, and a public timeline debate between two of the most prominent figures in AI — marks a new chapter in the US-China AI rivalry. The contest is no longer just about who builds the most capable models, but who gets to use them, and on what terms.

The U.S. Export Control Order That Started It All

On June 12, 2026, at 5:21 PM Eastern time, U.S. Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei. The letter, citing national security authorities, ordered Anthropic to suspend all access to its two most advanced AI models — Fable 5 and Mythos 5 — for any foreign national, anywhere in the world.

The stated concern was that the models' built-in safeguards could be bypassed through jailbreaking, potentially allowing military intelligence users in China, Russia, or other countries of concern to exploit the technology. Anthropic publicly disagreed, saying the government had only provided "verbal evidence of a potential narrow, non-universal jailbreak."

Because Anthropic had no way to verify the citizenship of every user in real time, the company concluded that the only path to compliance was to disable both models for all customers globally. The move was described in multiple reports as the first export control action targeting access to an advanced AI model itself, rather than the hardware or infrastructure around it.

Z.ai's GLM-5.2: A Launch Hours After the Ban

Within 24 hours of the Anthropic announcement, Z.ai (formerly Zhipu AI, a Beijing company that grew out of Tsinghua University) made its own move. On June 13, 2026, Z.ai announced GLM-5.2, its new flagship model, available immediately to all subscribers of its GLM Coding Plan.

Key specifications of GLM-5.2, documented in Z.ai's official release notes from June 16, 2026:

• 1-million-token lossless context window, which Z.ai says improves long-horizon task capability and reduces context drift and goal forgetting.
• Two thinking-effort levels: High mode for standard tasks, Max mode for complex coding.
• 131,072 max output tokens per response.
• MIT-licensed open weights, with full open-source release scheduled for the following week.
• Z.ai claims the model achieves "open-source SOTA performance on coding and long-horizon task benchmarks."

Notably, Z.ai shipped GLM-5.2 without publishing benchmark scores at launch. Third-party evaluations were pending, and developers immediately noted the absence of numbers on standard tests like SWE-bench Verified or HumanEval.

The timing was unmistakable. Z.ai's founder Tang Jie explicitly referenced the Anthropic ban in GLM-5.2's announcement, stating: "We deeply regret the sudden restrictions on specific frontier models. The sudden blocking of access to frontier models for non-technical reasons has strengthened our conviction that science should be global."

The Musk–Tang Exchange: A Timeline Measured in Months

On June 18, 2026, AI commentator @teortaxesTex posted an estimate on X: that Z.ai's GLM-5.2 was performing at roughly the level of Claude Opus 4.7–4.8, and that a Chinese "Mythos-class" or Fable-equivalent model could arrive by November or December 2026.

Elon Musk replied: "Probably Q1" of 2027.

Z.ai co-founder Tang Jie directly countered Musk, posting that China "won't take that long." Musk then responded: "In terms of benchmarks, that would be impressive. But if measured by true practicality, even by Q1, it would be very impressive."

The significance lies in what the two sides agreed on. Both Musk and Tang were arguing over a timeline measured in months, not years. The debate was not about whether China would close the gap, but exactly when.

Why This Sequence Matters

Four takeaways:

1. Export controls have moved to model access. The U.S. directive targeted foreign access to Anthropic's most advanced AI models, not just hardware. The legal mechanism was a "deemed export" rule, treating display of controlled technology to a foreign national within the U.S. as an export.

2. China's response was immediate and public. Z.ai's GLM-5.2 launch, with its framing as an open-weight, MIT-licensed alternative to the now-restricted Anthropic models, was explicitly positioned as a countermove.

3. The gap is narrowing. The Musk–Tang exchange centered on whether China would reach Fable-class capability by late 2026 or Q1 2027. This is a debate over months, suggesting parity is on the horizon.

4. Open-source is becoming a geopolitical fault line. Z.ai emphasized GLM-5.2's open-source and open-weight availability at the same time U.S. restrictions were limiting foreign access to Anthropic's leading models. The contrast could not be starker.

The chain of events — an export ban, a rapid Chinese model launch, and a public timeline debate between two of the most prominent figures in AI — marks a new chapter in the US-China AI rivalry. The contest is no longer just about who builds the most capable models, but who gets to use them, and on what terms.