The partial rollback comes as Chinese developers tout cheaper open-weight AI models that researchers and investors say are rapidly closing the gap with top U.S. labs.
The White House allowed Anthropic on Friday to resume limited access to its Mythos 5 model for some companies and federal agencies after a two-week shutdown tied to an export control directive, while Fable 5 remains restricted. OpenAI also said it would limit the rollout of its GPT 5.6 models following a government request. The moves have intensified debate over whether U.S. national security restrictions on leading domestic AI systems could give Chinese rivals more room to catch up. That concern has sharpened around Zhipu’s open-weight GLM 5.2 model, released earlier this month. Researchers cited in the report said it can perform on par with top U.S. labs on some cyber benchmarks, with some commentary suggesting it can match Anthropic’s Mythos capabilities in selected areas. Jefferies strategist Christopher Wood wrote, citing industry sources, that GLM 5.2 is "almost equal to Anthropic as a competitor for the corporate market" and costs about one quarter as much per token. The competitive pressure is landing as corporate buyers shift away from "tokenmaxxing" toward efficiency and return on investment. Companies are increasingly drawn to cheaper open-weight models that can be downloaded and run on their own servers, broadening access to Chinese systems such as GLM 5.2, Moonshot AI’s Kimi K2.7, Alibaba’s Qwen 3 and DeepSeek’s models. Industry executives say the trend could accelerate adoption while also heightening cybersecurity concerns as increasingly capable open-weight systems become easier to use.