Follow the latest coverage, related explainers and connected technology stories.
CISA, NSA, and the FBI said that six Chinese artificial intelligence companies carried out extensive operations to extract outputs from advanced U.S. models through billions of tokens and millions of requests since late 2024. The agencies recommend improving monitoring of API abuse and sharing indicators of these campaigns among companies.
Abliteration.ai provides access to modified versions of open-weight models after removing refusal mechanisms, claiming this enables cybersecurity teams to simulate attackers’ behavior. At the same time, however, the service opens a wider door to using these models for harmful activities, amid the absence of a mechanism to verify customers’ identities and incomplete accountability controls.
Tencent announced the release of the open-source Hy4 Preview model, with a context window exceeding one million tokens and the ability to activate 49 billion parameters per operation. The company says the model outperformed GLM-5.3 and Kimi K3 in an internal blind test, while the results remain limited to Tencent’s evaluation and the model is an initial release.
Z.ai confirmed that the mysterious Ox Alpha model belongs to the company and released it under the name GLM-5.3-Flash, making its weights available through Hugging Face. The model targets programming and long-term agentic tasks and is based on 320 billion parameters, with around 18 billion activated per query and a context window of up to one million tokens.
Parviz Sayed Mohammed believes that Z.ai's GLM-5.3-Flash model requires rethinking the allocation of AI workloads, thanks to its low cost and open weights. He proposes dividing tasks into three spending tiers, directing about 45% of task volume to the low-cost model after testing it in practice.
z.ai has made its GLM-5.3 model available through an API at the same price as the GLM-5.2 model, with a lower cost for cached inputs. This gives developers a new option for experimenting with coding tasks and software agents, while the timing of the model weights release and licensing remains unspecified.
Chinese AI startup Z.ai has launched the GLM-5.3 model, which relies on scaling post-training for the GLM-5.2 model rather than building a new base model. The company says its cyber capabilities detected a potentially serious vulnerability in Cursor, while initial availability will be limited to the GLM Coding Plan and the ZCode environment until safety evaluations are completed.
Writer announced its new Palmyra X6 model alongside a redesigned orchestration layer and governance tools to monitor agents’ token consumption and spending. The company says that combining the model and platform reduces costs by an average of 52%, while improving speed by 48% and quality by 10%.