Approved government security teams can now use GPT-5.6 Cyber through Daybreak Red in AWS GovCloud (US-East), keeping vulnerability research and security testing inside the AWS environment they already use for critical missions. The model can investigate software weaknesses, reproduce vulnerabilities in controlled settings, and assess fixes, while qualified practitioners review its findings before implementation. Access requires Daybreak Red approval and AWS provisioning; inference data is not used to train models.
Phonon lets developers ship text-to-speech that runs offline on phones, laptops, single-board computers, and browser tabs, keeping user text on the device instead of sending it to a server. The models are about 65 MB and 75 MB, cover English, French, German, Spanish, and Portuguese, and include more than 100 voices. Phonon 31m is Apache 2.0 for commercial use, while the five-language Phonon 40m is non-commercial.
Nous Research’s Arete-27B gives developers an open 27B-parameter model tuned to change how Qwen3.8-27B handles politically sensitive topics, without changing its architecture. The team used supervised fine-tuning across seven value categories, raising its fact score from 2.32 to 2.98 out of 5 and reducing answers containing propaganda from 22% to 0.9%.
Qwen3.5-4B-3D-Coder lets developers generate interactive Three.js scenes from text, and its Q4_K_M and Q8_0 versions can also use a reference image, making small local models more useful for rapid 3D prototyping. The independent fine-tune runs with llama.cpp; attach the image, add a prompt such as “create this scene using voxels 3d,” and it returns scene code. The preview was tested with a 64K context, while its optional vision projector adds about 644 MiB.