Google DeepMind released DiffusionGemma, a text diffusion model built by retrofitting Gemma 4 with less than 10% of the original training token budget, achieving 1,500 tokens per second.
Backflip AI's new tool transforms 3D scans into editable CAD models in minutes, addressing a critical gap in digital manufacturing.
Grok Imagine Image 2.0 is now generally available as Quality Mode on grok.com/imagine and iOS/Android apps, ranking second in text-to-image and image editing benchmarks.
OpenAI says Astra, an upcoming model, may have critical cybersecurity capabilities, based on recent evaluations.
Erik Kaunismäki trained a static embedding model, lattice-retrieval, scoring 0.4749 NDCG on decontaminated BEIR after fine-tuning, outperforming sentence-transformers/static-retrieval-mrl-en-v1.
Bytedance is training an AI model with up to ten trillion parameters, three times the size of Moonshot's Kimi K3.
Hugging Face's Ideogram-4 inpainting model, released in August 2026, offers fine-grained control through JSON bounding boxes and differential diffusion, according to the blog post.
OpenAI announced it is removing text-based chat limits for all ChatGPT users, including free users, as the platform recently surpassed 1 billion weekly users.
OpenAI announced improvements to GPT-5.6 Sol in ChatGPT, enhancing reliability and focusing answers, with free users gaining access to Luna for unlimited text chats.
NVIDIA announced the release of Cosmos 3, an open-world model family for physical AI, achieving top rankings in benchmark evaluations.
Alibaba's Qwen3.8 Max achieved a score of 56 on the Artificial Analysis Intelligence Index, surpassing GLM-5.2 but falling short of Kimi K3's 57.
DeepMind's WeatherNext model improves cyclone forecasts by 20% in 48-hour predictions, outperforming existing systems.