News reporting focused on AI and machine learning, covering the companies behind these technologies, their real-world applications, and the ethical concerns they raise. This includes areas like generative AI (large language models, text-to-image and video), speech tech, and predictive analytics.
NVIDIA's Vera Rubin platform is ramping up with 300 global partners, delivering 10x more tokens per megawatt than Grace Blackwell NVL72 in benchmarks. CoreWeave, Google Cloud, Microsoft Azure, and Mistral are among early adopters. The platform's extreme co-design across seven chips and five rack trays achieves the highest performance per watt and lowest token cost for AI factories.
.jpg)
Nvidia is promoting its new Vera Rubin chip system, which combines CPUs and GPUs into a single platform. The company is positioning itself as a supplier of complete AI systems, not just GPUs. New benchmarks show significant performance and efficiency gains, with early customer OpenAI already using a rack.
Researchers at Google DeepMind have developed GenCeption, a model that repurposes a pre-trained video generator for classic computer vision tasks like depth estimation and segmentation. The system builds on an open-source video model from Alibaba and delivers results in a single forward pass, guided by text prompts. It trained on just a small set of synthetic videos and needs far less data than competing approaches. In benchmarks, GenCeption matches established specialized models and transfers its abilities to real-world footage and untrained categories like animals. The authors say this supports the contested idea that video generators can serve as the basis for universal world models in computer vision.
A new benchmark pits Claude Fable 5 against GPT-5.6 Sol on an unpublished NP-hard optimization problem. Fable 5 produced the best overall solution and showed remarkable consistency. The native /goal mode helped in most trials but made average performance worse for both models, revealing that persistence features can amplify both good and bad decisions.
Kimi has released K3, a multimodal open-weight model with 2.8 trillion parameters and a one million token context window. In benchmarks, it approaches Claude Fable 5 and GPT-5.6 Sol but trails them slightly. Priced at $3 per million input tokens and $15 per million output tokens, K3 signals an end to ultra-cheap Chinese AI models, though it remains cheaper than top Western competitors.
Cohere has launched Cohere Transcribe Arabic, an open-source 2-billion-parameter ASR model designed for Arabic speech recognition. The company claims it is the most accurate open-source Arabic speech-to-text system, outperforming Whisper Large V3 and the standard Cohere Transcribe model in benchmarks. The model handles dialect variety, code-switching, bilingual conversations, and specialized vocabulary.
Perplexity introduces 'Search as Code', a new architecture where AI models write their own search workflows as Python code instead of calling fixed APIs. The approach reduces token usage by 85% in a cybersecurity benchmark and beats rivals on four out of five tests. The system is rolling out in Perplexity Computer and the Agent API.
Phoronix benchmark results show NVIDIA's Vera CPU, powered by custom Olympus cores, delivers strong performance in agentic AI workloads. The 88-core processor achieved up to 1.5x overall performance advantage over a latest-generation 128-core x86 chip, with 1.2TB/s memory bandwidth and 90% sustained peak bandwidth. Vera compiled a Linux kernel in 20 seconds and showed a 1.6x geometric mean increase over the Grace CPU. Memory efficiency is notable with under 30W memory power versus over 100W for DDR5.
Microsoft launched MDASH, a security system with more than 100 specialized AI agents that detected 16 new Windows vulnerabilities, including four critical ones. The tool uses a four-stage process to analyze code and debate findings, achieving 88.45 percent on the CyberGym benchmark. Backed by experts from a DARPA challenge winner, MDASH is in limited preview.
Alibaba's Qwen-Image-2.0 model achieves twice the image compression and reduces generation steps from 40 to 4 through key technical advances. It features a more efficient VAE, updated transformer architecture, and a prompt expansion module. The model excels in benchmarks and handles complex outputs like posters with text.
OpenAI has updated ChatGPT to use GPT-5.5 Instant as the default model, cutting hallucinations by 52.5% on topics like medicine, law, and finance. The model shows gains across benchmarks in math, science, and visual reasoning. A new memory sources feature lets users see and edit the personal context used in responses.
OpenAI has launched GPT-5.5 Instant, replacing GPT-5.3 Instant as the default model for ChatGPT. The update cuts down hallucinations in fields like law, medicine, and finance while keeping low latency. It also boosts performance on math and multimodal benchmarks, adds context tools, and shows memory sources to users.