AMD Quantizes Qwen3.8-Flash-Next for Ryzen AI Halo
AMD introduces UINT4 quantization for Qwen3.8-Flash-Next, reducing checkpoint storage while testing accuracy tradeoffs.
735 articles
New AI models as they ship — large language models, multimodal systems, and open-weight releases. Capabilities, context windows, pricing, and availability, summarized so you can compare what is actually new.
AMD introduces UINT4 quantization for Qwen3.8-Flash-Next, reducing checkpoint storage while testing accuracy tradeoffs.
Salesforce launches SOV-Transcribe-1, an open-weight speech model with 4 billion active parameters, designed for non-streaming automatic speech recognition (ASR) and enterprise applications.
Google AI Edge Gallery now includes on-device ASR and TTS capabilities, enabling offline, multilingual voice interaction across platforms.
Google DeepMind introduces Gemini 4 Hydrogen, a model optimized for rapid execution in coding and agentic workflows, with pricing and availability across multiple platforms.
Hugging Face introduces ZONOS3-Mini, an open-source Text-To-Speech model with real-time voice cloning capabilities and 100M active parameters, available for use across multiple platforms.
Google DeepMind announced two new AI models: Gemini Robotics 2.5 for advanced robotics and AlphaGenome 2 for gene prediction, expanding their capabilities in both fields.
DeepL's new Aure model enhances real-time translation by preserving tone, emotion, and speaking pace across languages, supporting eight languages at launch.
Radical Numerics introduces Omnii Defense, an AI model designed to detect natural and AI-generated pathogens using a laptop and USB gene sequencer, without internet access.
AMD outlines the GLM-5.3-Flash model's architecture and performance on the MI300X server, focusing on parallelism strategies and KV cache management for high-concurrency workloads.
AWS announced new AI models and tools in October 2026, offering improved performance, cost efficiency, and flexibility for developers building AI agents.
NVIDIA has released a synthetic dataset designed to help developers evaluate privacy protection in AI applications processing Japanese business documents.
DeepMind's new model improves prediction of how DNA variations affect gene expression in specific human cell types, aiding disease research.