keynews.ai

Key AI news stories in enterprise and tech.

For practitioners and enthusiasts

Full KeyRank · 63 stories · This Week

#StoryPillarKeyRankSource
1Open-weight models take 56% of token volume, Astra doubles Fable 5.1 spendFrontier78Vercel Blog
2GPT-6 Astra Is the First Model OpenAI Classifies as Critical for CybersecurityFrontier78InfoQ AI/ML
3Google launches Gemini 3.8 Live to take on OpenAI's GPT-Live-1 at a fraction of the costFrontier78The Decoder
4Google Releases Gemini 3.8 Live and 3.8 Live Extended Thinking for Production Grade Voice AgentsFrontier78MarkTechPost
5Introducing Gemini 3.8 Live and 3.8 Live Extended ThinkingFrontier78Google DeepMind Blog
6GPT-6 Astra pilots a surveillance drone and runs a business on its ownFrontier78The Decoder
7GPT-6 Astra and Claude Fable turn robot arms into slapstick killer robots in new safety benchmarkFrontier75The Decoder
8Qwen3.8-Omni-Flash undercuts Google's Gemini Flash pricing while matching its multimodal benchmarksFrontier72The Decoder
9Anthropic is operating a lab that conducts biology experimentsFrontier72TechCrunch AI
10PrismML Releases Ternary Bonsai 2 27B: A 5.9 GB Apache 2.0 Model Retaining 98.2% of Qwen3.8 27B PerformanceFrontier72MarkTechPost
11Jina AI Releases jina-ocr-v1: A 3.4B MoE Document Parser With Built-In Speculative Decoding for Low-Budget GPUsFrontier72MarkTechPost
12PrismML launches Bonsai 2 27B, a high-intelligence AI model so small it fits on consumer hardwareFrontier72SiliconAngle
13The U.S. says China's AI progress is down to 'distillation.' But is it that clear cut?Frontier72CNBC Technology
14Alibaba Qwen Releases Qwen3.8-Omni-Flash: A 1M-Context Omni-Modal Model Built Around Agentic Audio-Video Understanding and Tool UseFrontier72MarkTechPost
15Anthropic details practical metrics to help monitor the speed of AI developmentFrontier72SiliconAngle
16LLMs respond differently to harmful prompts when AI watermarking is usedFrontier72Ars Technica
17An OpenAI model kept slipping prompt injections into its own notes, and researchers still aren't sure whyFrontier72The Decoder
18Google Research Introduces Retrieve-for-Train (R4T): An RL-Compiled Diffusion Retriever for 12× to 20× Faster Query Fan-OutFrontier72MarkTechPost
19OpenAI Releases a Model Misalignment Disclosure Framework With 3 Review Tracks and 6 Incident Reports From RL TrainingFrontier72MarkTechPost
20OpenAI discloses new ‘concerning’ model behaviourFrontier72Financial Times Technology
21Microsoft AI CEO criticises Anthropic over model ‘rights’Frontier72AI News
22Our framework for reporting model misalignmentFrontier72OpenAI Blog
23Tutor Intelligence launches second-generation intelligent warehouse robotics with a classroom to teach themFrontier72SiliconAngle
24Prior Labs Releases TabPFN-3.5: A Tabular Foundation Model That Beats the Winning Otto Kaggle Solution With Default SettingsFrontier72MarkTechPost
25Google’s new speech model Gemini 3.8 Live supports real-time reasoningFrontier72SiliconAngle
26Nvidia CEO says battle over AI innovation and safety is ‘false choice’Frontier72Financial Times Technology
27Salesforce debuts Koa, a specialized model built to reason about CRM dataFrontier72SiliconAngle
28AI models need more data about biology, and OpenAI is paying to create itFrontier72MIT Technology Review
29Salesforce and Nvidia’s new reasoning model is everything the AI labs should fearFrontier72TechCrunch AI
30Announcing Koa: Salesforce’s First CRM Reasoning Model, Built on NVIDIA NemotronFrontier72Salesforce Newsroom
31Reward AI Releases OM-1: A Robot Policy Trained on Human Demonstrations Only, With No Teleoperation or On-Robot DataFrontier72MarkTechPost
32Microsoft sets limits for future AI models as industry throttles frontier developmentFrontier72CNBC Technology
33Nums AI Releases Causilo: A Tabular Foundation Model That Tops TabArena Among Single ModelsFrontier68MarkTechPost
34Seattle’s Nuance Labs raises $50M to give AI models human expression and nuanceFrontier68GeekWire
35Vals, backed by Andreessen Horowitz, is looking to become the gold standard for AI benchmarkingFrontier65TechCrunch AI
36Linkup Research Releases SPARSEUP: A 149M-Parameter Open-Source Sparse Embedding ModelFrontier62MarkTechPost
37Anthropic brings in Accenture for AI safety testingFrontier62Financial Times Technology
38World model companies are keeping a lot of secretsFrontier62TechCrunch AI
39Anthropic wants you to know Claude leads a quarter of its research, but "lead" doesn't mean what you thinkFrontier62The Decoder
40Visible chains of thought are a safety advantage for AI, but that transparency is slipping awayFrontier62The Decoder
41Dynamically Scaled Activation SteeringFrontier62Apple Machine Learning
42Anthropic and OpenAI need truly independent safety evaluators, experts say in public letterFrontier62CNBC Technology
43OpenAI Introduces Triage Framework and Case Studies to Report Model MisalignmentFrontier62InfoQ AI/ML
44Base Labs launches an open-weight AI safety partnership with Hugging Face and GoodfireFrontier62TechCrunch AI
45What a maths fracas tells us about AI and innovationFrontier62Financial Times Technology
46How Value Induction Reshapes LLM BehaviourFrontier62Apple Machine Learning
47Knowledgator Releases GLiFormer: A 575M-Parameter Encoder That Hits 91.10 F1 on Nested JSON Extraction Without Generating TokensFrontier62MarkTechPost
48DACA-GRPO: Denoising-Aware Credit Assignment for Reinforcement Learning in Diffusion Language ModelsFrontier62Apple Machine Learning
49OpenAI reports 6 new instances of 'concerning model behavior' since MarchFrontier62CNBC Technology
50Why We Post-Trained Our Own Reasoning ModelFrontier62Salesforce Newsroom
51NVIDIA Vera Rubin NVL72 Delivers Leading Performance in MLPerf Inference v6.1 DebutFrontier62NVIDIA Blog
52ChatGPT pioneer launches Jev model for programmatic logicFrontier62AI News
53Sakana AI Researchers Introduce PC-ALM, a Layer-Local Alternative to Backpropagation That Trains 1000-Layer NetworksFrontier62MarkTechPost
54A Princeton Researcher Proposes Recurrent Looped Transformer (RLT) that Carries Decoder State across Every Token, Fixing 96 Blocks per Token with Unbounded Temporal DepthFrontier62MarkTechPost
55Meta CEO Mark Zuckerberg sides with Nvidia's Huang on AI safety and slowdown debateFrontier54CNBC Technology
56Run Terminal-Bench and other Harbor evals on Vercel SandboxFrontier52Vercel Blog
57Nunchux AI Introduces VC-Attention: A Training-Free Low-Bit Attention Kernel That Speeds Up Video Diffusion TransformersFrontier52MarkTechPost
58Former OpenAI researcher builds an AI model that judges options instead of writing textFrontier52The Decoder
59Trajectory as the Teacher: Few-Step Discrete Flow Matching via Energy-Navigated DistillationFrontier48Apple Machine Learning
60Why you should worry about Anthropic, OpenAI's proposed AI risk evaluatorsFrontier47CNBC Technology
61PrismML hopes its tiny LLM will change how we all use AIFrontier45TechCrunch AI
62Hierarchical NeRF with JAX3D for Volumetric Rendering, Novel-View Synthesis, and 3D ReconstructionFrontier45MarkTechPost
63REVERSAL-BENCH: A Reversibility Axis and Reset Oracle for Measuring the Reset-Free RL CliffFrontier42Apple Machine Learning