Updated daily175 verified stories
New frontier, open-weight, multimodal, reasoning, and specialist AI models—plus the evidence behind their capability claims.
Every story links to the strongest available evidence: original sources when available, otherwise clearly attributed reporting.
What happened, why it matters, and what to watch — no jargon tax.
When the signal is thin, we publish nothing rather than padding the feed.
A growing stream of verified perspectives for people who need to understand AI without chasing hype.
IBM’s Granite 4.2 family includes 3B, 8B and 30B open-weight models with a native 128,000-token context window. Ars Technica reports that the two larger versions received specialized training for tool use and agentic tasks.
A new arXiv paper proposes PuzzleKV, a training- and calibration-free method that independently compresses completed pages of large language model KV caches. It reports more than 96% of Full KV performance at about 60% of original storage, and more than 93% at 18.7% of original storage with quantization.
A new arXiv paper reports that a multi-armed bandit controller selected prompting strategies for LLM essay scoring with comparable accuracy to exhaustive search while reducing calls by 78.4%.
Seoul Economic Daily reports that Alibaba, Nvidia, OpenAI and Anthropic are increasing outreach to South Korean startups through developer events, AI tools and Claude credits, seeking to anchor their ecosystems in the country’s young technology companies.
TechCrunch reports that Bloomberg identified China’s Z.ai as the lab behind the anonymously released Ox Alpha open-weight model. Z.ai reportedly confirmed the model belongs to its GLM series and said its weights would be released Wednesday.
A new arXiv preprint proposes giving large language models compact frequency-domain evidence, alongside indexed observations, to help detect anomalies in recurring time-series patterns.
An evaluation of 12 open-weight language models finds that they frequently mistake indirect or reversed relationships for direct causal connections, while expressing high confidence in many incorrect judgments.
A new arXiv preprint proposes Flexible Self-Consistency, a method that estimates how uncertain a language model is about a question and adjusts the number of sampled reasoning paths. The authors report token savings of up to 76% while maintaining accuracy comparable to standard self-consistency across various…
A new preprint describes FireRedAudio, a 9-billion-parameter audio language model that combines audio understanding, multilingual speech recognition, speech synthesis and speech editing through separate continuous representations.
The New Stack reports that Shopify CEO Tobi Lütke threatened to ban Claude Code because it does not natively read AGENTS.md, creating extra synchronization work for teams using multiple coding agents.
A new arXiv preprint reports that quantizing large language models can affect Bangla understanding unevenly: GPT-OSS lost up to 57.35% accuracy on reasoning-heavy tasks in one format, while Qwen and LLaMA were generally more stable.
Monthly Mixing reports that Adobe has launched three Firefly audio-generation tools: Generate Music, Generate Speech and Generate Sound Effects, alongside new Firefly AI Assistant features and support for Google’s Gemini Omni Flash.
One useful briefing each week
Get the week’s verified AI news, original data, useful tools, learning picks, and fresh AI jobs.
Hiring an AI professional or launching a useful AI product? Put it in front of people who came here to learn and act.
Post an AI job Submit an AI tool