Generative AI & Models
New models, releases and capabilities from the frontier labs and the open-source community.
-
Amazon Bedrock Adds Marengo 3.0 for Video, Image, and Audio Search
Amazon announced the general availability of TwelveLabs Marengo Embed 3.0 as an embedding model in its Bedrock Knowledge Bases, enabling natural‑language search across video, audio, and image assets. The new multimodal…
1 source primary sourceAWS Machine Learning Blog -
YuE2-3B Music Model Sets New Benchmark with Editable Scores
YuE2-3B is a newly released open‑source music generation model that outperforms the commercial Suno v5 on the WildSongBench benchmark, achieving a SongBench average of 6.9632 versus Suno’s 6.8721. The model supports…
1 source primary sourcehuggingface.co -
Nex-AGI Launches 1.6‑Trillion‑Parameter Agentic Model Nex‑N2.5‑Max
Nex‑AGI has unveiled the Nex‑N2.5 family, a new line of agentic models designed for long‑horizon tasks in real‑world settings. The lineup includes mini, Pro, and Max variants, with the Max model built on a…
1 source primary sourcehuggingface.co -
NInfer Studio: Desktop App for Local LLM Inference with GPU Presets
NInfer Studio is a new cross‑platform desktop shell built with Tauri 2 that turns the ninfer‑serve inference engine into a user‑friendly application. It offers per‑GPU presets, Hugging Face model downloads, live token…
1 source primary sourcegithub.com -
GPN‑Star Model Predicts Genome‑Wide Functional Constraints
GPN‑Star is a new genomic language model that predicts functional constraints across the human genome by leveraging multispecies sequence alignments. Developed by a team including Ye, Benegas, Albors, Li, and Song, the…
1 source primary sourceNature Machine Learning -
Model-Agnostic PII Detector for LLMs
A new model-agnostic detector for personally identifiable information (PII) has been released, designed to run on any large language model (LLM) managed on Amazon Bedrock. The detector, evaluated on five public PII…
1 source primary sourceAWS Machine Learning Blog -
OpenAI Unveils GPT‑6 Astra: Record‑Breaking 3D Rendering, Loop‑Transformer Architecture
OpenAI’s new GPT‑6 Astra was released last week, quickly becoming the most powerful LLM in the author’s hands. It outperforms its GPT‑5.6 predecessor across the board, but its biggest leap is in 3D rendering and…
8 sources primary source HN 512The DecoderAhead of AI (Sebastian Raschka)OpenAILast Week in AI +2 more -
Clinically Oriented AI Model for Intraoperative Pathology
Researchers have developed CRISP, a foundation model designed to support intraoperative pathology in precision surgery. CRISP was trained on over 100,000 frozen sections from ten medical centers and evaluated on nearly…
1 source primary sourceNature Machine Learning -
New Open Model Releases and Licenses
In the world of artificial intelligence, several new model releases and their associated licenses have been announced. Motif-3, a model from Motif-Technologies, is now available under an MIT license, showcasing…
1 sourceInterconnects -
AI Models May Add Watermarks to Text Outputs
On August 11, Anthropic announced that all future Claude models will include a watermark in their text outputs, identifying them as AI-generated. Google and OpenAI also use text watermarks. The EU AI Act mandates text…
1 sourceIEEE Spectrum AI -
Deepseek V4.1-Flash Reduces AI Agent Memory Usage
Deepseek has released its new AI model V4.1-Flash, which significantly reduces the memory requirements for AI agents. This model achieves this by shrinking the buffer that agents need for processing long texts. The…
1 sourceThe Decoder -
Claude Fable 5.1 shows longer, less hedged responses than Fable 5
Anthropic’s latest Claude model, Fable 5.1, has shifted its writing style compared with the earlier Fable 5. An analysis by Arena.ai of tens of thousands of high‑reasoning Text Arena outputs found the new version uses…
1 sourceThe Decoder -
OpenAI launches GPT‑Live‑1 API enabling simultaneous speech and listening
OpenAI has opened its new GPT‑Live‑1 speech model to developers via an API that can both listen and speak at the same time, a capability called full‑duplex. The model, already integrated into ChatGPT, lets developers…
2 sources primary source HN 46The DecoderOpenAI -
Animated Map Shows Mercator vs Equal Earth Projections
A new interactive animated map, created by GPT-6 Astra, visually demonstrates the differences between Mercator and Equal Earth map projections. Users can slide a bar or play a button to see the transition between these…
1 sourceSimon Willison -
OpenAI Expands ChatGPT Images with 2.5 Model
OpenAI has released ChatGPT Images 2.5, a new state-of-the-art image model that offers improved quality, faster generation times, and enhanced editing capabilities. Key features include more natural lighting, richer…
2 sources primary source HN 383Simon WillisonOpenAI -
GPT-6 Astra Generates Fabergé Egg, Codex Builds Blender Model
Simon Willison demonstrates the creative potential of OpenAI’s latest tools by generating a Fabergé‑egg image themed after the TV show Pluribus using ChatGPT Images 2.5. He then feeds the image into Codex running GPT‑6…
1 sourceSimon Willison -
Google launches WeatherNext 3, adding satellite data for hourly AI weather forecasts
Google has unveiled version 3 of its WeatherNext AI weather model, the latest upgrade in its push to make machine‑learning forecasts as accurate as traditional physics‑based systems while using far less compute. The…
2 sourcesThe Guardian AIArs Technica AI -
3.8B LLM Trained to 0.384 CORE Score for Under $1,000 Using Consumer GPUs
A solo researcher demonstrated that a 3.8 billion‑parameter language model can reach a 0.384 CORE benchmark score after processing 65 billion tokens, all for just $998 in cloud compute. The training ran on a mix of a…
1 source HN 112hugovergnes.github.io -
Claude Platform: Cost Reduction and Performance Improvement
Anthropic Engineering's Claude Platform has introduced new features to reduce costs and maintain or improve application performance. The key fixes include maximizing prompt cache hit rates, removing anti-patterns in…
1 source primary sourceAnthropic Engineering -
IBM Releases New Time Series Forecasting Model
IBM has released a new version of its Granite Time Series PatchTST-FM-r2 model, which combines an updated architecture, larger pretraining corpus, and probabilistic forecasting capabilities. This model is now the top…
1 source primary sourceHugging Face -
RTK Token Savings Debunked: Cost Benchmarks Disagree
RTK, a popular tool for compressing terminal output, was promoted as a way to reduce AI coding costs. One post claimed it could cut Claude Code tokens by up to 60%. However, JetBrains’s SkillsBench found no savings.…
1 source HN 51quesma.com -
Cognition's SWE-2 Breaks Terminal-Bench 2.1 with 92.8%
Cognition's SWE-2 model has achieved a remarkable 92.8% on the Terminal-Bench 2.1 benchmark, a significant improvement over the 73.0% of its predecessor, DeepSWE 1.1. The model, which has 2.8 trillion total parameters…
1 source HN 61tokenstead.ai -
Microsoft Releases Record 974 Patches, Including 2 Zero-Days
Microsoft released a record 974 patches across its products, including two previously exploited zero-day vulnerabilities. The first, CVE-2026-85880, allows local attackers to escalate privileges on Windows systems. The…
2 sourcessecurityweek.comArs Technica AI -
Suno launches v6 AI music model with record‑industry licensed data
Suno unveiled its v6 AI music model, the first built on a dataset that includes licensed recordings from Warner Music Group, BMG and Believe alongside user‑generated content. The company says the new training set…
1 sourceThe Verge AI -
Universal Music Launches AI Music Platform with ElevenLabs
Universal Music Group is launching a new AI-powered platform that allows users to remix, mashup, and create new versions of licensed tracks from its catalog. The platform, developed through a multiyear agreement with…
1 sourceThe Verge AI -
Slack Enhances Chat with AI-Generated Surfaces
Slack is introducing a new feature called Slackforce Surfaces that allows users to build interactive reports and tools directly within chats. This feature, which uses AI to gather information from relevant…
1 sourceThe Verge AI -
OpenAI Temporarily Pauses Pro Plan Due to Astra Demand
OpenAI, the creators of ChatGPT and Codex, has paused subscriptions for its $200-per-month Pro plan due to unprecedented demand for their newest and most powerful model, Astra. OpenAI's product leader, Thibault (Tibo)…
1 source HN 7TechCrunch AI -
GPT-5.6 Sol Streamlines Quantum Computing Experiments
OpenAI's GPT-5.6 Sol has been successfully integrated into MIT's quantum computing experiments. Beatriz Yankelevich, a graduate student, used GPT-5.6 Sol to automate routine measurements on superconducting qubit chips.…
1 source primary source HN 148OpenAI -
Claude No Longer Available to Minors
Claude, a consumer product, is now restricted to users over 18 years old. Users must verify their age through Yoti, a third-party age verification platform. If detected under 18, accounts will be disabled. Users have…
1 source HN 64support.claude.com -
Demystifying Anthropic's J-Space: A Mathematical Primer
In a recent article, Anthropic researchers introduced the J-space, a mathematical model inspired by the global workspace in the human cortex. This framework allows LLMs to be audited for alignment purposes. The authors…
1 sourceTowards Data Science -
GPT-6 Pro Tops ChessBench with 2,340 Elo Rating
ChessBench, a platform for benchmarking AI models, has released new results. GPT-6 Pro (Max) from OpenAI has been added to the platform with a benchmark Elo rating of 2,340, placing it at rank 10. This is the highest…
1 sourcechessbench-ai.github.io -
Datasette Security Patches Released
On September 11, 2026, Datasette, a database server, released two security patches: 1.0a39 and 0.65.4. These updates address vulnerabilities found through an extensive security audit conducted by Claude Fable 5.1,…
1 sourceSimon Willison
.jpg)
