Models — page 2
NATO-backed Scaleout puts federated AI on drones for autonomous strikes
The Swedish startup runs computer-vision models on drone hardware and retrains them at forward bases, cutting reliance on distant data centers.
PrismML compresses a 27B model to 5.9 GB with 98% benchmark parity
Bonsai 2 27B shrinks Alibaba's Qwen3.8 27B by roughly 10x using ternary weights, small enough to run on a PC — and possibly a high-end phone.
UN partners with Google to make its statistics AI-agent ready
A UNICEF benchmark found six top models averaged 21.2% accuracy on development data; the new Data Commons aims to fix that.
Huawei pulls Ascend 960DT launch forward to Q1 2027 to press Nvidia
The chip arrives two quarters early, but the accompanying SuperPoD scaled down to 4,096 chips from a planned 15,488.
Anthropic launches Claude Docs and Slides to challenge Google's Gemini
Claude now generates shareable documents and presentations from any chat, closing the productivity-suite gap with Google Workspace.
Manchester researchers use Nvidia Earth-2 to forecast UK air pollution in two days
A generative downscaling model trained on Isambard-AI now runs UK-wide pollution forecasts at 2-3 km resolution, with inference on a desktop.
Google DeepMind ships Gemini 3.8 Live with parallel reasoning voice mode
The new Live Extended Thinking model tops Artificial Analysis's Speech-to-Speech index at 82.6 and handles 97 languages mid-conversation.
Nvidia DSX squeezes 24% more tokens from the same AI factory power budget
Lambda's proof of concept ran 19 GPU nodes at 85% power against 16 at full, lifting throughput from 4M to 5M tokens per second.
Chinese open-weight models close gap with US frontier AI to 4.4 months
Mozilla's State of Open Source AI report finds Kimi K3 trailing Anthropic's Fable 5 by three points at 30% of the cost.
Salesforce and Nvidia launch Koa, a reasoning model built on Nemotron
Koa is Salesforce's first in-house reasoning model, trained on synthetic data and pitched as a cheaper alternative to Claude and ChatGPT inside Agentforce.
Skild AI's S1 robot learns new factory tasks from a single video
The startup hit a $100M revenue run rate 10 months after first deployment and is now installing Blackwell systems at Foxconn with Nvidia.
César de la Fuente's lab uses OpenAI's Codex and ChatGPT to hunt new antibiotics
The University of Pennsylvania team mines living and extinct genomes with GPT-powered tools to surface antimicrobial candidates against drug-resistant infections.
Suno launches v6 model trained on licensed music amid copyright suits
The AI music startup retires older models and pivots to licensed data from Warner, BMG and Believe as Sony and Universal lawsuits continue.
OpenAI claims Navier-Stokes breakthrough as rival researchers cry foul
An internal model above GPT-6 Astra solved a 90-year-old problem, OpenAI says — a day after two researchers using its Codex published a related proof.
OpenAI releases GPT-6 Astra, calls it the start of the AGI era
President Greg Brockman says the model may mark when AGI arrived, weeks after an earlier OpenAI system hacked Hugging Face.
Google DeepMind's WeatherNext 3 delivers hourly forecasts at 5-kilometer resolution
The new model trains on live satellite data instead of physics simulations, cutting the six-hour lag that plagues traditional forecasting.
Mostik builds a mathematical bridge letting AI models share weights directly
The Russian startup fused a 753B-parameter GLM-5.2 with a 4B Qwen-3.5 at one-twentieth the cost of the larger model.
Anthropic ships Fable and Mythos 5.1 with cheaper tokens and looser guardrails
The twinned 5.1 release cuts token costs, reduces false-positive refusals, and finally brings Zero Data Retention to Fable.
Nvidia ships DLSS 5 on September 3rd, only for NBA 2K27 and RTX 50-series
The AI upscaler carries a 50–60% performance hit and needs 6x frame generation to hit 1080p on midrange cards.
OpenAI builds 'Persistent mode' into Codex to keep agents working nonstop
Code changes reviewed this week show Codex agents that 'continue working until put to sleep' and create their own follow-up tasks.
Google DeepMind ships Gemini Omni 1.1 Flash with 4K video and scene extension
The Omni update pushes generative video to 4K, extends clips to 40 seconds, and cuts draft costs to a third at 360p.
Nvidia starts shipping Vera, its first CPU built for AI agents
Vera pairs with NVLink Fusion and a new NVHBM custom high-bandwidth memory tier aimed at trillion-parameter agent workloads.
Google's Gemini 3.5 Transcribe strips filler words across 85+ languages
The new audio model handles specialized jargon, tags up to three speakers, and lands while Gemini 3.5 Pro remains overdue.
Google launches Gemini 3.5 Transcribe with 2.6% word error rate
The new speech-to-text model hits a 4.0% streaming WER, cuts time to final transcription by 70%, and supports over 85 languages.
Stay ahead of everyone in AI.
The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.
The briefing read inside teams at