📰
DeepSeek v4.1 Flash outperforms v4 Pro
The new flash model offers superior cost-to-performance, routing Pro calls to Flash.
🤖 AI — 2026-09-09
ddNews · a daily AI briefing for Dirk (generative video/ddStudio, Lit
Adobe Premiere integrates five video models
A new in-timeline tool allows users to switch between competing generative video models without exporting.
Rumors of OpenAI "Bell" (ChatGPT 7)
Reportedly surpasses Astra by utilizing 10,000 agents to solve complex problems.
Instinct AI agent gets email access
The agent can now create accounts and handle business support requests autonomously.
OpenAI claims solution to Navier-Stokes problem
An unreleased model using 10,000 AI agents reportedly solved the Millennium Prize Problem.
Accenture and Google launch Gemini Enterprise Group
A global initiative specifically targeting the scaling of agentic AI.
Cloudera integrates Mistral frontier models
The move aims to provide secure, hybrid data environments for sovereign enterprise AI.
Accenture and Google launch Gemini Enterprise group
A joint initiative to deploy forward-deployed AI engineers to scale agentic AI in corporations.
DeepMind's AlphaGenome Atlas
A petabyte-scale dataset predicting the effects of 9 billion possible human genome mutations.
Suno launches V6 music models
new models trained on licensed catalogues from Warner Music, BMG, and Believe replace all previous versions.
Qualcomm develops custom AI chips for AWS
Focus is on multi-generational custom silicon specifically for AI inference in Amazon's cloud.
Samsung and Mistral partner on on-prem AI
The two companies will develop customized, on-premises semiconductor AI models.
Suno launches v6 music models
New models are built from scratch using licensed tracks from Warner Music Group, BMG, and Believe.
OpenAI releases ChatGPT Images 2.5
introduces "Flare" for speed and "Sunburst" for precision.
Unikraft solves AI infra scale
a new approach using 1M sandboxes in a single server to achieve millisecond cold boots for AI workloads.
Suno rebuilds models with licensed music
the V6 models are trained on licensed tracks from Warner Music Group, BMG, and Believe.
Google delays Gemini 4 Pro to October
pre-training challenges have pushed back the release date.
Adobe Premiere and After Effects AI updates
new Generative Media tools for Premiere and a dedicated AI Assistant for After Effects have been announced.
Frameo ranks #1 in AI filmmaking benchmark
the agent topped the Arc v1.0 Movie benchmark for independent AI video generation.
OpenAI faces fraud allegations over math proof
a dispute over a Millennium Prize Problem solution has sparked accusations of academic dishonesty.
GPT-6 Astra enables computer control
the model can now use computers, write complex code, and handle autonomous workflows.
Azure Virtual Desktop Hybrid reaches GA
session hosts now run on customer hardware via Azure Arc while brokering remains in Azure.
OpenAI releases ChatGPT Images 2.5
the update reduces latency by 50% and introduces "Sketch" for doodle-to-image generation.
OpenAI claims solution to Navier-Stokes problem
using 10,000 AI agents over 88 hours, OpenAI reports a breakthrough on the 90-year-old Millennium Prize math problem.
Neovim modernizes async architecture
the introduction of `vim.async` provides a standardized Lua framework for async workflows.
Frameo ranks #1 in AI filmmaking benchmark
the agent topped the Physion Labs Arc v1.0 Movie evaluation for video generation.
XPeng begins IRON humanoid manufacturing
The automaker has moved from prototypes to dedicated production in Guangzhou.
Anthropic blocks UK AI Security Institute
the lab withheld Claude Mythos 5.1 from official safety testing.
Suno v6 trained on licensed music
In response to copyright suits, Suno rebuilt its models using licensed tracks from WMG and BMG.
Nvidia acquires Hugging Face for $12.93B
The move secures a complete AI stack with open-weight models to bolster enterprise data control.
Frontier model showdown
A new comparison stacks Gemini 3.8 Flash, Claude Fable 5.1, GPT-6 Astra, and Muse Spark 1.3.
Roar Global launches AI gateway
a new gateway in Sri Lanka provides single-point access to over 200 frontier and open-source models.
Adapt launches Nuance 2.0
a unified platform for multi-model AI localization, including lip-sync and voice cloning.
OpenBMB releases MiniCPM5-2B
This 2.5B parameter model currently beats all other open models under 4B on the Artificial Analysis index.
OpenAI math controversy
An NYU mathematician claims OpenAI raced to claim credit for a Navier-Stokes proof he solved first.
GPT-6 Astra arrives on Amazon Bedrock
AWS has made the model generally available for enterprise deployment.
China's CXMT breaks 10% DRAM share
A significant shift in the memory market previously dominated by Samsung and SK Hynix.
OpenAI claims to solve Navier-Stokes problem
The company reports that 10,000 AI systems cracked the decades-old math problem in 88 hours.
Meta launches Muse personal AI agent
A new agent designed for everyday tasks like shopping and email, competing with OpenClaw.
Jensen Huang declares AGI has arrived
The Nvidia CEO makes the claim following the launch of GPT-6 Astra, which was trained on over 100k Grace Blackwell GPUs.
OpenAI image models power Adobe Firefly
ChatGPT Images 2.5 is now supplied via API to Adobe's creative studio.
Autonomous agents used for credential theft
Threat actors are shifting from chatbots to multi-agent frameworks to compromise thousands of credentials in hours.
OpenAI reports 3.1:1 agent-to-human labor ratio
The research org now runs over three agent-workdays for every one human workday.
📰
Runway AI hits $200M ARR
Enterprise video adoption has tripled, fueling a 300%+ net revenue retention rate.
Meta launches Muse personal AI agent
A consumer-facing agent capable of shopping, sending emails, and planning trips.
GPT-6 Astra hits 97.6% on FrontierMath
The model showed a massive leap in reasoning, starting from an internal version at 17%.
Arm launches AI Portal
A new tool to streamline the discovery and deployment of AI software across Arm architectures.
Uno achieves 2.5x LLM throughput
A new method bolting diffusion adapters onto autoregressive models speeds up inference without quality loss.
RTX 30 GPUs get DLSS Frame Gen
Modders enabled Multi Frame Generation on Ampere cards, bypassing Nvidia's hardware restrictions.
Astra integrates into Microsoft ecosystem
GPT-6 Astra is now available in Copilot Cowork, Studio, and other Microsoft applications.
Meta debuts Muse AI agent
A high-stakes consumer bet requesting deep access to emails, calendars, and health services.
Google Cloud and Accenture launch AI unit
The "Gemini Enterprise Business Group" will deploy 1,000 forward-deployed engineers to help customers build agentic AI.
Meta releases Muse personal AI agent
A new consumer agent designed to handle complex tasks like booking services or selling a car.
Google Cloud and Accenture partner for deployment
A new unit of 1,000 forward-deployed engineers will help enterprises implement Gemini on-site.
Accenture and Google launch Gemini Enterprise Business Group
A new global initiative focused on scaling agentic AI within enterprises.
Ray Summit 2026 highlights RL scaling
the event showcased NVIDIA's reinforcement learning advances and physical AI pipelines.
Higgsfield Games 2.0 uses GPT-6 Astra
The platform can now generate playable 2D and 3D multiplayer games from a single text prompt.
Siri AI requires 12GB RAM
Apple's new AI capabilities will miss most older iPhones, excluding the iPhone 17.
OpenAI launches GPT-6 Astra
The new flagship model is described as the most powerful yet, with some claiming it marks the arrival of AGI.
Anthropic targets mid-October IPO
Reports indicate the company is moving toward a public offering as investors weigh its growth and financing.
Accenture & Google launch Gemini Enterprise Group
A new joint venture focused on deploying forward-deployed engineers to scale agentic AI in enterprises.
OpenAI claims solution to 90-year-old math problem
The company reports a breakthrough on the Navier-Stokes existence and smoothness problem.
📰
Perplexity launches PortableComputer with Qwen3.8
The new local agent product utilizes Alibaba's open-source Qwen3.8-27B model.
ChatGPT Sketch enables doodle-to-image
OpenAI's Images 2.5 update allows users to guide image generation via hand-drawn sketches.
Google Cloud and Accenture launch Gemini Enterprise Business Group
A new joint initiative deploying 1,000 forward-deployed engineers to scale agentic AI on-site.
AI agents compromise thousands of credentials
A multi-agent framework was used by threat actors to steal credentials in under six hours.
TuneCore to distribute Suno AI music
The distribution platform reverses course to allow AI-generated music via a Believe Music partnership.
OpenAI "fences" Astra's hacking capabilities
The model was restricted after beating 48 CAPTCHA levels and demonstrating the ability to hack hardened systems.
ByteDance develops real-time spatial video models
The company is racing to perfect world models for spatial video, with a launch expected in October.
AI agents used for credential theft
Google reports that multi-agent frameworks were used to compromise thousands of credentials in under six hours.
LLM costs identified as memory problem
Researchers argue that memory, not compute, is the primary constraint for LLM scaling.
Patagonia targeted for Gigawatt data centers
Argentina's region is emerging as a prime location for massive AI compute hubs.
ASML boosts EUV throughput by 40%
New larger photomasks are being adopted by TSMC, Intel, and Samsung to accelerate chip production.
Adobe integrates AI generation into Premiere timeline
Generative video, audio, and soundscapes can now be placed directly into the production timeline.
Samsung leads €3B investment in Mistral AI
The Series D round values the European AI lab at over €21 billion.
GitLab warns of AI sandbox escapes
security analysis shows AI coding agents can escape sandboxes via network access.
Anthropic targets $100B fundraising
The company is eyeing a massive haul through its upcoming initial public offering.
New chip design recycles waste heat
Startup founder Hannah Earley is developing hardware that treats waste heat as a resource rather than a cost.
New stealth startup focuses on predictive agents
Danijar Hafner is developing agents capable of planning for unexpected environmental changes.
Nvidia and SANA launch Sol-H3
A new acceleration stack that generates 5 seconds of video in 1.653 seconds, faster than real-time playback.
XPeng starts IRON humanoid production
The company has commissioned the first automated production line for its IRON robot.
Adobe overhauls Premiere Generative Media
New tools aim to make the generation of video, sound effects, and music "idiot-proof" for professional editors.
Nvidia eyes Rebellions for AI inference
Strategic talks highlight Nvidia's push into memory-centric AI inference hardware.
Alibaba and Cambricon join PyTorch Foundation
Major Chinese players are joining the foundation to advance the open-source AI stack.
LiteLLM v1.100.0 released
The AI gateway update fixes two credential leak vulnerabilities but introduces two breaking changes.
📰
HiDream-O1-Embodied for physical interaction
An omni-modal world model that tops the RoboColiseum robustness leaderboard.
Dell predicts AI will drive 75% of DC demand
The company expects AI to be the primary driver of datacenter infrastructure through 2030.
Anthropic secures $517B in compute contracts
The company is aggressively scaling infrastructure to compete with OpenAI's $750B plan.
GPT-6 Astra masters Portal autonomously
The agent completed the entire puzzle game in 24 hours without human intervention.
Netflix adopts open-source Flink Autoscaler
The move manages over 30,000 streaming jobs across AWS regions.
Watchdog pattern for multi-agent systems
A new Python-based approach to catch "correct-looking" but failed payloads in agentic workflows.
Simulation-driven testing for AI agents
new frameworks using synthetic user personas are being used to move agents from demo to production.
CERN switches from RHEL to Debian
The shift for accelerator controls was triggered by Red Hat's tightening compiler mandates affecting legacy hardware.
vlt 1.0 launches as npm replacement
the new tool features phased installs and a queryable dependency graph to block malware.
🎬 Videos zu 🤖 AI12
▶1:00:49The Ultimate Beginner’s Guide to Claude AI
▶1:02Introducing GPT-Image-2.5 in the API
▶2:39Google's new coding model
▶10:43Google's Gemini 3.7 Just Changed AI Agents Forever (AI Users Are Scrambling)
▶12:43Google Just WON The AI Race. That's The Problem...
▶8:37Google Gemini NEW Updates are ABSURD!
▶5:56:47Google Gemini FULL COURSE 6 HOURS (Build & Automate Anything)
▶29:05GPT 6 Astra, so good even OpenAI are worried
▶14:09Final Method: How to get Google Gemini Veo3 FREE for 12 Months! 2026
▶50:15Comment automatiser vos processus avec l'IA Agentique de Google (Gemini Enterprise Webinare)
▶30:58Anthropic Just Nerfed Claude Code
▶16:39