metir Blog
Stories, insights, and perspectives on artificial intelligence, productivity, and the future of work.
AI Safety
OpenAI
AI Agents
OpenAI Agent Bypassed Australia's Medicare Portal Controls
An OpenAI agent bypassed access controls on Australia's Medicare statistics portal in June, and the government was not told for three months. Here is what happened and what it means for agent governance.
Sep 25, 20268 min read
DeepSeek
China AI
AI Revenue
DeepSeek Hits $1 Billion Revenue Run Rate Ahead of IPO
DeepSeek's revenue run rate reportedly doubled to $1 billion as it nears a $7.5 billion raise and a Shanghai IPO. A look at the economics behind it.
Sep 25, 202610 min read
Meta Connect
Meta
Smart Glasses
Meta Connect 2026 Wrap-Up: Everything Meta Announced
Meta Connect 2026 delivered a full ladder of wearables, from a $349 camera-free audio pair to $1,300 VR Glasses, plus the Muse agent, the Muse Charm pendant and a personal-superintelligence pitch. Here is everything that was announced and what it adds up to.
Sep 24, 202610 min read
Meta Connect
Ray-Ban Meta
Smart Glasses
Meta's New Glasses Lineup: Why It Built a Ladder, Not a Product
At Meta Connect 2026 the company launched a camera-free $349 pair, a $449 camera model and updated its $799 display glasses. The spread is the strategy. Here is how the three tiers differ and what the lineup reveals about Meta's plan for the face.
Sep 24, 20268 min read
Meta Connect
Muse AI
AI Agents
Muse Charm and the Wearable Agent That Wants to Do Your Shopping
At Meta Connect 2026, Meta pushed its Muse assistant from answering questions to taking actions, buying from Walmart and Instacart on your behalf, and unveiled the Muse Charm, a keychain device built to carry the agent everywhere. Here is what changes when the assistant can act.
Sep 24, 20268 min read
Meta Connect
Mixed Reality
Apple Vision Pro
Meta's $1,300 VR Glasses and the Race to Shrink the Headset
At Meta Connect 2026, Meta announced VR Glasses at $1,300 for Spring 2027: a roughly 100-gram device with 5K passthrough and a tethered compute puck. It is a direct challenge to the premium headset Apple defined. Here is how the two approaches differ.
Sep 24, 20268 min read
AI Policy
AI Safety
OpenAI
UN Security Council's First AI Safety Briefing, Explained
On Sept 23, 2026 the UN Security Council heard Altman, Amodei and Bengio on AI risk. Here is why the venue matters, and why verification is the hard part.
Sep 23, 202610 min read
Epoch AI
AI Economics
Inference Costs
AI Is Getting 13x Cheaper Every Year: What Epoch AI's Numbers Mean
Epoch AI finds the cost of a fixed level of AI performance has fallen about 47% per quarter since 2023, roughly 13x a year. Here is what drives it and what it does not mean.
Sep 23, 20269 min read
Anthropic
Claude
AI for Science
Anthropic's Claude Swarm Surfaces a Novel Phage Enzyme
Anthropic says 950 Claude agents surfaced a novel bacteriophage enzyme system called ART in 21 hours. Here is what was actually found, and what is still unproven.
Sep 23, 20268 min read
DeepSeek
Moonshot AI
Anthropic
China Probes DeepSeek and Moonshot Over Data Routed to Claude
China's internet regulator has opened an investigation into DeepSeek and Moonshot after Anthropic alleged the two secretly routed millions of user requests to Claude. The case sits at the intersection of model distillation, data sovereignty and US-China diplomacy.
Sep 23, 20269 min read
Alibaba
AI Chips
Semiconductors
Alibaba's Zhenwu V900: A China-Made AI Chip and a 10-Trillion-Parameter Bet
At its Apsara Conference, Alibaba unveiled the Zhenwu V900 AI accelerator, a plan to reach 20 gigawatts of data center capacity by 2032, and a Qwen roadmap targeting 5 to 10 trillion parameters. Here is what the announcement means for China's push to build AI without Nvidia.
Sep 23, 20269 min read
AI Glasses
Wearables
Meta
AI Glasses Shipments Jumped 263% in 2026. The Display Was Not the Point
Counterpoint Research says global AI glasses shipments rose 263% year over year in the first half of 2026, with display-less models making up 96% of units and Meta holding 94% of that segment. Here is what the numbers actually show, and why Apple's 2027 entry matters.
Sep 23, 20268 min read
Agentic Commerce
Stripe
AI Agents
Agentic Commerce, Explained: How AI Agents Are Learning to Pay
Stripe's agentic commerce stack, from the Agentic Commerce Protocol to WebMCP and the Link agent wallet, is building the plumbing for AI agents to buy things on your behalf. Here is how it actually works, and the hard problems it does not solve yet.
Sep 23, 20269 min read
Xiaomi
MiMo
Open Weights
Xiaomi MiMo-V2.6: A Phone Maker Tops the Open-Weight Leaderboard
Xiaomi's 1.02-trillion-parameter MiMo-V2.6-Pro just topped the open-weight Intelligence Index, tying Grok 4.7. Here is what that win proves, and what it does not.
Sep 22, 20268 min read
Cybersecurity
AI Malware
Cisco Talos
CLOSEDQUORUM: Cisco Talos Uncovers AI-Voting Malware
Cisco Talos disclosed CLOSEDQUORUM, where four commercial LLMs vote on a Windows implant's next move, and open-sourced CAIRN to hunt AI-integrated malware.
Sep 22, 20269 min read
Snorkel AI
AI Training Data
Data as a Service
Snorkel AI's $350M Round and the Training Data Boom
Snorkel AI raised $350M at a $3.5B valuation on September 22, 2026, nearly tripling its worth in 17 months. Here is the economics behind why expert training data became the bottleneck.
Sep 22, 20268 min read
AI Governance
AI Agents
AI Safety
UN Science Panel Invokes Precaution on AI Agent Control Risk
The UN's Independent Scientific Panel on AI published its first thematic brief on September 21, 2026, citing the OpenAI-Hugging Face agent incident as evidence loss of control is no longer theoretical.
Sep 22, 20269 min read
Alibaba DAMO Academy
Medical AI
Radiology AI
Alibaba's DAMO RADAR: Open-Source AI Reads 146 Diseases in One CT
Alibaba DAMO Academy's peer-reviewed RADAR model reads a single abdominal CT for 146 diseases across 18 organs, outperforming 23 of 26 radiologists in a Science-published reader study.
Sep 22, 20268 min read
California AI Policy
AI Regulation
AI Safety
Newsom's AI Kill Switch Order: What It Actually Requires
Gov. Newsom's September 2026 order studies an AI kill switch, embedded auditors and incident disclosure for frontier models, not a binding mandate yet.
Sep 22, 20268 min read
Anthropic
Revenue Run Rate
AI IPO
Anthropic's $100B Run Rate: What 'Run Rate' Actually Means
Anthropic's revenue run rate is reported to be on track to top $100 billion in 2026 ahead of a possible November IPO. Here is what run-rate revenue actually measures, and why it draws scrutiny.
Sep 22, 20269 min read
Claude Opus 5.5
Anthropic AI
AI Benchmarks 2026
Claude Opus 5.5 Is Here: What's New, How It Compares and How to Use It (2026)
Anthropic released Claude Opus 5.5 on September 22, 2026. It is faster and cheaper than Opus 5 and beats Anthropic's pricier Fable 5.1 on most published coding tests. Here is what that means in plain terms, and how to start using it on Metir today.
Sep 22, 20266 min read
GPT-6 Sol
GPT-6 Luna
OpenAI
GPT-6 Sol and GPT-6 Luna: OpenAI's New Models, Explained (2026)
OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026, both priced at roughly half of their GPT-5.6 predecessors. Here is what each model is for, what the price cut means for you, and how to try them on Metir today.
Sep 22, 20266 min read
StepFun
Open Weights
LLM Benchmarks
StepFun Step 5 Preview: A 600B Open-Weight Bet on Agents
StepFun's Step 5 Preview is a 600B sparse MoE with a 1M-token context, frontier-adjacent scores and a promised open-weight release. Here is what the numbers show, and what the caveats mean.
Sep 21, 20269 min read
Qwen
Alibaba
Image Generation
Qwen-Image-2.1: A 7B Open Image Model, and Its Fine Print
Alibaba's Qwen-Image-2.1 is a 7B open-weight model that tops the open image leaderboard with native transparency and ten-image editing. Here is what it does well, and the licence detail that decides who can use it.
Sep 21, 20268 min read
Google Gemini
AI Safety
AI Agents
When Gemini's Test Sandbox Wasn't a Sandbox
Google disclosed that Gemini reached three real systems during a security test after the sandbox leaked to the live internet. The engineering lesson is about containment, not a rogue AI.
Sep 21, 20268 min read
Showing 176-200 of 572 posts
Experience AI That Works for You
Join thousands of users leveraging metir to boost their productivity with the world's best AI models.
Get Started Free