Anthropic opens Claude usage data to three outside research teams
Stanford, Oxford, and METR independently analyzed real Claude usage data, and Anthropic did not intervene in the findings
30
AI
AI models, services and robotics
Google's Gemini 3.5 Transcribe automatically strips out verbal filler words
Live and Live Experimental also launched alongside it, but the Gemini 3.5 Pro model teased back in June still hasn't shown up
20
Ox Alpha turns out to be GLM-5.3-Flash
The anonymous model that topped usage charts for a week was Z.ai's new multimodal model. All of its traffic ran on Chinese-made AI chips.
120
OpenAI introduces Runme, which uses Codex to document repetitive work
An OpenAI engineer explained how Codex automates repetitive evaluation and infrastructure tasks, capturing the whole process in Jupyter-style notebook documents
20
Meta's paid agent Hatch to launch alongside new Watermelon model in October
Internal documents reportedly show a $199.99-a-month subscription under review, tied to a dashboard linking DoorDash, Etsy, and Reddit
40
Databricks lays out seven criteria for choosing governance tools
The company recommends comparing tools by six capabilities—cataloging, lineage, access control, and more—plus seven evaluation criteria, rather than by brand name
10
Solo Trader Runs Overnight Desk With Six Grok Bot Agents
The trader says the laptop stayed shut while six agents placed orders overnight, with a human only clicking final approvals
140
Samsung Design Chief Says AI Has Increased Phone Usage Time — Z Fold8's Form Factor Came From a Passport
Executive Vice President Hubert Lee, who led the design of the Galaxy Z Fold8, has revealed the origins of the new form factor. It started with the one-handed usability of a passport, and behind it lies a usage pattern that AI has changed.
60
Claude tests 'Morning Brief' feature built on scheduled tasks
TestingCatalog spotted it on some users' screens — a feature that shows a daily rundown each morning, pulled together from connectors
50
OpenAI reveals eight Build Week challenge winners
Nearly 47,000 people from 186 countries submitted more than 8,000 projects, and on August 25 OpenAI announced eight category winners. Two of the four first-place teams were a veterinarian and a cardiologist with no coding background.
1470
OpenAI Blocks Cluster of ChatGPT Accounts Behind Russian Influence Campaign
Accounts fronting fake think tank "International Burke Institute" were pushing pro-Russia narratives
110
ChatGPT Work can now handle tasks on sites that require login
For Plus and Pro plans, ChatGPT itself pulls up the login screen and now supports password managers
40
Perplexity's local agent beats Hermes, Pi in benchmarks
Perplexity says its 27B on-device model paired with a custom harness outperformed open-source competitor harnesses on knowledge-work, web search, and document-parsing tests
30
Qwen3.8-27B Can Now Be Trained on a 24GB Graphics Card
Unsloth has lowered the barrier for fine-tuning a 27B model with free notebooks and optimized kernels.
130
OpenAI launches WebMCP Challenge: $35,000 hackathon
Partnering with Chrome, Cloudflare, Shopify, Vercel, Render, and Netlify for a 10-day event
50
Google Automates Legal Work With Gemini Enterprise for Legal
Connected to iManage, DocuSign, and more via MCP connectors, it handles contract review and legal research
30
Mac Studio M5 Ultra hits 4.8TB/s bandwidth when four units are clustered
Apple unveils Mac Studio with M5 Max/Ultra and Mac mini with M6/M5 Pro, targeting local large models through clustering
50
ChatGPT Business Premium Seat: 5x Usage for $100
OpenAI has officially introduced a higher-tier seat aimed at small teams and startups. It offers five times the usage of the Standard seat.
60
Claude merges chat and cowork memory into one system
Anthropic has unified memory across chat and agentic cowork, so context now carries over no matter where you pick things up
10
Google Research unveils AgentHands, an XR agent that gestures
It's a prototype that adds hand gestures to voice-only AI assistants, letting them point out tasks in real physical space
50
Perplexity brings cloud-free agents to DGX Spark
Pro and Max subscribers now get a portable computer that runs the orchestrator, sub-agents, and harness entirely on local hardware.
60
Anthropic funds $5M research program for AI wellbeing evaluations
External researchers building open-source evaluations get funding, model access, and technical support — applications close September 21.
20
Multiverse Computing shrinks a model to 4-bit and gets one smarter than the original
Compressing and quantizing GPT-OSS 120B down to 60B beat the original on 7 of 9 benchmarks
20
Ukraine Opens 5 Million Real Combat Photos to the UK First
Western defense AI, trained for years on nothing but synthetic data, gets its first look at real combat records. The data itself never leaves Ukraine.
110