OpenRouter: AI Agent Token Use Jumps 14x, Overtakes Humans
Since February 2026, agent token usage has grown 14-fold, compared to just 2.8x for humans
50
AI
AI models, services and robotics
Claude Code drops npm install in favor of script-based installation
Anthropic's coding agent, now past 140,000 GitHub stars, has made curl, Homebrew, and WinGet the official recommended install methods instead of npm
410
Adding a 'mind' variable to world models boosted accuracy from 63 to 88
A new framework adds belief and intention to world models that, like Sora and Genie 3, only track physics
40
OpenHuman: A Local AI Agent That Builds Memory From Personal Data
An open-source tool that builds memory from personal data and commands a fleet of agents is drawing attention
80
MCP's New Roadmap Enables Stateless, Horizontally Scalable Servers
MCP, the standard for connecting AI tools, has revised its roadmap just five months after the last one. Servers no longer need to hold onto state, and agent identity authentication is up next.
90
Turn technical book PDFs into AI skills, save up to 51x on tokens
book-to-skill, an open-source converter for use with Claude Code, Copilot CLI, and Amp
230
Anonymous model Ox Alpha matches GLM-5.2 on all 60 tokenizer tests
Community tokenizer comparison shows 60/60 match with zero average error — but that doesn't confirm it's the same model
70
Reddit developer releases 'Unswarm' to auto-switch between multiple local LLMs
An open-source tool that unifies local LLM operations — previously spread across separate containers and scripts — behind a single proxy
70
Karpathy's LLM coding critique, distilled into a single CLAUDE.md file
A guideline file that curbs Claude Code's habit of making assumptions and bloating code has hit GitHub's trending page
110
OpenAI Codex CLI: The Open-Source Coding Agent That Runs in Your Terminal
Here's what you need to know about the open-source coding agent that works with your ChatGPT plan or API key — from installation to how it differs from the IDE, app, and web versions
130
LinkedIn's AI slop report button passes 1 million clicks
The reporting feature, launched last month, has drawn heavy use — and views on flagged posts have dropped sharply
60
NVIDIA Swaps Harness, Lifts AI Agent Score from 30% to 100%
Custom harness on Claude Opus 5 hits perfect score on ARC-AGI-3 — the scaffolding, not the model, made the difference
160
Netflix Pilots In-House Language Model GenRec in Recommendation Engine
GenRec converts viewing history into sentences instead of thousands of hand-crafted features, posting meaningful gains in online testing
50
RayNeo unveils camera-free AI glasses "iO Glasses"
Waveguide glasses with no speaker or camera that only display text, with a subscription adding ChatGPT and Gemini on top
860
AI Safety Scores Can Be Gamed Just by Refusing More
UK AI Security Institute and others used psychometric methods to dissect 8 safety benchmarks and found they were measuring three different things
30
Apache Foundation incubates local-first AI agent tool 'Maka'
A local-first agent that logs every model message, tool call, and permission decision — currently officially supported only on Apple Silicon Macs
20
ComfyUI Open-Sources Local MCP Server
Comfy Org open-sourced a local MCP server on August 18. Claude, Cursor, and Codex can now directly open ComfyUI on your own machine, checking your GPU specs first and pulling only models it can handle.
50
Hume AI Measures Benchmark Memorization in Speech Recognition Models
Even after silencing unspoken numbers, some models still transcribed them correctly 30-40% of the time
40
Georgia Tech Traces Roots of Social Reasoning Using Fully Open Olmo 3
Sampling 5.68 million of 1.26 billion training documents to find out which texts build which capabilities
290
OpenAI cuts GPT-5.6 Sol API pricing by over 20% for three months
API and credit prices drop citing efficiency gains, fixed-plan subscription usage unchanged
120
Google Research unveils AI framework that understands places through mobility data
By adding people's movement rhythms to AI that previously only read text, the framework improves predictions of business hours, price range, and crowd levels
630
OpenAI Rolls Out Per-API-Key Spend Tracking and Limits
API key filters added to the Usage/Spend dashboard let developers see exactly which app is spending how much
10
Google Research unveils AI that prioritizes depression biomarker candidates from wearable data
A multi-agent system analyzed 9,279 participant-observations to find a link between sleep variability and depression
20
Claude Security scans code with new Mythos 5 model
Anthropic launches enterprise-only public beta — model stays hidden behind the scan, only results are surfaced
20