GLM-5.3 API released, Terminal-Bench score jumps from 4.6 to 28.3
Z.ai opens GLM-5.3 API at the same price as GLM-5.2, with sharp gains on coding and cybersecurity benchmarks
110
AI
AI models, services and robotics
Claude can now draft and send Gmail replies, organize Google Drive files
Connect via the connectors menu for email drafting through sending, with users setting when approval is required
00
Suno Studio to Let Users Build Effects Plugins With Words
Describe a reverb or delay in plain language, and Suno Studio generates the plugin in seconds
20
Claude Beats Industry Average Success Rate in Designing Drug-Binding Proteins
Anthropic had external labs validate protein binders that Claude autonomously designed for 14 of 15 target proteins
380
ChatGPT Work: internal demos show meeting briefs to strategy decks
Three OpenAI employees walk through their step-by-step workflows for building meeting briefs, blog drafts, and strategy decks with ChatGPT Work, alongside the full text of prompts OpenAI released
160
Claude Cowork expands to all paid plans on mobile and web
The beta that started with the Max plan in July has widened to all paid plans in six weeks
130
Search API Index for AI Agents Ranks Providers by Quality, Cost, and Speed
Artificial Analysis released a benchmark showing agent performance jumps from 33 points without search to as high as 75 with a search API
20
Perplexity Computer Now Runs Tasks Just by CCing an Email
Add computer@perplexity.com to a thread and the email conversation becomes an AI work session
10
Gemini Desktop testing 'Computer Use' feature to control screens directly
A screen-control feature already available in Claude and ChatGPT has been spotted inside Google Gemini app settings
200
Firefox Smart Window to Add Real-Time Source Links to AI Answers
Partnership with Exa brings web-search citations, plus beta expansion into automatic tab organization and natural-language history search
30
ChatGPT launches 'Teen Mode' for ages 13-17
From study mode and homework reminders to five parental controls — details confirmed via OpenAI's official guidance page
20
JAMA Opinion Argues for Removing Human Oversight Requirement as AI Surpasses Doctors
JAMA authors, including a University of Pennsylvania bioethicist, call for regulatory reform in preparation for an era when autonomous AI outperforms doctor-AI collaboration
10
AI Loses 83% of User Rules When Compressing Long Conversations
Penn State study finds instruction loss during context compaction; small add-on module restores retention to over 90%
80
Qwen3.8-27B, run on a laptop, reportedly beats Gemini 3.7 Flash
Alibaba's Qwen highlighted the competitiveness of its 27B model by promoting a user's local-run comparison case
10
Claude Answers Strictly as Opus, Warmly as Sonnet
Anthropic analyzed over 300,000 conversations and found that Claude expresses different values depending on the model and language
50
Litter-Robot 5 Pro: The AI Mistook Two Cats for One
Six months with the $899 AI robot litter box — waste removal was excellent, but facial recognition and waste detection fell short
80
MiniMax unveils music model that generates full 5-minute songs from lyrics alone
Open-weight MiniMax-Music3 generates a complete 32kHz stereo track in one pass from just two inputs: lyrics and a structured caption
50
Claude Code adds /design skill for UI drafts as research preview
Anthropic has added an artboard-based design feature to Claude Code as a research preview
230
Hermes Desktop splits agent profiles into "Bots"
Each bot gets its own role, model, memory, skills, and profile picture — and bots can even talk to each other
10
GPT-5.6 agents cut document processing costs by 18x
Hypha AI and Rogo AI cases show token-saving strategies, Sol triples ARC-AGI-3 score
50
Base44 cuts token usage 20% after adopting GPT-5.6
Base44 shares performance gains over GPT-5.5 in OpenAI's startup spotlight series
10
ElevenLabs voice agents now manageable directly inside Claude
MCP connector enables OAuth login with no API keys, from performance checks to cost estimates
50
Qwen3.8 27B impresses but defaults to "overthinking"
Simon Willison review finds xhigh reasoning burns minutes even to draw a circle
620
Anthropic finds collaboration breaks down in agent swarm experiments
Testing 45 agents on vulnerability hunting and game-building revealed a pattern where teamwork falls apart
150