GLM 5.3 Flash vs DeepSeek V4 Flash: user comparison

a16z's Erik Torenberg, Ben Horowitz, and Jennifer Li interview Vals founder and CEO Rayan Krishnan on the challenge of measuring model improvement as public benchmarks saturate and models optimize for them.

Cohere's video examines which tasks and occupations AI tools concentrate around, based on an analysis published at mcp-landscape-article.vercel.app. It asks whether AI tools are truly everywhere or clustered in specific areas.

Okta analyzed a 7 GB infostealer dump from Telegram containing data from 5,871 infected machines across 162 countries. Of 44,791 JWTs, 555 were likely AI-service auth tokens, and 2,937 JWEs were mostly set by OpenAI.

Ramp data shows 56% of its 70,000 business customers paid for AI products in August, up just 0.4% from July. Spend per employee at top 1% of firms fell nearly 10% to $7,205, as average token costs dropped to $0.68 per million from a March peak of $1.15.

QwenLM released cua-driver-rs v0.20.0 through v0.20.5, providing prebuilt binaries for macOS (codesigned/notarized), Linux (unsigned), and Windows (unsigned). The releases also publish the @qwen-code/cua-sdk npm package.