Mozilla's inaugural State of Open Source AI report finds open models now match closed ones on capability, but only 51% reach production. The report highlights real-world adoption across sectors, from Māori language preservation to finance.

open sourceFriday, July 17
Open source AI hits a capability milestone
Today's big story is that open source AI has caught up to closed models on pure capability, according to a major new report from Mozilla. But the same report reveals a deployment gap that's keeping most models out of production, while China's Kimi K3 model stirs debate about what 'open source' really means.
Capability vs deployment
Two reports today frame the state of open source AI: one says the models are good enough, the other says getting them into production is still the hard part.

A separate take on the same report emphasizes the 'harness layer' and permission model gaps as the real barriers, not model quality. The 3% capability gap is essentially closed.
China's open-weight contender
Moonshot AI's Kimi K3 is drawing attention for its open-weight strategy and low pricing, but a closer look questions how open it really is.
Moonshot AI's Kimi K3 is being hyped for its coding ability, low API pricing, and open-weight release. It signals a shift toward more accessible AI, but questions linger about its true openness.

A technical deep dive questions whether Kimi K3 is truly open source, given its model weights are available but training data and code are not. The piece also estimates what it would take to recreate it.
Security and infrastructure
Open source tools for security and infrastructure keep evolving, from Capital One's AI code scanner to Linux kernel review with AI.
Capital One open-sourced VulnHunter, an AI code security tool with agentic reasoning. It's a defensive response to the growing threat of AI-powered vulnerability discovery.
Linus Torvalds endorsed Sashiko, an AI kernel reviewer that caught 53% of bugs human reviewers missed. Someone even forked Linux over it, which Torvalds reportedly encouraged.
Also today10
Beyond the Hardware Hub: How ALC Taipei Turned Computex 2026 into a Milestone for the Apache Waynews.apache.org
kurama/dealve-tui: Cross-Store Game Deal Tracker for Your Terminalblog.brightcoding.dev- Ponytail: The Agent Skill That Forces AI to Write Less Code (74K Stars)byteiota.com
Critical WordPress Flaw Allows Code Executioncyberwebspider.com
First Contributions: Your Launchpad to Open Source Masteryblog.brightcoding.dev
These 5 Unix ideas from the 1970s are why Linux still works so wellwww.howtogeek.com
4 uncomfortable truths about Home Assistantwww.howtogeek.com
It's all fun and games until the lights won't turn on.
Patreon begins blocking AI training bots after years of warning them offsuperintelligencenews.com
Patreon is blocking AI scraping with Cloudflare tools, replacing robots.txt requests with direct enforcement to protect creators' work. The post Patreon begins blocking AI training bots after years of warning them off appeared first on Superintelligence News - Artificial Intellig
Netflix dumps gRPC for SSE, eyes more AIwww.thestack.technology
Any company that decides to build a graph database alternative on Cassandra probably has novel ideas worth listening to...
Yazi: Fast Rust-Based Terminal File Manager for Linuxwww.tecmint.com
The post Yazi: Fast Rust-Based Terminal File Manager for Linux first appeared on Tecmint: Linux Howtos, Tutorials & Guides . Yazi is a terminal-based file manager written in Rust that lets you browse, preview, and manage files. It uses asynchronous The post Yazi: Fast Rust-Based
More roundups today
wellness roundupvolufiline, Blue Zones, and burnout
sustainability roundupAnimation gets its own green certification
real estate roundupHousing market pain deepens across the U.S.
memoir roundupMemoir as lens on body, music, and loss
literature roundupNolan's Odyssey and Paris Review classics
interviews roundup