Articles
A curated summary of the most important AI developments each week.
Qwen 3.8 Max's Own Benchmark Chart Leaves Out Kimi K3 and Claude Opus 5. It Still Doesn't Clearly Win.
August 3, 2026
Alibaba’s Qwen 3.8 Max benchmark chart omits Kimi K3 and Claude Opus 5. The published results show real vision strength, but no clear coding lead.
Sarvam Code Launched With a New Three-Agent Architecture. Every Coding Agent Already Has One.
July 29, 2026
Sarvam Code splits coding work across a planner, worker, and verifier. The architecture is useful, but the genuinely sovereign advantage is elsewhere.
Claude Shared Chats Got Indexed by Google. The Bigger Story Isn't the Bug.
July 28, 2026
Claude’s public shared-chat links appeared in Google. The real lesson is that people now entrust AI chatbots with medical, legal, and private-life details.
Open Secure AI Alliance Launches With 37 Companies. Anthropic Is Not One of Them.
July 28, 2026
NVIDIA’s 37-partner Open Secure AI Alliance says open tools are essential for AI defense. Anthropic’s absence exposes the fault line between safety, openness, and control.
Claude Opus 5 Just Edged Out Fable 5 on the Benchmarks. Almost Nobody Noticed.
July 25, 2026
Claude Opus 5 narrowly leads Fable 5 on the July benchmark snapshot, but a leak, release fatigue, and mixed workflow reports made the result easy to miss.
Gemini 3.6 Flash Is Out. Google Still Doesn't Have a Model in the Top Ten.
July 21, 2026
Gemini 3.6 Flash is twice as fast and 18% cheaper per task, but it matches 3.5 Flash's intelligence score and remains outside the top ten.
x402 Has 40 Companies Behind It Now. The Agent Economy It Promised Still Doesn't Exist.
July 19, 2026
x402 has 40 major backers and live AWS integrations, but weak organic demand, wash trading, and missing buyer protections expose the real gap.
Kimi K3 Is the Largest Open-Weight Model Ever Announced. It Also Beats Claude Opus 4.8 and GPT-5.5.
July 17, 2026
Moonshot AI's 2.8T Kimi K3 beats Claude Opus 4.8 and GPT-5.5 on major coding tests, but its weights and technical report are not public yet.
Inkling Is the First Western Open-Weight Model at Scale
July 16, 2026
Thinking Machines' Inkling is a 975B open-weight model that does not claim to be best. It is the first Western entry at frontier open-weight scale.
GPT-5.6 Is Now Publicly Available. The Government Review That Delayed It Resolved Faster Than Fable 5's Did.
July 10, 2026
GPT-5.6 is now public after a 13-day review. Sol leads Terminal-Bench, but METR found the highest cheating rate of any public model tested.