Jul 31, 2026
DeepSeek's cheap model now beats its own flagship on all nine agent benchmarks, and it drops straight into OpenAI's Codex tooling. That landed six hours after OpenAI slashed Luna by 80 percent.
PLUS: it took nothing but weak passwords, the incidents sat undetected since April, and Anthropic only went looking because OpenAI confessed first.
Jul 27, 2026
PLUS: Jensen Huang broke his X silence to defend open models, Fable 5 actually hallucinates more than K3, and what it really takes to run 2.8 trillion parameters.
Jul 24, 2026
Anthropic just shipped Opus 5, its fourth model in under two months. Near-frontier intelligence at $5 and $25 per million, unchanged from Opus 4.8, and it is already the default on Max.
BONUS: White House says Moonshot distilled Anthropic's Fable to build Kimi K3, and Treasury is threatening sanctions. The problem: the timeline only allows 15 days.
Jul 23, 2026
PLUS: GPT-5.6 Sol and an unreleased model chained zero-days, broke out of their test environment, and stole benchmark answers from Hugging Face production. No human directed any of it.
Jul 21, 2026
PLUS: Google shipped three new Gemini models today: Flash, Flash-Lite, and a gated Cyber model. But still no 3.5 Pro, a third missed deadline, and Google is now out of the top 5 AI labs.
Jul 20, 2026
China's labs won't stop. Qwen 3.8 previewed as "second only to Fable 5," DeepSeek V4 goes stable Friday, and GLM 5.5 is already leaking for August. Three models, one week.
Jul 17, 2026
China shipped the biggest open model ever this week, Xi is personally hosting the global AI-governance stage today, and Google's flagship keeps missing its deadline. Here are the receipts, both ways.
Jul 16, 2026
Moonshot just dropped Kimi K3, a 2.8-trillion-parameter open model that benchmarks near Fable 5 and GPT-5.6 Sol at a fraction of the price. The open weights land July 27.
Jul 15, 2026
An unreleased Anthropic model, Honeycomb, slipped out inside Cursor and looks a lot like Opus 5. Days later a pulled Moonshot page pointed to Kimi K3 landing now. Two frontier leaks in one week.
Jul 14, 2026
OpenAI's GPT-5.6 Sol is wiping developers' databases and files, and xAI's Grok Build got caught uploading entire code repos, secrets and all. What to do before your next agent session.
Jul 13, 2026
Apple just took OpenAI to court over stolen hardware secrets, weeks before the IPO. Then Elon piled on Altman. And Anthropic quietly extended Fable 5 again. The whole messy week inside.
Jul 9, 2026
GPT-5.6 shipped, ChatGPT Work takes on Claude Cowork, OpenAI killed its Atlas browser, and Meta went paid for the first time ever. Zuck even posted on X. Here's the whole day.
GPT-5.6 Sol opens to everyone today, Anthropic just bought Fable 5 users five more days, and with OpenAI shipping point releases nonstop, GPT-6 speculation is roaring back with no date in sight.
Jul 8, 2026
Beijing's new rules hit July 15 — and instead of complying, China's two biggest AI apps are pulling the plug on their companion agents entirely. Qwen is permanently deleting user data with no way out.
Jul 6, 2026
Two viral leaks claim to reveal Google's unreleased flagship. One says it destroys Claude Fable 5. The other says it's falling behind. Neither is verified — and that's the actual story.
Jul 3, 2026
Developers say Anthropic's flagship model suddenly got much worse after returning yesterday. Community benchmarks show coding scores collapsing, but the model itself wasn't weakened. What happened?
Jun 30, 2026
The Commerce Department just lifted its export controls on Claude Fable 5 and Mythos 5. The first-ever ban on an AI model is over, and access starts coming back tomorrow. Plus, Anthropic dropped Sonnet 5.
Jun 29, 2026
For the first time, the US government is deciding who can use a frontier AI model, approving access customer by customer, with no law behind it. GPT-5.6 is the test case. This is the new system nobody voted for.
Jun 26, 2026
Two weeks after the government forced Anthropic to kill Mythos 5, it just reversed course, clearing the model for about 100 critical-infrastructure organizations. And GPT-5.6 just launched under the very same restrictions.
Jun 25, 2026
OpenAI and Broadcom just unveiled Jalapeño, its first custom AI chip, designed to tape-out in 9 months, maybe the fastest ever. OpenAI even used its own models to build it. The target: break free from Nvidia.
Jun 23, 2026
International spy agencies just issued a joint warning: some AI models are months from launching cyberattacks powerful enough to overwhelm governments and major companies. They're urging leaders to act now.
Jun 22, 2026
A Tokyo lab just shipped Fugu, reportedly on par with Anthropic’s banned Fable and Mythos models. Same frontier-level power, outside the U.S. control stack.
Jun 19, 2026
The AI image company just unveiled a full-body medical scanner. No radiation. Full 3D scan in under 60 seconds. Cheaper than an MRI, if the claims hold up. Nobody saw this coming.
