Fugu Ultra: The AI That Conducts Other AI Models
Sakana AI's Fugu Ultra v2 isn't a new flagship model — it's a system that routes tasks to other AI models, and it beats much bigger names on one benchmark doing it.
Read more →Sakana AI's Fugu Ultra v2 isn't a new flagship model — it's a system that routes tasks to other AI models, and it beats much bigger names on one benchmark doing it.
Read more →The Apple Watch Series 12 listens in on conversations and summarizes them — without recording any audio, and with solid encryption. For now, you can only use it outside the EU.
Read more →Three articles from this blog got an update — one of them an obituary. Flowise is history, MiniMax is getting pricier, and OpenRouter is building a European door.
Read more →Fable 5.1 doubles a key coding benchmark and makes caching notably cheaper. The people who actually benefit are mostly the ones already paying.
Read more →GPT-6 Astra is OpenAI's newest and most capable model: it can use software like a human would, but it costs real money and carries a serious new security risk.
Read more →Nvidia is reportedly buying Hugging Face for roughly $13 billion. The internet's biggest hub for open AI models would suddenly belong to a chip company.
Read more →A mystery AI model quietly dominated the leaderboards for weeks. Now we know who built it, and how they pulled it off.
Read more →Since August 18, ChatGPT for 13- to 17-year-olds runs in a protected mode by default. Age is estimated, not verified — with all the weaknesses that implies.
Read more →Google's AI agent Gemini Spark sorts your inbox, books appointments and works your browser — except if you live in the EU. What it does, what it doesn't, and why Europe is missing.
Read more →OpenAI gives free users unlimited text chats with its new GPT-5.6 Luna model. Images and file uploads, though, remain limited.
Read more →Alibaba celebrates Qwen3.8-Max as an open giant model. The download really is free — just not quite what the announcement promised.
Read more →Claude Code will stop asking permission for every command starting August 14 — a classifier handles the safety check instead. The feature isn't new; the default setting is. A look at the numbers, the limits, and whether that trust is earned.
Read more →On-device AI finally works in everyday phones: Apple's chip and Qualcomm's Snapdragon process AI requests locally. Here's what's really behind it – no hype.
Read more →ChatGPT gets its own health feature with Apple Health integration and a medical records dashboard – but only in the US so far. Here's why Europe is excluded.
Read more →Anthropic has released Claude Opus 5 — its strongest model yet, with a new effort dial, but available only to paying Pro and Max subscribers.
Read more →FLUX 3 from German company Black Forest Labs generates images, videos with sound, and robot actions from a single model — but for now only through a waitlist and with no announced price.
Read more →Gemini 3.6 Flash is here – more efficient than its predecessor, free to use in the Gemini app and AI Studio, and already your new default model.
Read more →A new Chinese model sparks such a rush that its maker shuts the door. We explain what Kimi K3 can do, how to get in anyway — and whether switching is worth it.
Read more →Talk instead of type: GPT-Live and Gemini Live make chatting with AI surprisingly natural. What's free, where it stumbles — and who it's worth it for.
Read more →Anthropic is giving verified US teachers free access to Claude — with its own privacy package built for the classroom. What's in it, and why the rest of the world is still left out.
Read more →After three weeks offline, Claude Fable 5 is back worldwide with a new safety filter — but free-tier users still get none of it.
Read more →OpenAI's GPT-5.6 lineup brings three models: Sol for top performance, Terra for balance, Luna for speed. Most ChatGPT users can't try them yet — but the rollout is coming in weeks.
Read more →Elon Musk's rocket company is taking over AI coding assistant Cursor — the largest venture-backed startup acquisition in history. Here's what that means for you.
Read more →Gemini CLI was a beloved open-source AI tool with over 100,000 GitHub stars — now Google is pulling the plug. The replacement, Antigravity, has worse quotas and no open code.
Read more →Fable 5 is Anthropic's most powerful publicly available model. Free for paid users through June 22 — after that, usage credits apply.
Read more →At WWDC 2026, Apple unveiled a Gemini-powered Siri that finally works like a real AI assistant. Here’s what changed and what you can try right now.
Read more →OpenAI rebuilt ChatGPT's memory from the ground up. Dreaming V3 reads your old conversations and builds a profile of you automatically — no setup required, free tier rollout underway.
Read more →At Build 2026 Microsoft unveiled seven in-house AI models and began moving away from OpenAI. What's behind MAI-Thinking-1 — and what does it mean for you?
Read more →ChatGPT now shows ads — but only for free users and Go subscribers. What you'll see, what happens to your data, and what options you have.
Read more →Apple is breaking ChatGPT's exclusivity in iOS 27. A genuine improvement for iPhone users — and a feature Android has had all along.
Read more →On May 20, 2026, an OpenAI model autonomously solved a major open math problem for the first time. Here's what happened, what it means, and where AI still hits its limits.
Read more →Magnifica Humanitas: The Pope writes about AI. No hype, no bashing — just some points that actually land.
Read more →Since May 2026, all images from ChatGPT carry a hidden watermark. Anyone can check — but how reliable is it really?
Read more →