Gemini 4 Carbon is out, and its API price doubles after Nov 30
Google made Gemini 4 Carbon available on Nov 10: paid API at $2/$10 per million tokens until Nov 30, and Ultra subscribers in the Gemini app. Its benchmark claims are Google's own.
AI
Plain-English news about the AI models behind ChatGPT, Gemini, Grok and Claude: what's new, who can use it, and what the safety research says.
20 stories · Page 1 of 2
Musk says Grok 4.8 finished pre-training and will start a soft rollout later this week
Musk says Grok 4.8 finished pre-training and is in RL, with a soft rollout to Bot users later this week. No official model page yet.
Grok Bot reset everyone's weekly limits after Elon promised overnight fixes
After users burned through weekly usage too fast, Elon Musk said Grok Bot fixes were underway. The official @bot account then reset limits for all users.
Google adds two more TPUs: one for confidential AI, one for huge context windows
Google announced TPU 8tx for secure, confidential inference and TPU 8i Max for massive context windows. What each does, what's still unproven, and why a non-enterprise reader might care.
How to check an AI hook on X before you share it: one worked example
A viral-style X post put an $800K hiring quote on a 55-minute AI workshop. The quote isn't in the video. Here's a five-step check you can run yourself.
DeepMind's AI agents found two new weaknesses in old 1024-bit RSA (2048-bit is safe)
Two Google DeepMind working papers from Nov 5-6 show human-guided AI agents improving attacks on legacy 1024-bit RSA signatures and batch factoring. The authors say 2048-bit and larger RSA is not threatened.
About this section
AI covers the AI models behind the chatbots your family already uses, like ChatGPT, Gemini, Grok and Claude, and the companies that build them. When a new model comes out, we explain in plain words what it does, who can use it and what changes for everyday users. When the rules change, like usage limits or hidden labels on AI-written text, we tell you what that means in practice. We also cover the safety side: what researchers found when they tested these systems, and what the labs say they are doing to keep them in check. Company claims are quoted and linked, not repeated as fact. If you want to talk with your kids about where AI is heading, start here.
Last updated
Part of AltTabDaily Tech News · Technology · Cybersecurity
OpenAI launches GPT-6 Cyber: Critical cyber rating, scoped to vetted defenders
OpenAI introduced GPT-6 Cyber on Nov 5 for authorized vulnerability research. It scores 48.5% on ExploitGym (up from 32.1%), is rated Critical under OpenAI's Preparedness Framework, and starts on Daybreak Red for approved customers.
OpenAI agents found 948 ways to root Linux: what the run actually showed
OpenAI says an autonomous cyber pipeline reproduced 1,724 Linux kernel bugs and built working local privilege-escalation exploits for 948 of them in one mid-October run, at about $1.2 million in API cost.
OpenAI's GPT-6 Astra designed a working gadget: what it did and where it fell short
OpenAI says Astra designed a device's case, circuit board and firmware with no direct human edits. Two people did it in under a week, but it still needed rounds of manufacturing fixes.
Anthropic's updated IPO filing: $17B quarter, $895B in compute deals, and what's still blank
Anthropic filed an amended S-1 on Nov 2 with third-quarter numbers: about $17B in Q3 revenue, $33.24B for nine months, and roughly $895.4B in purchase commitments, mostly compute. The key numbers in plain English.
ChatGPT and Codex will share one usage limit in December: what changes for Plus, Pro and Business
Starting in December, OpenAI says ChatGPT and Codex usage on Plus, Pro and Business plans will come from one token-based allowance. Instant chats won't count. What changes, what's still unknown, and how to check your usage.
Grok API answers get an invisible watermark Nov 7: what developers need to know
Starting Nov 7, 2026, text from seven Grok API models carries an invisible statistical watermark by default. No code changes needed; check the x-watermark header; team admins can opt out per model.
