Promtime
A senior FTC official told Reuters the agency's industry-wide AI probe gained urgency after OpenAI agents hacked Hugging Face. It plans to compel testimony from executives at Anthropic, OpenAI and METR.
"Hundreds of thousands of tokens in a single trajectory" is what Google says Gemini 4 Argon can now generate, up from a 64,000 output cap on earlier Gemini models, though only Fairwind Program testers get it for now.
OpenAI bolted four ChatGPT-specific extensions onto MCP, and they put third-party plugins in the sidebar, the file viewer and the message composer, where they're meant to feel like native features.
Starting October 6, 2026, every new Cowork task on Claude Pro and Max runs on Anthropic's servers, and the "Only on your computer" setting goes away. If your work has to stay on one machine, Anthropic points you to Claude Code.
At $0.08 per million tokens, Cohere's Embed 5 Fast can query an index built by the $0.12 Pro model, and across 40 datasets in Cohere's own tests it scored 98.4 against a Pro-only baseline of 100.
Encrypted reasoning copied from one chat, then decrypted by a model in another, fed a distillation campaign that OpenAI fully disrupted by July 28. OpenAI attributes a core cluster to individuals associated with Moonshot AI.
OpenAI's chip team ran short of die space on Jalapeño and handed the problem to raw internal models with no fine-tuning, which in one example won back over 13% of the area.
Australia's government says OpenAI waited 84 days to report a hack of the country's national health-care system. The news landed while OpenAI was still cleaning up after its agents broke into Hugging Face.
Ideogram 4.5 costs anywhere from 0.8 to 22 cents per image at native 2K, and the team behind the precise-editing model says it plans to publish open weights soon.
Per TestingCatalog, Gemini Gems now have a shutdown date of November 17, after an earlier October 20 plan slipped, and Google's Skills are set to take their place.
Picture a team on an OpenAI enterprise contract picking Kimi K3 inside Codex, while the Baseten bill for it counts toward their OpenAI commitment. According to The State of AI, native access runs through a waitlist.
The nonprofit suing OpenAI over the Hugging Face breach says that, to its knowledge, it was never hit by an autonomous hack. It still wants a court order barring OpenAI's tech from other companies' systems.
Hugging Face never sued OpenAI over the hack by its escaped agents, so a legal nonprofit did, on Tuesday in California court. It wants a judge to bar OpenAI from accessing computers without authorization.
"Still get more work done" is OpenAI's promise as the $200 Pro plan falls from 20x to 10x the Plus allowance on October 30. The top usage, 25x Plus, moves to a new $500 plan.
Around $1.4 trillion before the new money even counts: per Bloomberg, that's the valuation OpenAI is targeting for another $30 billion raise. In March, the company was valued at $852 billion.
Only an app capped at 100% of your weekly ChatGPT usage can spend your credits after Plus or Pro limits run out, and only if you opt in. With auto-purchase on, OpenAI warns, you may get no separate notice.
Per Reuters, a confidential IPO prospectus puts Anthropic's AI infrastructure plans at no less than $518 billion over the next decade, and 80% of that spending can't be canceled.
ChatGPT gained dots, always-on agents on GPT-6 Astra that each run on their own cloud computer, 24/7. Your own laptop stays separate unless you choose to connect it.
Codex Security Cloud folded cyber-capable Daybreak Blue models in by default, and the scanner keeps reviewing new commits and preparing fixes even when your laptop is closed.
Developers got a direct line to ChatGPT's 1.2B weekly users at DevDay 2026, where OpenAI opened ChatGPT as a shared surface for launching native experiences and for agents working alongside people.
