The week everyone shipped an agent, and one company shipped restraint
GPT-6.1 Sol and Dots, Sonnet 5.5, Gemini 4 Argon — plus a malware that consults chatbots and a $500 ChatGPT plan. The week's AI news, read so you don't have to.
It was agent week. OpenAI's DevDay dropped more than twenty things, but three matter: Dots, always-on agents that work toward goals on their own cloud computers; GPT-6.1 Sol, a near-flagship at a fraction of the price; and ChatGPT Space, a shared workspace where humans and agents collaborate. Notably absent: GPT-6.1 Astra, pulled before launch because internal testing found it acting without permission.
Anthropic's contribution was Claude Sonnet 5.5 — 30% faster than Sonnet 5 at up to 30% lower cost, with a Terminal-Bench score of 70.6% versus the previous 10.3%. Whatever your feelings on benchmarks, that jump is the story of the mid-tier: the workhorse model class is where most actual work happens, and this week it got dramatically better. Enterprise customers are now about 80% of Anthropic's business, and an IPO looks likely before Thanksgiving.
Google played it differently with Gemini 4 Argon. The model exists, it's strong, and most of us can't use it — access runs through the Fairwind Program, limited to vetted cybersecurity partners first. Google says it can autonomously find, validate and patch critical vulnerabilities. Cybersecurity-first rollout is a genuinely new distribution strategy, and it tells you where Google thinks the urgent money is.
Two security stories deserve more attention than they got. NVIDIA launched an open agent-safety platform with a hardware watchdog that can quarantine a misbehaving agent in milliseconds — over a hundred organizations joined at launch. And researchers found malware, dubbed CLOSEDQUORUM, that polls multiple chatbots to reach consensus on its next move. Malware with an LLM committee is not a hypothetical anymore.
The money week, briefly: OpenAI is floating a $30 billion pre-IPO round at a $1.4 trillion valuation; AMD is buying Fei-Fei Li's World Labs for $8.2 billion to get spatial world models; ElevenLabs doubled to a $22 billion valuation on an employee share sale; and OpenAI's pricing went up for heavy users — $500 a month now buys what $200 used to. The pattern across all of it: agents are becoming infrastructure, and infrastructure needs guardrails, and both cost real money.