AI News Roundup: November 13, 2026 - Agents in the OS, a Spreadsheet With a Voice, and a Price Tag That Doubles
A week where AI stopped being a tab and started being the interface: Business Agents inside ChatGPT, an Anthropic office suite, Qwen inside Apple, and Gemini 4 Carbon's price set to climb.
The week of November 6 to 13 was less about a bigger model and more about AI moving into the surfaces people already use. OpenAI began testing Business Agents inside ChatGPT on November 10 and 11 - verified brand agents that can compare prices and complete purchases, with a dedicated profile and aggregated insight for the business. Anthropic, the same week, shipped an office suite: a meeting assistant, data dashboards, and a spreadsheet tool called Claude Sheets, pulling notes, sheets, and whiteboards into the chat. The model stopped being a chatbot and started being the front door.
On the model front, Google opened Gemini 4 Carbon to paid developers on November 10 at $2 per million input tokens, claiming the top spot on the refreshed Artificial Analysis Intelligence Index. Gemini Omni 1.2 Flash went generally available November 5 with 15-second video, conversational editing, and a 30 percent price cut to $0.07 per second. But the line that should worry budgets: Google said Gemini 4 Carbon's price doubles on November 30. The frontier is getting cheaper and more expensive in the same breath.
Anthropic's IPO roadshow continued through the week after its S-1 showed $16.3 billion in first-half revenue. The roadshow set a $70 to $78 range and a Nasdaq ticker, ANT. The signal for builders isn't the valuation - it's that a leading lab is now a public utility bet, which makes depending on it feel safer and switching from it feel harder at the same time. Meta's Muse Ember, at $0.10 per million tokens for high-volume agent work, is the counterweight: the cheap model for the boring loop.
Open weights kept creeping into the mainstream. MiniMax open-sourced H3-Image with 4K generation in seconds, and Pika published weights for H3 Flex, its six-step video-and-audio model. Meanwhile a Menlo Ventures report found 89 percent of US enterprise open-model spend went to Chinese models - Kimi at 61 percent, DeepSeek at 15 percent, GLM at 11 percent - a reminder that the model map and the political map no longer line up.
The throughline: AI is being woven into the OS, the spreadsheet, and the phone. That's good for reach and bad for anyone who wanted one clean vendor. My read is that the interesting question this quarter isn't 'which model is best' but 'which surface owns my workflow' - and the answer is being decided this month.
Frequently asked questions:
Q: Are ChatGPT Business Agents live for everyone?
A: They were in testing from November 10-11, starting with ecommerce and online-service brands, not a full public launch.
Q: Does Gemini 4 Carbon really get more expensive?
A: Google indicated the price doubles on November 30, so budget for it if you depend on that model.
Q: Is Anthropic's IPO finalized?
A: The roadshow launched with a price range and ticker; the listing itself comes after.