ANT's first day as a public company ended 17% above its IPO price after fading from a 34% opening pop. House Democrats are probing whether Nvidia swayed the Commerce Department's chip-smuggling inquiry into Megaspeed, Google opened its Genie 4 world model to developers through the Gemini API, and OpenAI shipped GPT-Live-1.1, a full-duplex voice model that listens while it speaks.
The print
1ANT's first day closes 17% above the IPO price - a volatile fade from a 34% opening pop
Anthropic shares closed their first trading day at $97.17, up 17% on the $83 IPO price, after opening at $111.50 and giving back much of that 34% pop through the session, Nikkei and AP report. The stock at one point pushed Anthropic's value past $2 trillion before settling below that mark; at the closing price, each co-founder's stake is worth between $23.3 billion and $23.7 billion, by the BBC's arithmetic. It is the first stand-alone frontier AI lab to list in the US, and its offering - the largest IPO on record - is expected to close today, Friday.
In plain English
The biggest IPO in history survived day one - up solidly, but well off its morning highs.
Why it matters
A 17% first-day gain on an $81 billion offering is a clear demand verdict for frontier AI, but the fade from the open shows how nervous this market is; day two's trade, with the offering closing, sets the reference price every private lab behind Anthropic will be measured against.
Representatives Sydney Kamlager-Dove and Gregory Meeks of the House Foreign Affairs Committee sent a letter Thursday to Commerce Secretary Howard Lutnick requesting records of communications between Nvidia and the office that handles chip-smuggling investigations, Bloomberg reports. The letter asks whether Bureau of Industry and Security head Jeffrey Kessler authorized Nvidia general counsel Tim Teter to contact BIS investigators working the Megaspeed case, and why the office paused its inquiry and left the Justice Department to lead it. Megaspeed is the Singapore cloud firm where Alice Huang was an executive; Commerce and Justice said in a joint statement Monday that Commerce began investigating Megaspeed for potential export-control violations in 2024, conducted an on-site check with Singaporean authorities in 2025, and interviewed Alice Huang, with Justice taking the lead after that enforcement action. The lawmakers want written answers and documents by December 1 - the day Jensen Huang is set to brief three House committees behind closed doors. Teter has separately agreed to a closed-door briefing on December 2, and the House Energy and Commerce Committee this week asked interim CEO Colette Kress for a briefing on Nvidia's export-control compliance and its Megaspeed sales. Megaspeed has denied wrongdoing.
In plain English
Congress is now formally asking whether Nvidia reached into a federal investigation tied to the scandal that ousted its own CEO.
Why it matters
Democrats take control of Congress in January and Meeks is in line to chair Foreign Affairs - the letter previews sustained scrutiny of the administration's relationship with the AI industry's central chip supplier and its China sales. Watch the December 1 deadline and the Huang and Teter briefings.
3Google opens Genie 4, its world model, to developers through the Gemini API
One day after launching Genie 4 to consumers in Project Genie, Google made the world-simulation model available to developers through the Gemini API and Google AI Studio. Google pitches it for prototyping interactive experiences, visualizing spatial designs, and training robotics and AI agents; every output carries an imperceptible SynthID watermark. Starter apps and an interactive canvas are live in AI Studio, with usage billed under Gemini API pricing.
In plain English
Google's world simulator went from a demo to an API any developer can call.
Why it matters
World models are the industry's next platform bet for agents and robotics; API access turns Genie 4 into infrastructure and puts Google in direct competition for the developers every lab is courting this quarter.
4OpenAI ships GPT-Live-1.1: a full-duplex voice model that listens while it speaks
OpenAI rolled out GPT-Live-1.1 in its API this week, an upgrade to its voice model that can listen while it speaks - full-duplex conversation rather than turn-taking. OpenAI's documentation says it transcribes speech more accurately and holds a more consistent accent and voice across languages than GPT-Live-1, and developers can describe in plain language when the model should hand a task to their backend for reasoning or tool use. Voice sessions are billed by duration: five cents per active minute and two cents per standby minute, a new lower rate that applies while inference is paused.
In plain English
OpenAI's voice model learned to listen and talk at the same time - and got a cheap idle mode for always-on assistants.
Why it matters
Full-duplex listening is the difference between a walkie-talkie and a phone call; paired with standby pricing, OpenAI is pricing voice agents to run all day, which squeezes every standalone voice-AI startup.
Checked against original announcements and named press reports on the morning of November 20, 2026. ANT's opening and closing prices are as reported by Nikkei and AP; the founders' stake values are the BBC's arithmetic at the closing price, and share prices move. The House Democrats' letter and the contacts it examines - between Nvidia's general counsel and Commerce officials - are as reported by Bloomberg and remain allegations under congressional review, not established findings; Megaspeed denies wrongdoing. Genie 4's availability and features are Google's own announcement. GPT-Live-1.1's capabilities and pricing are from OpenAI's documentation.