Google shipped a new frontier model with little warning, and it is the first one built to write book-length output in a single pass. Also today: who is already using it, a voluntary AI pledge with an embarrassing typo, and the revenue math behind the industry's spending.
7 Stocks You May Never Need to Sell
Buying a stock is easy. Finding one you may never need to sell is harder.
This free report reveals seven blue-chip companies selected for financial strength, durable demand and long-term potential to hold up when markets turn ugly.
TODAY IN AI
3 stories worth your next 90 seconds
1. Google releases Gemini 4 Argon, a model with a 1 million token output limit.
Argon can write up to 1 million tokens in one response, up from 64,000 in the previous model, which is roughly the length of several novels. Google reports 77.9% on DeepSWE v1.1, a coding test, and 51.3% on AutomationBench, where it says Argon ranks first. Those numbers come from Google and have not been independently checked. Access starts with a small group of cyber defenders, then widens to paid API customers and Google AI Ultra subscribers. Read Google's announcement.
2. Wiz scans public services for security holes with Google DeepMind.
The Scan for Good project looks for serious exposures in public services and critical infrastructure, then helps the owners fix them. Its page reports 17,761 domains and 326,891 public endpoints checked, with 475 critical exposures found so far. Wiz's page names Gemini 3.8 Flash Cyber as the model, while Google's launch post lists Wiz as an Argon partner, so the two tools may overlap. See the Scan for Good page.
3. Tech leaders sign a voluntary pledge on frontier AI responsibility.
The Joint Commitment on Frontier Responsibilities was signed by President Trump alongside leaders including Mark Zuckerberg, Jensen Huang, Dario Amodei and Sam Altman. The companies promise that their technology behaves as intended, that problems are found and fixed quickly, and that they keep independent board oversight and internal controls. TechCrunch reports that it carries no legal or regulatory consequences if a company breaks it. Read TechCrunch's report.
Blu Dot surpasses 2,000% ROAS with self-serve CTV ads
Blu Dot used Roku Ads Manager to drive incredible results for its furniture sales event. Its strategy hinged on custom audiences and retargeting, where intent was strongest.
“Roku has been a top performer,” said Blu Dot’s Claire Folkestad. “We have seen…CPMs lower than any other CTV partner we've worked with.”
FROM THE FRONTIER
Revenue is racing. The bill is racing faster.
The surge. OpenAI's annualized revenue is approaching $70 billion, Quartz reports, up more than 70% since the start of the third quarter, with business sales more than doubling over that stretch.
The hole. Leaked audited financials showed OpenAI spent $34 billion last year against $13 billion in revenue, with losses growing year over year. Fast growth and deep losses are running side by side.
The math. Bain & Co. estimates the data center buildout needs $6 trillion in annual revenue by 2031 to pay off. Existing AI services account for $1.8 trillion of that, which leaves $4.2 trillion still to be created.
The reframe. Nobody is arguing anymore about whether people will pay for AI. The open question is whether enough of them will pay, soon enough, to cover what is being built. A 1 million token model is impressive, but every extra token is also a cost someone has to recover.
You're Running Three Databases. You Only Need One.
Events in one database, embeddings in another, analytics in a third. That's three systems and the pipelines between them, for one app. TimescaleDB does all three in Postgres: hypertables, pgvector, continuous aggregates. One system, no pipeline, no drift.
IN THE KNOW
What people are actually watching and sharing
A typo at the signing. The signed pledge shared on Truth Social showed "Unites States" instead of "United States" beneath the President's signature, and people noticed quickly.
Defenders first. Google's Fairwind Program launched on September 2 with more than 650 partners. It pairs Gemini 3.8 Flash Cyber with CodeMender to produce patches in minutes, and members must limit access to internal security teams and use protections like multi-factor authentication.
Price check. Argon costs $2 per million input tokens and $10 per million output tokens at the introductory rate, rising to $4 and $20 afterward. Cached input tokens are priced 95% below the input rate.
Beyond code. Google also reports 91.7% on LVBench, a long-video test, and 68% on CWE-bench v1, where it says Argon is tied for first. Again, these are Google's own figures.
PROMPT STATION
Run a mock interview that scores you
Most interview prep stops at reading questions. This prompt works in Claude, ChatGPT or Gemini and plays the interviewer for real: one question at a time, a score for each answer and a Most interview prep stops at reading questions. This prompt works in Claude, ChatGPT or Gemini and plays the interviewer for real: one question at a time, a score for each answer and a stronger version of what you said. Run it before your next interview, performance review or pitch.
Act as a tough but fair interviewer hiring for this role: [JOB TITLE] at [COMPANY OR TYPE OF COMPANY].
MY BACKGROUND: [TWO OR THREE LINES ABOUT YOUR EXPERIENCE]
Run a mock interview with five questions. Ask one question at a time and wait for my answer before moving on. Mix behavioral, situational and role-specific questions, and make at least one of them uncomfortable.
After each answer, give me a score out of 10, one thing I did well, one thing that weakened the answer, and a stronger version of my answer in my own voice.
After the fifth question, summarize my three biggest patterns to fix and give me a one-line opening answer for "Tell me about yourself".
Do not be encouraging by default. If an answer is vague, say so.