Today’s through-line is the fight over the search entry point, the bundling of AI with basic tools in education, and financing for research agents and inference infrastructure: OpenAI publicly said ChatGPT search-like query volume is about one-fifth of Google’s; Google switched campus Gemini and related capabilities to off by default while placing Google Search in the same control group; DeepMind showed cheminformatics skills on Antigravity; and Fal closed a new round at a valuation of about $16 billion.

1. OpenAI: ChatGPT search-like queries reach about one-fifth of Google’s

According to The Verge (Hayden Field, about 18:30 UTC on November 3, 2026) and Big Technology (Alex Kantrowitz), Adam Fry, who leads ChatGPT consumer products at OpenAI, said at a Monday-evening media event in New York that “search-like” query volume on ChatGPT is now roughly equivalent to about one-fifth (about 20%) of Google Search query volume. Fry defined “search” as cases in which a user enters a question and actually triggers web retrieval, excluding simple paraphrases or basic how-to instructions; he did not specify the source or definition of the Google-side figure.

Other scale figures disclosed at the same event included: ChatGPT weekly active users rising from about 1 billion to about 1.2 billion in roughly 60 days; voice-related weekly active users (including dictation and voice conversation) totaling about 250 million; and about 1.5 billion images generated per week. Fry also said advertising “has not had a negative impact on the product or its usage,” and that most people do not seem to mind ads in the product. The Verge noted that even at about one-fifth of Google’s scale, ChatGPT remains far from “replacement”; the piece cited Microsoft CEO Satya Nadella’s 2023 antitrust-trial remark that Bing held about a 3% search share for contrast. The comparison is OpenAI’s own query-volume framing, not an independently audited market share, and is not the same as referral share to third-party websites.

Takeaway: The company is quantifying the “Ask Chat” habit as a search-volume narrative comparable to Google’s, while answering growth costs with a commercialization path that runs ads alongside subscriptions.

Sources: The Verge; Big Technology

2. Google for Education: AI off by default, but turning AI off also means losing Search

Chris Phillips, vice president and general manager of Google’s education business, wrote on November 3 that the company will change K-12 student-account controls: Gemini, Gemini Notebook, Gemini in Classroom, and Google Search will all be off for K-12 student accounts until an administrator explicitly enables them; schools will receive specifics in the coming months before the change takes effect. Guided Learning, launched in parallel for Search and Classroom, uses step-by-step questions to prompt student thinking when it detects certain homework-help intents, rather than giving a final answer outright; Guided Learning in Classroom is available globally, while the Search version will first pilot in selected U.S. schools.

Multiple outlets flagged the bundling logic: Ars Technica (Ryan Whitwam) and AP among others reported that the products sit in the same control group, so schools cannot keep Search while turning AI off. Phillips told The New York Times that “search and AI are deeply integrated”; if a school chooses not to enable AI, students “will not be able to use Search” and will see a school notice that it is unavailable. Once enabled, schools can choose between the standard Gemini path and the Guided Learning path. Google said the changes will roll out over the coming months to give students and administrators time to prepare.

Takeaway: “Off by default” answers campus AI controversy, but tying Search to generative capabilities on the same switch effectively turns “whether to allow AI” into “whether to keep using Google Search.”

Sources: Google Blog; Ars Technica; AP News

3. DeepMind: Antigravity cheminformatics skills complete a drug-discovery iteration in about an hour

Google DeepMind published a blog on November 3 introducing Cheminformatics Skills for Google Antigravity. Using lung-cancer-related tyrosine kinase inhibitors crizotinib and lorlatinib as examples, the piece shows an agent retrieving thousands of compounds, running matched molecular pair analysis, and proposing novel rigidified macrocycle candidates with synthetic routes; the full demo took about an hour, whereas comparable human iteration often takes days.

The skills stress a division of labor: the language model handles workflow orchestration, database retrieval and scientific reasoning, while deterministic chemistry tools (built on the open-source toolkit RDKit and connected to ChEMBL, PubChem and others) perform molecular computation. The article also shows an interactive compound browser, potency-prediction models and reaction-mechanism (electron-pushing) diagrams, and stresses that computational designs remain hypotheses pending experimental validation; sparse data and inter-lab measurement noise limit precision. The skills are available in Antigravity via the Science plugin; enterprise users can sign in with a Google Cloud account; DeepMind also said it will offer up to 10,000 scientists six months of free access at the highest usage tier (rolling selection, subject to eligibility and capacity).

Takeaway: Packaging “retrieve—analyze—design—propose synthesis” into an agent workbench moves research agents from general chat toward reproducible domain toolchains.

Sources: Google DeepMind

4. Fal raises about $1.3 billion at a valuation of about $16 billion

According to TechCrunch (Marina Temkin, about 11:34 a.m. Pacific Time on November 3, 2026), San Francisco–based inference company Fal, which hosts and runs AI video, image and audio models, closed a latest round of about $1.3 billion led by Bessemer Venture Partners, with participation from Nvidia, Sound Ventures, T. Rowe Price and others, at a valuation of about $16 billion—roughly double the about $8 billion valuation in April this year. Bloomberg also reported the valuation. The company disclosed Series E and Series F together, saying the two rounds raised about $1.65 billion combined (an earlier round led by Sequoia).

Coverage said Fal’s annualized revenue earlier this year already exceeded about $800 million; the round comes amid surging demand for open-source model inference. Peers such as Baseten, Fireworks AI and Modal are also expanding quickly and raising at high valuations, but people familiar with the matter said inference companies’ gross margins remain pressured by GPU purchase and rental costs. Fal was co-founded in 2021 by former Coinbase machine-learning lead Burkay Gur and former Amazon developer Gorkem Yurtseven.

Takeaway: Capital keeps doubling down on the inference layer of “how to run models after training,” and generative-media inference is becoming a financing through-line alongside large-model training.

Sources: TechCrunch

---

One-line takeaway: OpenAI says ChatGPT search-like query volume is about one-fifth of Google’s, Google answers campus AI controversy with defaults off while bundling Search with AI, DeepMind writes a drug-discovery workflow into Antigravity, and Fal raises about $1.3 billion at a valuation of about $16 billion.