ANG⁺ #6 · Anthropic's IPO Price Range, +5 More
Anthropic set an IPO range of $70–$78 a share. Plus Gemini 4 Carbon for everyone, a decision API that bills zero output tokens, business profiles inside ChatGPT, new freelance research, and dates for your calendar.
ANG⁺ #6 · Anthropic's IPO Price Range, +5 More
about 4 min
ANG⁺ = briefings sorted by AI and checked by the editor
Welcome to the sixth ANG⁺ Briefing. Briefing #5 (November 10) covered the price and terms of the top models. This edition is about where money and work are moving. On November 10–11, Anthropic put a price range on its IPO, Google opened to everyone the model it gave defenders first last month, and OpenAI shipped news for businesses and freelancers. Everything is built on first-party material: official company announcements and developer changelogs.
1. Follow-up: Anthropic launches its IPO roadshow — price range $70.00–$78.00
This is the next step after the IPO filing covered in #3 (filed October 26). On November 10 (U.S. time), Anthropic launched its roadshow, the series of investor presentations, and put numbers on the offering for the first time.
- Shares: The offering is 825,000,000 shares of Class A common stock. Of these, 801,351,354 are offered by Anthropic and 23,648,646 by existing stockholders. Anthropic receives no proceeds from the shares sold by existing stockholders.
- Price: Expected between
$70.00and$78.00per share. The underwriters get a 30-day option to buy up to 123,750,000 more shares. - Where: The Nasdaq Global Select Market under the symbol "ANT", as covered in #3.
- Not yet: The registration statement has not become effective, and the company says shares cannot be sold, nor offers to buy accepted, before then. The final price and listing date are not in this announcement.
Our take: Even the bottom of the range puts the offering at about$57.75 billion(825,000,000 shares ×$70). Big as that is, what matters more is that a listed company reports results every quarter — readers get a regular window into what an AI company actually earns and spends.
Source: Anthropic launches roadshow for proposed initial public offering (Anthropic)
2. Follow-up: Google's Gemini 4 Carbon goes GA — defenders first, everyone two weeks later
Gemini 4 Carbon, covered in #3, opened to developers, enterprises, and consumers on November 10 (16:00 UTC). Two weeks earlier it went first to cybersecurity defenders through the Fairwind Program.
- Price: The API model is
gemini-4-carbon. Through November 30 it carries an introductory price of$2per million input tokens and$10per million output tokens (the same as Gemini 4 Argon); after that,$4input and$20output. - The company's numbers: #1 on CursorBench 4.0, Perplexity's WANDR, and Zapier's AutomationBench, at more than 2x lower cost per task than the next best external model. On MKA, the multimodal agent benchmark Google released alongside it, Carbon ranks #2 at more than 6x lower average cost per task than the top model.
- Where to use it: The Gemini API paid tier, Gemini for business, and GitHub Copilot (generally available the same day). Google AI Ultra subscribers get it in the Gemini app, AI Mode in Search, and Antigravity, where
/swarm— splitting one prompt into a team of agents — is now in preview. - Use them side by side: Google says Argon is still the right choice for most coding and knowledge work; Carbon is for jobs that need to dig deeper, even at the cost of more tokens.
Our take: Next to Fable 5.2 from #5 ($8input,$40output), the introductory price is a quarter. But it ends on November 30 — if you are running comparison tests now, do the math at December's price ($4/$20).
Source: Gemini 4 Carbon is now available (Google) · Gemini API changelog, November 10, 2026 (Google) · Gemini 4 Carbon in GitHub Copilot (GitHub)
3. Google's Gemini Decisions API — a model that chooses instead of writing
On November 11 (18:00 UTC), Google shipped a new kind of API. The model is Gemini Decision 1 (gemini-decision-1). It does not write answers as text; it scores a set of options you define with probabilities and confidence scores.
- What you can ask: Three types — yes/no, picking among 2 to 255 choices, and grading on an ordered rubric starting at 0. You can send up to 1,000 questions at once, over a single input of up to 256K tokens that can mix text, images, audio, video, and PDFs.
- Speed and price: 150–250 ms end to end, at
$0.10per million input tokens, with zero output tokens billed. On Google's benchmark suite it is nearly 5x cheaper than doing the same with Gemini 3.5 Flash-Lite. - What it's for: Choosing which tool or model to send a task to, checking policy rules, grading answers — decisions with a fixed set of options. Every answer carries a 0–1 confidence score, so code can send only the unclear cases to a person.
- Same direction: On November 10, Hugging Face published a guide to running "decision models" with Transformers and llama.cpp.
Our take: Anyone who has run agents knows a large share of the bill comes from short "what next?" questions. Pull those out of a writing model and both cost and latency drop. If you are building automation, start by sorting each step: is it choosing, or writing?
Source: Gemini Decisions API: Fast, calibrated multimodal decisions (Google) · Gemini API changelog, November 11, 2026 (Google) · Decision Models in Hugging Face (Hugging Face)
4. OpenAI's "Businesses in ChatGPT" beta — a business profile inside ChatGPT
On November 10 (10:00 UTC), OpenAI opened Businesses in ChatGPT in beta, letting businesses manage their own information inside ChatGPT. The company says ChatGPT has 1.2 billion weekly users.
- Claim your business: Once verified in Business Hub, a business can update key details and upload its own product feeds with pricing and availability. Any eligible business where ChatGPT is available can claim; Shopify merchants can connect their store to speed up verification.
- The numbers you see: How many times the business is mentioned and how many website clicks it gets from ChatGPT answers, plus common topics in conversations where its site is cited. Conversations themselves are not shared, and insights appear only in aggregate once enough interactions meet a threshold.
- Business Agent: When someone clicks a business profile in an answer, they can start a separate conversation with the business's own agent. That conversation is labeled and kept apart from ChatGPT's own answers. For now, only U.S. ecommerce and online service businesses can build one, and it can also serve as the destination for ChatGPT Ads.
Our take: This is the first dashboard for the shift from "where do I rank in search?" to "how often am I mentioned in AI answers?" The agent is U.S.-only for now, but claiming a profile and seeing the numbers are not — if you run a business with its own website, check this week whether you can claim it.
Source: Introducing Businesses in ChatGPT (OpenAI)
5. OpenAI × Upwork research — how freelance work is shifting in the AI era
On November 10, OpenAI published "Work at the Frontier: AI's effect on freelance work" with the freelance marketplace Upwork, comparing Upwork job postings with the tasks people do directly in ChatGPT.
- How big the market is: Upwork's research finds 38% of skilled U.S. knowledge workers freelance, and freelance knowledge work generated an estimated
$1.5 trillionin U.S. earnings in 2024. - Demand is moving: In 43 of the 63 work categories compared, work that is common in ChatGPT lost share in Upwork postings and less common work gained share; the other 20 showed the reverse. The company adds that ChatGPT use alone does not explain the shift.
- Work is being combined: From early 2023 to late 2025, the mix of tasks in AI-related postings became 45% more diverse — the equivalent of going from about 14.6 to 21.2 equally common tasks. One posting might ask for AI-generated video, editing, and motion graphics together.
- Budgets: Posted hourly rates were similar, but AI-related projects had larger fixed-price budgets and involved more hours of work.
Our take: A more accurate picture than "AI eliminates work" is "the bundle of work handed to one person is getting bigger." If you freelance, read this as a signal that demand is gathering around delivering two or three skills that rarely went together, in one package.
Source: Work at the Frontier: AI’s effect on freelance work (OpenAI)
6. New this week, and dates for your calendar
Smaller changes from November 10–11, plus the dates within a month when prices or code change.
- NVIDIA donates OpenShell to the CNCF (November 10): Its open source runtime for securing and governing autonomous agents is now a Cloud Native Computing Foundation Sandbox project. NVIDIA also contributes to Agent Substrate, a Google-led framework for long-running agent workloads.
- IBM × Microsoft "Imagine Enterprise" (November 10): An agent platform aimed at regulated industries. IBM Consulting brings more than 6,000 prebuilt AI agents and assets, and engineers from both companies work on site with customers.
- Cursor Projects goes multiplayer (November 10): Projects, launched in September (one coordinator agent directing many agents), now lets you invite teammates to share a task board, context, and documents.
- Google's MKA benchmark for multimodal agents (November 10): Built with Artificial Analysis, with experts from 10 verticals and 83 occupations writing 200 tasks. Each task takes a seasoned professional three to eight hours and is graded on about 70 checks (75% factual accuracy, 25% visual design). Experts agreed with the automatic grader 96% of the time.
- Google Cloud's "Gemini for healthcare" preview (November 10): Connects health records, billing, and clinical guidelines to handle work such as medical records synthesis and prior authorization review under human oversight. The post says clinicians lose nearly 28 hours a week to paperwork.
- November 30: Gemini 4 Carbon's introductory price ends (then
$4input,$20output), and Claude Sonnet 4.5 retires from the API (covered in #4). - December 1:
gemini-2.5-proandgemini-2.5-flashswitch automatically (#4). - December 3:
gemini-omni-1.1-flashswitches automatically togemini-omni-1.2-flash(#5).
Our take: What this week's smaller items share is "running agents as an organization, not alone" — security runtimes are becoming shared open source, task boards are going team-wide, and grading is moving to public yardsticks. November 30 is a day when two prices change, so put it on the calendar first.
Source: NVIDIA Expands Open Source Commitment With OpenShell Donation and New AI Infrastructure (NVIDIA) · IBM and Microsoft announce a new AI agent platform designed for industry-specific transformation (Microsoft) · Cursor Projects is now multiplayer (Cursor) · Multimodal Knowledge Agent benchmark (Google) · Gemini for healthcare (Google) · Gemini API changelog (Google)