AI & SaaS News, Oct 21: Anthropic S-1, Notion AI Credits

Last checked October 21, 2026 · News roundup

The short version for October 21, 2026: The biggest AI and SaaS news this morning is money. A draft of Anthropic's IPO prospectus, reported by the Wall Street Journal, shows $16.3 billion in first-half revenue next to an $80.7 billion net loss, while Reuters says OpenAI is rushing a roughly $30 billion raise before Anthropic lists. For everyday tools, Notion rebuilt its AI agents and changes how credits work on November 2, Gemini on Android Auto can now use your Gmail and Calendar if you opt in, and Yelp's app can phone a restaurant for you.

Below: five AI stories first, then four SaaS and business stories, each with the change, the people it touches and the next step, if there is one.

Part 1 · AI

Anthropic IPO filing: $16.3B revenue and an $80.7B loss in the draft S-1

Anthropic's draft IPO prospectus shows first-half 2026 revenue of $16.3 billion and a net loss of $80.7 billion, according to the Wall Street Journal (October 20). Reuters, citing the Journal, adds that the loss a year earlier was $13 billion and that about $72 billion of this year's loss comes from accounting charges on convertible notes and other instruments Anthropic used to raise money, so most of it is not cash spent. Secondary coverage of the WSJ report puts first-half 2025 revenue at $1.1 billion and says a large share of 2026 revenue came from two unnamed customers.

The filing lands as OpenAI tries to get ahead of it. Reuters reported on October 19 that OpenAI is racing to raise about $30 billion at a valuation that could top $1.4 trillion, with SoftBank and Abu Dhabi's MGX in talks, before Anthropic's expected listing in mid-November. The same report says some investors argue for an Anthropic valuation near $1.5 trillion rather than the $2 trillion the company wants.

Who it affects: investors, and anyone tracking which AI vendor will be around in five years. What changes for Claude users today: nothing; prices and plans are not part of the filing. The public S-1, once filed, will be the first audited look at these numbers.

Gemini on Android Auto: Personal Intelligence can now use your Gmail and Calendar

Since October 20, Gemini in Android Auto can personalize its answers with Personal Intelligence, which draws on Gmail, Calendar, Photos, Maps, Search and YouTube, 9to5Google reports. In practice you can ask "is my flight delayed?" without saying the flight number, or "take me to my appointment" without an address, because Gemini pulls those details from your email and calendar.

The feature is opt-in. If you already switched on Personal Intelligence for the Gemini app on your phone, it comes to Android Auto automatically; if you haven't, the car still handles navigation, music, calls and general questions, just without personal answers. Cars with Google built-in (Android Automotive) get it "in the coming months," and there it only uses your data while your phone is connected. Google hasn't said which regions or paid plans are required.

What to do: if you share a car or don't want your inbox read aloud on the road, check the Personal Intelligence setting in the Gemini app before your next drive. For Google's other Gemini news this month, our guide to what Gemini 4 Argon costs and who can use it is still current.

How popular is the Meta Muse app? 10.2 million US downloads in 40 days

Meta's Muse AI companion app has passed 10.9 million downloads since its September debut, 10.2 million of them in the US, according to Sensor Tower data published by TechCrunch on October 20. Only 11 other apps have reached 10 million US downloads within 40 days; the last was Threads in July 2023.

Retention looks unusually strong for an AI app: 30% of US users were still active after seven days, against an average of 19% for ChatGPT, Claude and Grok. Daily active users reached 2.5 million for the first time on October 16. The growth is not only organic, though: Sensor Tower says Meta put 39% of its total ad spend behind Muse in October, up from 23% in September.

Who it affects: Muse is available only in the US and Canada, so the comparison with ChatGPT's global launch (about 15 million downloads in 40 days) isn't like for like. If you use Muse with Marketplace or other Meta accounts, review its permissions; agent mistakes there made news earlier this week.

Yelp "Check for Me": the app can now call a restaurant to ask about wait times

Yelp added Check for Me on October 20, a button on restaurant pages that sends a voice agent to call the restaurant for you, TechCrunch reports. You enter your party size and phone number, the agent asks about the wait, and Yelp texts you the answer.

The Yelp AI Assistant also gained OpenTable booking inside the app, memory of preferences you mention, a 30-day chat history and custom reminders. New "Been Here" and "Want to Go" lists replace Collections, and weekly personalized recommendations based on those lists start in November. For business owners, Yelp launched Smart Lead Reply, an agent that answers and qualifies leads for $30 a month.

Worth knowing: Yelp hasn't published a consumer price or country list for Check for Me. If the restaurant already uses its own AI phone agent, your call may be one bot talking to another, which TechCrunch notes can break down. Restaurants that would rather not field automated calls should check how their phone system handles them.

Can GPT-6 write COBOL? OpenAI's new tests say yes, mostly

OpenAI published two new coding evaluations on October 19 aimed at "low-resource" languages such as COBOL, Fortran, Visual Basic and OCaml, the code still running many banks, insurers and government systems. On MultiLangBench, which asks a model to build a SQLite-compatible database engine in 20 languages, GPT-6 Astra averaged an 87.4% pass rate at maximum reasoning, up from 26.8% for GPT-5.6 Sol. Its COBOL score went from 10.9% to 72.1%.

On a separate COBOL migration test, which checks whether converted programs behave the same, GPT-6.1 Sol scored 86.6% at low reasoning effort (GPT-5.6 Sol: 5.4%) and 97.9% at higher effort. OpenAI says GPT-6.1 Sol did it at 39% lower cost than its predecessor at the same setting. The benchmark setup and per-language results are on GitHub.

Who it affects: teams planning mainframe migrations and the consultancies that sell them. These are OpenAI's own benchmarks, so treat them as a starting point and test on your own code. If you want software built without writing any code at all, our prompt-to-app and no-code builder comparison covers a different route.

Part 2 · SaaS & business

Notion AI credits change on November 2: what Business and Enterprise seats include

Notion rebuilt its AI agents and its credit system, co-founder Akshay Kothari wrote on October 19. Each Business and Enterprise seat now includes 1,500 Standard credits a month (Notion picks the model) and 500 Select credits a month (you pick the model, or run Custom Agents). Admins can buy extra workspace credits for shared agents and expensive models. The new setup is live for new customers; existing customers move over on November 2, 2026, and customers in the middle of a contract keep their current allowances until renewal.

The bigger change is under the hood. A new Auto router sends each task to a cheaper model when it can; in Notion's sample of 20+ use cases it matched results from top models at about 90% lower average cost. Agents can now work on one task for more than 24 hours, up from five minutes, and gained computer and browser use. Plus, Pro and Go ChatGPT subscribers can connect their account to a personal Notion Agent, and Enterprise bring-your-own-key is in alpha. An AI Command Center with spending limits is "coming soon."

What to do: Notion says most users won't run out of credits, but heavy Custom Agent users should check usage before November 2. We track changes like this on our SaaS pricing changes page.

Microsoft 365 E7 agent controls: risk scores and a kill switch for AI agents

Microsoft unveiled new tools on October 19 to monitor and restrain AI agents inside companies, GeekWire reports, as part of Microsoft 365 E7 ($99 per user per month, launched in May) and its Project Perception security work. Each agent gets a continuously updated risk score based on what data it can reach and whether it strays from its task. When the score climbs, admins can switch the agent to read-only, require a human to approve high-stakes actions, or shut it down.

Microsoft can now detect 60 kinds of agents running on employee computers, up from 25, including Claude Code, OpenAI's Codex and GitHub Copilot. New Guardian Agents investigate why a score changed and can recommend or take containment steps. Microsoft security chief Hayete Gallot framed agents as potential insider threats in a company blog post, writing that "frontier models, regardless of their origin, cannot be trusted," because they are trained to finish tasks rather than follow company norms.

Availability: private preview now, with broader availability planned for Microsoft Ignite, November 17–19 in San Francisco. IT teams already on E5 or E7 should ask their Microsoft rep about the preview, especially if developers run coding agents on company laptops.

Otter.ai raises $320 million: what it means for the meeting notes app

Otter.ai raised $320 million from General Catalyst's Customer Value Fund, TechCrunch reported on October 20. The fund doesn't take equity; it earns returns tied to how well the marketing spend it pays for performs. Otter's previous funding totaled about $50 million, per PitchBook, with its last round in 2021. No valuation was disclosed.

Otter says it passed $100 million in annual recurring revenue last year and now has 40 million users, up from 25 million. The money goes to enterprise sales, international expansion and product work: Otter is testing a pre-meeting brief built from past conversations and "Live Assist," which lets you ask the AI questions during a meeting. It faces Fireflies.ai, Granola, Read AI and newer rivals including Calendly and Superhuman, which bought Fathom.

For users: TechCrunch reported no pricing changes. Expect more enterprise features and, likely, more marketing.

Pigment raises $250M at a $2.3B valuation to take on Excel for planning

Paris-based Pigment, which makes AI software for budgeting, forecasting and scenario planning, raised $250 million at a $2.3 billion valuation, Bloomberg reported on October 20. The Scaleup Europe Fund led the round, with Nvidia and World Innovation Lab joining existing backers Iconiq, IVP and Meritech, according to Crypto Briefing's summary of the deal.

Pigment says annual recurring revenue has passed $100 million and doubled year over year. It pitches itself against Excel and PowerPoint as well as planning suites from Anaplan, Oracle, SAP and Workday, and names L'Oréal, Unilever, Uber and Siemens as customers. Its last round was a $145 million Series D in April 2024.

Who should care: finance teams choosing between spreadsheets and a planning platform. Pigment's site points buyers to a demo request rather than a public price list, so plan for a sales conversation and ask to trial it on one budget cycle first.

Your questions on the Anthropic filing, Notion credits and Gemini in the car

When is the Anthropic IPO?

Anthropic hasn't set a public date. Reuters reported on October 19 that the listing is expected in mid-November 2026. The Wall Street Journal has seen a draft prospectus, but the numbers become official only when Anthropic files a public S-1 with the SEC, which must happen at least 15 days before the investor roadshow begins.

Will I run out of Notion AI credits after November 2?

Probably not if you use Notion AI for everyday writing and Q&A. Each Business or Enterprise seat gets 1,500 Standard and 500 Select credits a month, and Notion says most people won't hit the limit. Heavy Custom Agent use draws on Select credits faster, and admins can buy more for the workspace.

Can Gemini read my email in my car?

Only if you opt in. Personal Intelligence on Android Auto uses Gmail, Calendar, Photos, Maps and other Google apps to answer questions like flight status or where your next appointment is. If you never turned it on in the Gemini app on your phone, Android Auto keeps giving general, non-personal answers.

Is Yelp's Check for Me free?

Yelp hasn't published a price for Check for Me, and TechCrunch's launch report doesn't mention one, so check the restaurant page in the Yelp app. The paid product announced the same day, Smart Lead Reply at $30 a month, is for business owners who want an AI agent to answer and qualify customer leads.

Sources (accessed October 21, 2026): Reuters via Investing.com on the WSJ S-1 report (Oct 20) · Reuters on OpenAI's fundraising (Oct 19) · Tech Startups on the S-1 figures (Oct 20) · 9to5Google (Oct 20) · TechCrunch on Muse (Oct 20) · TechCrunch on Yelp (Oct 20) · OpenAI (Oct 19) · Notion blog (Oct 19) · GeekWire (Oct 19) · TechCrunch on Microsoft (Oct 19) · TechCrunch on Otter.ai (Oct 20) · Bloomberg on Pigment (Oct 20) · Crypto Briefing on Pigment (Oct 20)