Wed, Nov 11, 2026Wednesday, November 11, 2026 · 20 stories · 8 min read

Anthropic sets its IPO range and moves into office software

Ramit KoulFounder, Software Engineer & Innovator · Published 5:05 AM ET

A share price range of $70 to $78 would value Anthropic at up to $1.4 trillion, and the company opened its investor roadshow alongside new Claude tools for meetings, spreadsheets and whiteboards.^{1, 2} Google is competing on price: Gemini 4 Carbon is now on the paid API at $2 per million input tokens through November 30, which VentureBeat says is a quarter of Claude Fable 5.2's standard rate.³ Adobe, meanwhile, plans to let outside agents edit native Photoshop and Premiere files over MCP.⁴ Here are the 5 stories that matter most, then 15 briefs. Numbers in the text link to the references at the end.

Models

Gemini 4 Carbon reaches the paid API at an introductory price

Google released Gemini 4 Carbon to developers on the paid tier of the Gemini API in AI Studio and Android Studio, and describes it as its strongest model for long-horizon agentic work.⁵ The release came nearly two weeks after Google announced the model with access limited to its Fairwind cybersecurity program.³

Carbon costs $2 per million input tokens and $10 per million output tokens until November 30, then rises to $4 and $20.⁶ VentureBeat calculates that this is a quarter of Claude Fable 5.2's standard rates now and half after the increase.³ It adds that Fable 5.2 has since overtaken Carbon on Artificial Analysis's general intelligence index.³

Google is keeping Gemini 4 Argon as the default choice for most software engineering and positions Carbon for the hardest tasks.⁶ Antigravity is also previewing /swarm, which turns one prompt into a coordinated team of agents, with Carbon handling design, coordination and auditing while lighter models do parallel work.⁶ Carbon is also coming to Gemini for business and to Google AI Ultra subscribers in the Gemini app.³

Why it matters for builders

The low price lasts less than three weeks, so benchmark cost per finished task rather than per token. VentureBeat notes that a model with cheaper tokens can still cost more once you count retries and corrections.³

Products

Anthropic brings Claude into meetings, spreadsheets and whiteboards

Anthropic launched Claude Whiteboards and Claude Sheets in beta and made Claude Dashboards generally available under a new name, Claude Data.^{7, 8} Claude Meetings adds Claude to video calls as a visible participant that drafts plans and pulls figures from connected apps while people talk. TechCrunch and Fortune report that Meetings is in beta on Team and Enterprise plans, starting with Google Meet.^{8, 2}

In meetings, Claude identifies itself and posts a notice in the chat. It will not join calls with external guests unless an admin authorizes it, and it leaves if one joins.⁸ Sheets supports tabs, formulas and dropdown menus, outlines every edit it makes and exports to Excel or CSV.^{8, 2} Anthropic product lead Nicholas Lin called Sheets fairly basic and said it is not meant for financial modeling.⁹ On Enterprise plans, Whiteboards and Sheets start switched off, and Claude Data turns on by default on November 19.⁷

Upgraded interactive connectors open full apps such as Adobe, Atlassian and HubSpot inside Claude.^{8, 10} Microsoft shares were down about 3% at midday after the news.⁹

Why it matters for builders: SaaS vendors now have a way to run their full app inside Claude through interactive connectors.⁸ Enterprise admins should review Claude Data before it turns on by default on November 19.⁷

Research

Stanford's AI100 panel says scaling is reaching its limits

Stanford's One Hundred Year Study on Artificial Intelligence published its third report, AI Goes Mainstream: The Scaling Era and the Rise of Generative AI, with 12 takeaways on the past five years.^{11, 12} Oxford's Michael Wooldridge chaired the panel, and its members include Peter Norvig, Toby Walsh and Sara Hooker.¹²

The panel says scaling drove the period's main advances but is running short of high-quality data, and compute for frontier models now costs too much.¹¹ It expects progress to come from better algorithms, smaller models and more time spent reasoning.¹¹ It also finds that investment has surpassed earlier booms while AI's economic impact does not yet appear in macroeconomic data, concluding that "AI is not yet paying its way".¹¹

On harms, the report cites AI finding serious security flaws, teams of agents carrying out cyberattacks, agents escaping test sandboxes, and deepfakes ending the assumption that recordings are reliable evidence.¹¹ Its last takeaway is that the report will be out of date by the time readers see it.¹¹

Why it matters for builders: The panel's bet on efficiency over scale favors teams working on smaller models and inference-time reasoning.¹¹

Products

Adobe will let third-party agents edit native Creative Cloud files

At Adobe MAX, Adobe expanded its Creative Agent across Firefly and the Creative Cloud apps and announced local MCP integrations for Photoshop, Illustrator, Premiere and After Effects.^{13, 4} Through MCP, outside agents will work on native project files on the user's own computer, and layers, vectors and sequences stay editable. A private beta is due in early 2027.^{4, 14}

Firefly is becoming what Adobe calls an AI creative studio, with models from Adobe and Topaz Labs plus partners including Google, OpenAI, ByteDance, Kling and ElevenLabs.^{13, 15} The agent now offers a Plan Mode for reviewing its approach before it runs, and a model selector.¹⁵ Always-on agents that handle recurring work in the background are in preview and are due in beta in early 2027.¹⁵

In Premiere, the agent can apply effects across every clip in a sequence and adjust color and exposure for the whole cut.¹⁶ Firefly Builder, in private beta, turns a text description into a prototype website or app.¹⁵ Adobe also told TNW it is considering one-off purchases of AI credits.¹⁵

Why it matters for builders: Starting in early 2027, agent builders will be able to work with Adobe's native file formats directly instead of flattened exports.⁴

Funding & deals

Anthropic targets up to $1.4 trillion in its Nasdaq listing

Anthropic's amended registration statement, filed November 10, offers 825 million Class A shares at $70 to $78 each under the Nasdaq ticker ANT.^{17, 1} The sale would raise $57.75 billion to $64.35 billion, and the top of the range implies a fully diluted valuation of $1.404 trillion.¹ Reuters, calculating from outstanding shares only, put the valuation at up to $1.02 trillion.¹⁸ Trading is expected to start around November 19.^{1, 17}

In a separate private placement, AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC will buy $26 billion of non-voting Class N stock at the IPO price, with Google and Nvidia taking $10 billion each.¹ Up to 20% of the offering may go to retail investors.¹

Third-quarter revenue was $17 billion, up from $1.4 billion a year earlier, and run-rate revenue reached about $81 billion at the end of October.¹⁹ The filing shows a nine-month net loss of $104 billion and compute and infrastructure commitments of $900.9 billion, about $70.7 billion of which falls due within 12 months.^{1, 17} The founders' Class F share carries 50.1% of the voting power, and the Long-Term Benefit Trust elects a majority of the board.¹

Why it matters for builders: Anthropic says it must make many of its compute payments even if demand falls short.¹⁹ That fixed cost is worth watching for teams that depend on Claude's capacity and pricing.

In brief

Products
Google Cloud launched Gemini for healthcare in preview. The product connects to health records, billing systems and Medicare data to run tasks such as summarizing medical records and reviewing prior authorizations. Its links to Epic and Oracle Health records are read-only, and HCA Healthcare, Highmark Health and Northwell Health are among the launch partners.^{20, 21} Customers can route work to models from Google, Anthropic or OpenAI.²¹
Research
Anthropic says 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in about 30 hours. Anthropic's own engineering post shows the agents started with 60 target statements, a one-sentence outline of the proof route and 47 lemmas from its September formalization, which took 11 days. TechTimes found no independent review of the new proof.²²
Products
Photoshop's Relight, Generative Restyle and Quick Compare become generally available this week. Relight lets users add up to three light sources anywhere in 3D space, and a new beta feature, Convert to Text Layers, rebuilds text from a flat image as editable type. Premiere 27.0 also shipped, with its Color Mode grading workspace now generally available.²³
Products
OpenAI launched Businesses in ChatGPT, a beta that turns its tested dashboard into a full program. Through Business Hub, verified businesses worldwide can claim their profiles, update product details, and see how often ChatGPT mentions them and how much traffic it sends their sites. Eligible U.S. e-commerce and online service companies can also publish Business Agents, which chat with users in separate, labeled threads and are free for now.^{24, 25}
Models
OpenAI's CFO says its Luna model costs a fraction of Chinese open-weight models. Sarah Friar said Luna costs about one-fifth to one-sixth as much as a Chinese open-source model, and that usage rose tenfold after OpenAI cut its price by about 80%. Luna ranks fourth by token usage on OpenRouter, behind open-weight models from DeepSeek and Z.AI.²⁶
Models
Meta's Muse models are now available on Google Cloud. Muse Ember, Muse Spark 1.3 and Muse Wonder are in Google's Model Garden, where developers can use them through managed APIs without running their own infrastructure.²⁷
Policy & legal
Canada joined the Paris Call for AI Safety and Security alongside OpenAI, Anthropic and Cohere. Six countries and 10 AI companies committed to allow some independent testing of their models, to warn early about incidents and to shut down systems if necessary, but the text does not say how it would be enforced. The U.S. and Chinese governments did not take part, and neither did Google or Meta.²⁸
Products
Kora unveiled a Linux operating system with AI models and agents built into the system itself. Kora's Cortex inference engine pools a PC's CPU, GPU and NPU, and Snapdragon X2 Elite machines can run a 26 billion-parameter model locally. The first PCs, due in 2027, will run HUMAIN OS, and Kora itself follows later that year.²⁹
Models
Zyphra released ZONOS3-Mini, an open-weight text-to-speech model small enough to run in a browser. The Apache 2.0 model is a mixture of experts with 800 million total and 100 million active parameters. It runs through WebGPU, MLX, ggml or SGLang, and all of its benchmark results are self-reported.³⁰
Models
Fish Audio released Live 3, a full-duplex voice model that learns when to talk. Live 3 keeps listening while it speaks and can react with short acknowledgments in as little as 186 ms. It supports English only, costs $0.035 per minute through December 31 before rising to $0.07, and has no published benchmark scores.³¹
Infrastructure
An analysis syndicated on AOL argues Alphabet, Meta and Oracle are teaming up against Nvidia's biggest advantage. The piece says Nvidia's GPUs have long led AI chips on flexibility and performance, but the available excerpt does not describe what the three companies are doing together.³²
Funding & deals
OpenAI acquired Odessia, an AI travel agent, and shut the service down the same day. Founder Francis Davidson, a co-founder of Sonder, is joining OpenAI to work on travel. Odessia could book flights and hotels and held IATA accreditation, and Expedia and Booking Holdings shares fell on the news.^{33, 34}
Products
Premiere can now generate b-roll, music and sound effects inside the timeline. The Generative Media Tool uses Firefly and partner models from Google, Kling, Runway and Luma. Adobe also added two Topaz Labs enhancement models, Wonder 4 and Image HDR.³⁵
Policy & legal
A French advisory council has asked for stricter rules on AI assistants, especially for minors. Following Macron's call for child protections, the CIANum report delivered on November 10 stops short of a ban. It proposes turning off anthropomorphic and flattery features by default, detecting suicide and self-harm topics more reliably, and adding a button for reporting problem responses.³⁶
Funding & deals
The Associated Press put Anthropic's IPO valuation near $1.3 trillion at the midpoint of its range. AP reported that selling 825 million shares at the $74 midpoint would raise more than $61 billion, with a Wall Street debut expected this month.³⁷

That’s the edition.

The next one is ready tomorrow by 6 AM ET.

Tue, Nov 10 edition All editions

Written with AI from the sources in the references above.

AuthorRamit KoulFounder, Software Engineer & Innovator