Skip to content
Company · Claude

Anthropic: the company that makes Claude, and what you should know about it

Anthropic is a company that works on artificial intelligence and model safety and makes Claude: a chat with a voice mode, the Claude Code development system and the Claude Duet assistant. Alexander Volchek ranks it, together with OpenAI, as one of the two main market leaders. For an ordinary user it is above all a choice: which system to make your main one, and which limits and rules you will have to live with. Below is what the ToTheMoon hosts have said about Anthropic's products, its money, its IPO and its bans.

Updated November 16, 2026 · ToTheMoon editors · based on 95 episodes

What Anthropic is and what it makes

Anthropic's main product is Claude, but far from everyone knows the company. Anna Volchek recalls an acquaintance from her professional circle who did not know what Anthropic was. That is hardly surprising: ChatGPT, as Alexander Volchek notes, has become a generic name — like Pampers or Xerox — and most people see modern AI precisely through it.

Anthropic has several products. The Claude chat has a voice mode, Voice, which works with the Opus and Sonnet models. Claude Code is a development system; it has gained Projects, where a project has one goal and several parallel tasks, each in its own branch. Claude Duet is an assistant that keeps working on assignments even when the app is closed and calls the person in when a decision is needed; it launched in preview through a waitlist. Among the models, the episodes discuss Claude Opus 5.5, Fable 5.1 and Claude Sonnet 5.

The lineup keeps changing: on September 16 Cowork was merged with the chat, and the separate Claude Designs app is being shut down. Alexander sees this fragmentation as a product of project managers' narrow view, and Boris Cherny, one of the main creators of Claude Code, went as far as proposing to erase the product and rewrite it from scratch, calling it outdated. At the same time, Alexander is convinced that Claude Code is not just for programmers: it is worth connecting your email to it and starting to automate your own life.

How Anthropic differs from OpenAI — in the hosts' view

Alexander Volchek advises working with Anthropic and OpenAI directly, without wrappers and aggregators, but assigns them different roles. He finds it more convenient to connect email and services to Codex: behind it stands ChatGPT, which Alexander rates above Anthropic's system. For development, though, Claude Code is, to his taste, “better than Codex”: in Max mode the Opus 5.5 and Fable models are, in his experience, stronger than 6 Astra and 6.1 Sol.

The second difference is limits and audience. In ChatGPT, according to Alexander, good models came practically without restrictions, while at Anthropic they did not. OpenAI's Dots works independently of tokens, whereas Claude Duet spends Claude's shared tokens and stops when they run out. Hence his conclusion: Anthropic designs Duet not for an ordinary person's everyday affairs but for running a business — no wonder its examples feature QuickBooks, Stripe, HubSpot and Salesforce. Alexander himself stopped using Duet, calling it a half-baked copy of Dots.

Alexander believes in the future of both companies: within three to ten years, he forecasts, OpenAI or Anthropic should be worth more than ten trillion dollars, unless huge competition appears on the market. But if both gave up on building AGI for ordinary people, he says, he would write them off.

Money, limits and going public

Anthropic is preparing for an IPO: the company has published its S-1, the valuation is expected to exceed two trillion, and in its SEC filings it had to admit that AI will increasingly create itself. Ilnar Shafigullin considers the listing a huge plus: the filings show profitability, expenses and investment in compute — around half a trillion through 2030, if he remembers correctly. By his count, more than a hundred billion has already been raised, the offering will bring in another one hundred to one hundred and fifty, about a hundred is sitting in the bank, and together with current profit that will be enough to cover the obligations.

Alexander Volchek disagrees: the obligations come to almost a trillion, while only around ten percent of the company will be floated on the exchange, maybe four or five. Revenue, by his figures: five billion for the first quarter of 2026, sixteen for the first half. Ilnar thinks inference is cheap and a two-hundred- or five-hundred-dollar subscription is enough to pay for the servers; Alexander thinks heavy users are loss-making and that soon models of today's level will cost tens of thousands of dollars.

Money explains the limits too. In Ilnar's version, Anthropic cannot hand out limit resets the way OpenAI does because of a decision Dario Amodei talked about before the IPO: to build data centers for a pessimistic scenario. Demand turned out to be optimistic; according to the news, the company is now investing about seven hundred and fifty billion, but the capacity will only come online in three to four years. If you keep hitting the limits, try Ilnar's setup: break the task into parts in Claude and carry it out through open code on models like DeepSeek and Kimi, which are many times cheaper via the API.

By November 2026 the numbers had grown. Anthropic's quarterly revenue, according to figures Alexander Volchek cites in episode 194, reached seventeen billion, although the company, as he and Ilnar recalled, had wanted to report more. Infrastructure obligations stand at 895 billion dollars, seventy of them due next year; the loss of a hundred and four billion includes an accounting adjustment for financing, and making sense of such reporting, Alexander says, is very hard. Besides chips, the company also had to take care of energy: according to the Financial Times, Anthropic reserved gas generators worth three to four billion dollars — almost two gigawatts.

Rules, bans and safety: what matters to a user

Anthropic, unlike OpenAI, does not allow multiple accounts and keeps banning users en masse around the world. The company puts a lot of effort into customer verification (KYC) and above all fights distillation — training other companies' models on its own models' answers. It recently shut down tens of thousands of accounts, apparently Chinese botnets. If Anthropic asks whether you are a company or an individual, Alexander Volchek advises not rushing to answer: in your own business you may use a regular account if you are, roughly speaking, the only one in it.

A personal consumer Claude account is not allowed under eighteen, and the company promises to detect and block children's accounts, but, as Anna Volchek points out, Common Sense Media's March review did not find a single confirmed case. Training on your data at Anthropic, as in ChatGPT, is on by default but can be turned off; Alexander advises checking this setting, finding out how long deleted chats are kept and getting to grips with the laws of your country.

Safety is what Anthropic started with: according to Max Grigoriev, its mission and charter were about safety, and the company was built by AI-safety fanatics. Today, in his estimate, it is staffed mostly by mercenaries, and safety can no longer be considered the company's top priority. Systems of the Mythos class, Alexander says, have already carried out the first serious hacks, and only select companies have been given access to them, in coordination with the US government.

Fable 5.2 and Mythos: Anthropic's most powerful models

By November 2026 Fable 5.2 and Mythos 5.2 had come out — the biggest news of the week, according to Alexander Volchek. Fable comes from the Mythos family: these models are not available to ordinary people, were created for cybersecurity and vulnerability hunting, and in the spring, he says, more than ten thousand bugs were found in open-source systems with their help. In the summer Fable became the first model to be officially blocked by the US government. Alexander names several versions of the conflict with the White House — from the story of Maduro's capture, which the company's leadership opposed, to the theory that the block was marketing.

After the block, Fable came back with restrictions and limits. On many plans it is capped at half the limit, but not on the $200 one; on the $20 plan there is no point in it, and via the API it costs, as Alexander puts it, astronomical money. Max mode costs about three and a half times as much as Medium and can make a task's result worse. He advises using Fable for contracts, finance, finding errors in processes and ad analysis, and in complex programming Fable 5.2, he says, is several times better — he switched a task with too much money riding on it from Opus 5.5 to Fable.

The company's pace is speeding up: according to an outside estimate Alexander cites, Anthropic is improving its models' capabilities about twice as fast as before, largely thanks to AI itself. It has also described a memory system in which Claude does not just save notes but, during pauses, checks them against its work history, corrects mistakes and restructures its knowledge — which matters for long projects.

Questions

Anthropic: in brief

What is Anthropic?

Anthropic is the company that develops Claude and works on artificial intelligence and model safety. Its products are the Claude chat with a voice mode, the Claude Code development system and the Claude Duet assistant. Alexander Volchek counts Anthropic, together with OpenAI, as one of the two main market leaders, and the company is now preparing to go public.

How is Claude different from ChatGPT?

In Alexander Volchek's experience, ChatGPT and Codex are more convenient for everyday life and for connecting email, while Claude Code with the Opus 5.5 and Fable models is stronger for development. Anthropic's limits, he says, are stricter, and the company's products consume tokens. He advises choosing one main system and keeping the second as a backup.

When will Anthropic go public?

The episodes don't name an exact date, but preparations are under way: Anthropic has published its S-1, and the valuation is expected to exceed two trillion. Ilnar Shafigullin expects the offering to bring the company one hundred to one hundred and fifty billion; Alexander Volchek points to obligations of almost a trillion. An open question is whether AI subscriptions will get more expensive after the IPO.

Why does Anthropic ban accounts?

The company does not allow multiple accounts, verifies its customers and fights distillation — it recently shut down tens of thousands of accounts resembling Chinese botnets. Alexander Volchek advises studying the rules, not rushing to answer whether you are a company or an individual, and splitting tasks between Claude Code and Codex in advance in case you get blocked.

What is Claude Fable 5.2 and how much does it cost?

Fable 5.2 is Anthropic's most powerful model, from the Mythos family built for finding vulnerabilities; in the summer Fable became the first model to be officially blocked by the US government, and it was later brought back with limits. On the $20 plan, Alexander Volchek says, there is no point in it, and via the API it is very expensive. He recommends it for contracts, finance and complex automation.

Latest conversations

Recent episodes where it's the main topic

Open the dossier →

The list updates itself with every new episode of the podcast.

More guides

More guides

All guides →