From seven OpenAI researchers to the world's most valuable AI startup

Anthropic was founded in 2021 by siblings Dario and Daniela Amodei along with five other former OpenAI employees. Dario Amodei previously served as vice president of research at OpenAI and was one of the main architects of the GPT-3 model. The reason for their departure was differing views on safety direction — while OpenAI under Sam Altman accelerated commercialization, the group around the Amodeis wanted to build AI systems with an emphasis on safety and ethics.

The first version of the Claude model was completed as early as summer 2022 — before the public launch of ChatGPT. But Anthropic decided not to release the model right away. The company's leadership argued for the need for additional safety testing and concern about sparking a dangerous race for ever more powerful AI. The public launch didn't come until March 2023.

This caution paid off for the company. While OpenAI faces lawsuits, criticism for hallucinations, and questions about pricing sustainability, Anthropic has built a reputation as a "responsible AI company" — and attracted investors who have poured tens of billions of dollars into it. In May 2026, Andrej Karpathy, OpenAI co-founder with Slovak roots, also joined Anthropic, which only underscores the company's growing prestige.

Constitutional AI — a model that follows its own constitution

Anthropic's biggest technological differentiator is so-called Constitutional AI. While most models learn to "be good" through feedback from human annotators (RLHF method — Reinforcement Learning from Human Feedback), Claude is trained to evaluate its own responses according to a predefined "constitution" — a set of ethical principles derived from documents such as the UN Universal Declaration of Human Rights from 1948.

The principle is elegant: the model generates a response, checks it itself against constitutional rules, and corrects it if necessary. The result is an AI that is consistently cautious, less prone to toxic or dangerous outputs, and overall "more polite" than the competition. However, this also means that Claude sometimes refuses to answer even legitimate queries — a frequent complaint from users accustomed to the more permissive ChatGPT.

The company also invests heavily in mechanistic interpretability — research that seeks to understand exactly what happens inside neural networks. In 2024, Anthropic scientists used a technique called dictionary learning to identify millions of patterns of neural activity in the Claude model, including one associated with the Golden Gate Bridge. In other words — Anthropic doesn't just build AI, it also tries to truly understand it.

The Claude model family: From humble beginnings to Mythos

The journey of Claude models mirrors the rocket growth of the entire company:

Claude 1 and 2 (2023): The first public models. Claude 2 came with a context window of 100,000 tokens, which was groundbreaking at the time — users could insert an entire book into a single conversation.

Claude 3 (March 2024): Three variants — Opus (most powerful), Sonnet (balanced), and Haiku (fast and cheap). Claude 3 Opus surpassed GPT-4 for the first time on several benchmarks including MMLU and HumanEval.

Claude 3.5 Sonnet (June 2024): An intergenerational upgrade with an emphasis on coding and logical reasoning. It became a favorite tool for developers.

Claude 4 (May 2025): A major leap in performance. Opus and Sonnet received improved capabilities for complex enterprise tasks and agent workflows. At the same time, Claude Code was launched — a terminal developer tool that kicked off the phenomenon of vibe coding (programming by describing in natural language).

Claude Opus 4.6 and Sonnet 4.6 (February 2026): Another iteration with improved reasoning and speed.

Mythos (April 2026): The newest and most capable model. Before the public release, Anthropic discovered over 10,000 critical bugs, demonstrating a thorough safety process. The model is currently available only on a limited basis — Japanese researchers gained access to it through a special agreement.

Who funds Anthropic and why it's heading for the stock market

Anthropic's financial history is breathtaking. In 2022, the company raised 580 million dollars — of which 500 million came from Sam Bankman-Fried's cryptocurrency exchange FTX. After FTX's collapse, these funds became the subject of legal disputes.

Further investments came from technology giants: Google invested over 3 billion dollars in stages, Amazon nearly 8 billion in exchange for Anthropic using AWS as its primary cloud infrastructure. In November 2025, Nvidia and Microsoft joined with a 15 billion dollar investment and a commitment for Anthropic to purchase 30 billion worth of computing capacity from Microsoft Azure.

The company's own equity rounds grew at rocket speed: Series E in March 2025 brought in 3.5 billion (valuation of 61.5 billion), Series F in September 2025 another 13 billion (valuation of 183 billion), and Series G in February 2026 a record 30 billion dollars at a valuation of 380 billion dollars.

In May 2026, the company announced plans to go public (IPO) with a target debut in autumn 2026. If successful, it will be one of the largest technology IPOs in history. As an interesting point: today you can invest in Anthropic indirectly through "lab funds" from Harbor Capital, which allow betting on individual AI companies even before their IPO.

The Pentagon dispute: When ethics collides with national security

The year 2026 brought Anthropic its biggest reputational test in its history. It all started in June 2025, when the company launched Claude Gov — a version of the model for government use — and secured a contract from the U.S. Department of Defense worth 200 million dollars.

The problem arose when the Pentagon demanded the removal of safety guardrails that prevented the use of Claude for fully autonomous weapons and domestic mass surveillance. Anthropic refused. Defense Secretary Pete Hegseth threatened to remove the company from the supply chain. When Anthropic didn't back down, the Trump administration ordered government agencies to stop using Anthropic's models. In March 2026, the Pentagon officially designated Anthropic as a "supply chain risk" and canceled the 200 million dollar contract.

Anthropic turned to the courts. A federal judge issued a preliminary injunction, calling the Pentagon's actions "classic retaliation violating the First Amendment of the Constitution." The heads of Google, Amazon, Apple, and Microsoft sided with Anthropic in the dispute. The whole case opened up a fundamental question: do private AI companies have the right to determine how their technologies are used in war?

Claude in Czechia and Europe: Availability, pricing, and Czech language

For Czech users, Claude is available through the web interface claude.ai, mobile apps for iOS and Android, and desktop versions for Mac and Windows. Registration is possible from the Czech Republic without needing a VPN.

The model supports Czech at a solid level — it understands Czech queries, generates fluent Czech texts, and handles translations between Czech and other languages. It's not quite native-level Czech like specialized tools, but for everyday use, study, or work, it's fully sufficient.

In terms of pricing, Claude Pro is available for 20 USD per month (approximately 460 CZK at the current exchange rate). Enterprise customers can use Claude via Amazon Bedrock or Google Cloud, which enables deployment in compliance with European GDPR regulation. For developers, Claude Code is charged based on token consumption — prices range in single dollars per million tokens depending on the chosen model.

In the context of the EU AI Act, Anthropic is seen as a company that goes beyond legal requirements in the area of safety — which gives it a competitive advantage in the European market, where regulation tightens rules for all players. Claude is fully available in the European Union without restrictions, unlike some Gemini features or advanced voice modes of ChatGPT, which arrive in the EU with delays.