Anthropic

Updated June 2026

Anthropic is the safety-focused AI lab behind Claude. Founded in 2021 by former OpenAI researchers, it pioneered Constitutional AI, leads independent coding benchmarks, and in 2026 became the world's most valuable AI startup at a $965 billion valuation while its revenue overtook OpenAI's.

Website →API →Pricing →Status →

Anthropic is the AI lab behind Claude, founded in January 2021 by seven former OpenAI researchers who left over disagreements about how to develop powerful AI safely. It pioneered Constitutional AI, ships the Claude model family — the models that consistently top independent coding benchmarks — and is the only frontier lab whose models are sold on all three major clouds (AWS, Google Cloud and Azure). In 2026 it became the world’s most valuable AI startup, raising $65 billion in a Series H round at a $965 billion post-money valuation, overtaking OpenAI on both valuation and revenue.

The 2026 story is a safety lab that has become a commercial juggernaut without abandoning its founding posture. Anthropic’s run-rate revenue crossed $47 billion in May 2026 — up from roughly $9 billion at the end of 2025 — driven overwhelmingly by enterprise and by Claude Code. At the same time it remains the lab most willing to gate its own products on safety: in June 2026 it suspended its most capable models, Fable 5 and Mythos 5, worldwide to comply with a US export-control directive — an 18-day standoff that ended when the US Commerce Department lifted the controls on 30 June, with Fable 5 returning to general availability from 1 July 2026.

Quick facts

History and founding

The departures from OpenAI began in December 2020, and Anthropic was incorporated in January 2021 around siblings Dario Amodei (former VP of Research at OpenAI) and Daniela Amodei (former VP of Operations). The founders have consistently framed the split as a directional disagreement about safety — the belief that it should be a foundational design principle rather than a feature added after the fact — rather than a reaction to OpenAI’s Microsoft deal. The other five co-founders were Tom Brown (lead author of the GPT-3 paper), interpretability pioneer Chris Olah, now-CTO Sam McCandlish, scaling-laws researcher and now-Chief Science Officer Jared Kaplan, and policy lead Jack Clark.

Anthropic incorporated as a Delaware Public Benefit Corporation, letting directors balance shareholder returns against its stated mission of “the responsible development and maintenance of advanced AI for the long-term benefit of humanity.” Governance sits with a Long-Term Benefit Trust that holds special-class shares and appoints a growing share of the board; trustees cannot be employees, executives or major investors. The structure is designed to preserve mission control as commercial pressure mounts — a pressure that has only intensified at a near-trillion-dollar valuation.

Funding and valuation

Anthropic’s capital raises are among the largest in private-company history, with cloud and chip giants investing alongside traditional venture funds.

The cumulative cloud commitments are the backbone of the story. Amazon has invested roughly $8 billion as Anthropic’s primary cloud partner and committed a further $15 billion of hyperscaler capacity alongside the Series H; Google has invested several billion and, in an October 2025 deal, gave Anthropic access to up to one million TPUs; Microsoft and Nvidia committed up to $5 billion and $10 billion respectively in November 2025, making Claude available on Azure. The Series H drew an unusually broad book — Altimeter, Dragoneer, Greenoaks and Sequoia as leads, plus GIC, Coatue, ICONIQ, Blackstone, Fidelity, T. Rowe Price, and memory-and-silicon partners Micron, Samsung and SK hynix (Anthropic).

Despite the billions invested, no cloud partner holds a board seat — an independence Anthropic treats as central to its mission.

Models and the Mythos-class tier

Anthropic’s model ladder runs Haiku → Sonnet → Opus → Mythos-class, the last a new top tier introduced in June 2026 that sits above Opus. The mainstream flagship is Claude Opus 4.8; the most capable model, the Mythos-class Fable 5, was suspended 12–30 June 2026 under an export-control order but returned to general availability on 1 July, while its restricted twin Mythos 5 is trusted-access only (see Controversies). The full Claude lineup and live benchmarks render in the table below this page.

Two threads define the 2026 generation. First, context windows reached one million tokens across the Opus 4.x line, removing a long-standing gap with Google’s Gemini. Second, Opus 4.8 leaned into honesty over raw capability: Anthropic reports it is around four times less likely than Opus 4.7 to let flaws in its own code pass unremarked, framing it as “a modest but tangible improvement” rather than a leap.

The leap came on 9 June 2026 with Fable 5, the first generally-released Mythos-class model. On the memorisation-resistant SWE-bench Pro it scores 80.3%, well ahead of Opus 4.8 (69.2%), GPT-5.5 (58.6%) and Gemini 3.1 Pro (54.2%) (Anthropic). Fable 5 ships with safety classifiers that silently route a narrow slice of queries (cyber, bio-chem, model-distillation) to the weaker Opus 4.8; its no-classifier twin, Mythos 5, is reserved for vetted government and critical-infrastructure partners through Project Glasswing. See best AI for coding for how these rank against rivals.

Constitutional AI and the safety framework

Anthropic’s technical signature is Constitutional AI (CAI), introduced in a December 2022 paper as an alternative to standard human-feedback training. Instead of relying solely on human labellers, the model critiques and revises its own responses against an explicit written “constitution” — drawn from sources such as the UN Declaration of Human Rights — and a second model provides preference feedback (RLAIF). The pitch is scalability, transparency about values, and less reliance on humans reviewing disturbing content.

That philosophy is operationalised in the Responsible Scaling Policy (RSP) and its AI Safety Levels (ASL), modelled on biosafety containment tiers. Anthropic activated ASL-3 protections — enhanced model-weight security, deployment restrictions, and CBRN risk controls — with Claude Opus 4 in May 2025, and every Opus model since has shipped under them. The Mythos-class models are benchmarked at a still-higher alignment bar, and the June 2026 export-control suspension is the clearest real-world test yet of Anthropic’s willingness to pull a shipping product on safety or compliance grounds.

Products and ecosystem

• Claude apps — web, iOS, Android, Windows and macOS, with Free, Pro ($20/month), and Max ($100–$200/month) tiers.
• Claude API — the developer platform, now under the claude.com domain, with code execution, tool use, the Model Context Protocol, and prompt caching.
• Claude Code — the agentic command-line and IDE coding tool that is the company’s breakout product, passing a $2.5 billion annualised run rate by early 2026.
• Claude Cowork — a multi-step agent that carries out tasks on the user’s behalf; Microsoft’s enterprise Copilot Cowork is built on the same Claude engine.
• Enterprise — Claude for Enterprise (SSO, SCIM, audit logs) and deep integrations across AWS Bedrock, Google Vertex AI and Azure AI Foundry.

The Model Context Protocol, open-sourced by Anthropic in late 2024, has become an industry standard, adopted by OpenAI and Microsoft and now common across competing assistants.

Business and financials

Anthropic’s revenue growth is the fastest of any company at its scale: a run rate of roughly $1 billion in 2024, ~$9 billion at the end of 2025, then $14 billion (February), $19 billion (March), $30 billion (April) and over $47 billion by late May 2026 (Sacra, VentureBeat). That trajectory took Anthropic past OpenAI in revenue during 2026, even though OpenAI’s ChatGPT remains far larger in consumer reach. The growth is concentrated in enterprise and in developer tooling — Claude Code alone reportedly clears $2.5 billion annualised.

Headcount has grown roughly fivefold in a year, to about 5,000 by mid-2026, making Anthropic one of the most revenue-efficient companies ever measured on a per-employee basis.

Leadership

• Dario Amodei — CEO and co-founder; former VP of Research at OpenAI.
• Daniela Amodei — President and co-founder; former VP of Operations at OpenAI.
• Sam McCandlish — Chief Technology Officer and co-founder.
• Jared Kaplan — Chief Science Officer and co-founder; scaling-laws researcher.
• Mike Krieger — Chief Product Officer; Instagram co-founder, joined 2024.
• Jack Clark — co-founder and policy lead; author of the Import AI newsletter.
• Chris Olah — co-founder; interpretability pioneer.

Competition and market position

Anthropic competes most directly with OpenAI and Google DeepMind at the frontier. Its durable edge is coding: Claude leads independent agentic-coding benchmarks (SWE-bench Verified and the harder SWE-bench Pro), and Claude Code has turned that lead into the clearest revenue proof point in the industry. Its second edge is enterprise and regulated-industry trust, where Constitutional AI and the published safety framework are easier to take to a compliance team than black-box alternatives.

Where it lags is consumer reach and multimodal breadth. ChatGPT remains the default consumer chatbot, and Anthropic still does not offer image, video or full voice generation — it stays focused on text, code and vision. Its unusual position is that it monetises far better per user than its consumer share would suggest, and is now the revenue leader despite a smaller audience.

Controversies

• Export-control suspension and redeployment (June–July 2026). On 12 June 2026 Anthropic disabled Fable 5 and Mythos 5 worldwide to comply with a US government export-control directive barring access for all foreign nationals (Anthropic). After an 18-day standoff — during which Anthropic argued the order rested on a misunderstanding and an infosec coalition urged the controls be lifted — the US Commerce Department lifted them on 30 June. Fable 5 returned to general availability from 1 July with an improved safety classifier, and Mythos 5 was cleared for approved US organizations on 26 June (Anthropic).
• Bartz v. Anthropic copyright settlement. In late August 2025 Anthropic agreed to a $1.5 billion settlement — the largest copyright settlement in US history — over training on pirated books, covering roughly 482,000 works at about $3,100 each (NPR).
• Silent model routing. Fable 5’s classifiers route certain queries to a weaker model without telling the user, meaning the model answering is sometimes not the one selected — a transparency criticism even as it is a safety feature.
• Safety versus scale. Critics on one side argue Anthropic’s rapid capability scaling contradicts its safety mission; critics on the other argue its guardrails make Claude overly cautious. The near-trillion-dollar valuation sharpens both critiques.
• Multi-cloud dependence. Billions from Amazon, Google and Microsoft — all competitors — raise recurring questions about how independent a safety lab can remain, which the Long-Term Benefit Trust is meant to answer but has not been stress-tested.

Recent developments (2026)

• Series H ($65B at a $965B valuation, May 2026) made Anthropic the world’s most valuable AI startup, ahead of OpenAI.
• Fable 5 and Mythos 5 introduced the Mythos-class tier on 9 June 2026, were suspended worldwide three days later under a US export-control directive, and returned after the controls were lifted on 30 June — Fable 5 to general availability from 1 July, Mythos 5 to trusted-access partners.
• Opus 4.6 → 4.8 brought 1M-token context to the Opus line and shifted the headline improvement toward honesty and reliability.
• Revenue overtook OpenAI, crossing a ~$47 billion run rate, with Claude Code passing $2.5 billion annualised.
• Series G ($30B at $380B, February 2026) preceded the Series H, led by GIC and Coatue.

Where Anthropic excels

• Coding. Claude leads SWE-bench Verified and SWE-bench Pro, and Claude Code is the industry’s standout AI revenue driver.
• Safety as a product. Constitutional AI, the RSP/ASL framework, and a demonstrated willingness to pull models suit regulated and risk-averse buyers.
• Long context. 1M-token windows across the Opus line enable whole-codebase and whole-corpus work in a single pass.
• Multi-cloud reach. The only frontier model available on AWS, Google Cloud and Azure, giving enterprises deployment flexibility no rival matches.

Where Anthropic falls short

• Consumer brand. ChatGPT still defines AI chatbots for most people; Claude’s reach trails its capability.
• Multimodal gaps. No image, video or full voice generation — Anthropic stays text-, code- and vision-focused.
• Availability risk. The June 2026 suspension shows that compliance and safety gates can take its best models offline overnight.
• Cost of the frontier. Mythos-class models are priced at roughly twice Opus, and silent routing on restricted topics complicates the value proposition.

Developer resources

Anthropic’s developer stack centres on the Claude API (docs.claude.com), with the Claude Developer Platform, the Anthropic SDKs, code execution, tool use, prompt caching and the Model Context Protocol for connecting Claude to external data and tools. Claude is also available through AWS Bedrock, Google Vertex AI and Azure AI Foundry, and Claude Code ships as a command-line and IDE agent for engineering teams. Pricing and model IDs live on the pricing page; service status is at status.anthropic.com.

Frequently asked questions

Who owns Anthropic?

Anthropic is a private Delaware Public Benefit Corporation. Its co-founders Dario and Daniela Amodei lead the company, and a Long-Term Benefit Trust holds special-class shares and appoints part of the board to protect the mission. Amazon, Google, Microsoft and Nvidia are major investors, but none holds a board seat, and the company is governed independently of them.

How much is Anthropic worth?

Anthropic was valued at $965 billion post-money in its Series H round in May 2026, making it the world’s most valuable AI startup, ahead of OpenAI. It has raised over $125 billion in total across eight rounds.

What is Anthropic’s latest model?

The most capable Claude is the Mythos-class Fable 5 (and its restricted twin, Mythos 5), released on 9 June 2026. Both were suspended worldwide on 12 June under a US export-control directive, but the controls were lifted on 30 June: Fable 5 returned to general availability from 1 July 2026, while Mythos 5 is restored for trusted-access partners only. For most users the mainstream flagship is Claude Opus 4.8, with Fable 5 the frontier option above it.

What is Constitutional AI?

Constitutional AI is Anthropic’s training method in which a model critiques and revises its own responses against an explicit written set of principles (a “constitution”), and a second model provides preference feedback, reducing reliance on human labellers. It is meant to make a model’s values explicit, auditable and scalable compared with standard human-feedback training.

Is Claude better than ChatGPT?

It depends on the task. Claude leads independent coding benchmarks and is widely preferred for software engineering, long-document work and safety-conscious enterprise use, while OpenAI’s ChatGPT has far greater consumer reach and broader multimodal features such as image and video generation. See ChatGPT vs Claude for a head-to-head.

How much revenue does Anthropic make?

Anthropic’s run-rate revenue crossed roughly $47 billion in May 2026, up from about $9 billion at the end of 2025 — fast enough to overtake OpenAI in revenue during 2026. Claude Code, its agentic coding tool, reportedly contributes over $2.5 billion of that on an annualised basis.

Models

Apps

• Claude