Anthropic

  Executive Briefing

Anthropic is a San Francisco-based artificial intelligence safety company and Public Benefit Corporation (PBC) founded in January 2021 by Dario Amodei (CEO), Daniela Amodei (President), and seven fellow researchers who departed OpenAI over concerns that frontier AI development was outpacing the alignment and safety work needed to make it trustworthy.¹ The company's core product is the Claude family of large language models — spanning lightweight Haiku variants to the Mythos-class frontier — and an expanding ecosystem of applications including the consumer chat service Claude.ai, the developer Claude API, and the agentic coding tool Claude Code.

The lab is distinguished by its founding mission — responsible development of AI for the long-term benefit of humanity — and by the depth with which safety is institutionalized in its research. Constitutional AI, mechanistic interpretability, and its Responsible Scaling Policy (RSP) with explicit AI Safety Level classifications are not afterthoughts bolted onto commercial work but originated at Anthropic and shape every major model release. Jared Kaplan (Chief Science Officer, co-author of neural scaling laws), Chris Olah (Interpretability Research Lead, pioneer of circuit-level neural network analysis), and Jack Clark (head of the newly launched Anthropic Institute think tank) give the organization unusual scientific and policy depth alongside its commercial operations. As of mid-2026, Rahul Patil serves as CTO (joined October 2025), Ami Vora as CPO, and Krishna Rao as CFO.

Anthropic occupies a philosophically unusual position: it acknowledges it may be building one of the most transformative and potentially dangerous technologies in history, yet presses forward on the theory that safety-focused labs at the frontier are better positioned to steer outcomes than ceding that frontier to less safety-conscious actors. The commercial results are striking — reported annualized revenue run-rate of approximately $47 billion as of May 2026,² a confidential IPO S-1 filed with the SEC on June 1, 2026,³ and a Series H valuation reported at $965 billion that briefly made it the most valuable AI startup in the world, surpassing OpenAI.⁴ At the same time the company is navigating an escalating legal and political confrontation with the US government over its refusal to allow Claude to be used for autonomous weapons or mass surveillance of US citizens.

  At a Glance

  Origins & Founding

Anthropic was incorporated on 26 January 2021 by a group of nine researchers who left OpenAI amid disagreements about the pace of commercialization relative to safety research. Dario Amodei, who had served as VP of Research at OpenAI and was among the co-authors of GPT-2 and GPT-3 research, led the departure alongside his sister Daniela Amodei, who had been VP of Operations.¹ The other co-founders were Jared Kaplan (co-author of the seminal neural scaling laws work), Chris Olah (who had pioneered neural network visualization and circuit-level interpretability), Sam McCandlish (who became Chief Architect), Tom Brown (lead author of the GPT-3 paper), Jack Clark (former OpenAI Policy Director), Ben Mann, and Amanda Askell (moral philosophy research lead).

The founding group was especially concerned that large language models were being deployed at increasing scale without the interpretability and alignment tools necessary to understand or control them. Structuring Anthropic as a Public Benefit Corporation under Delaware law — rather than a standard C-corp or a non-profit — was a deliberate legal commitment to a declared social purpose, binding the company to weigh both profit and public benefit in decision-making. A long-term benefit trust (LTBT) structure was layered on top to insulate the mission against purely commercial pressures even as outside capital scaled up.¹ The founding charter committed specifically to developing AI systems that are reliable, interpretable, and steerable.

A seed round of $124 million closed in May 2021, giving the founders the runway to begin pre-training Claude 1.0 and publish foundational safety research — most consequentially, the Constitutional AI paper released in December 2022 — before the first public model launch in March 2023.

  History & Timeline

    2021–2022: Founding, early research, and Constitutional AI

The first two years were primarily a research and model-building phase. Anthropic recruited rapidly from academic and industry safety communities, with Olah's interpretability team establishing early results on features and circuits in transformer networks. Internally, the team completed training of Claude 1.0 during summer 2022. The December 2022 publication of the Constitutional AI paper on arXiv marked Anthropic's first major scientific contribution: a method for training models to critique and revise their own outputs according to a written constitution, using AI feedback (RLAIF) rather than solely human labeling.⁵ This substantially reduced the human-annotation labor required for alignment fine-tuning and became the methodological backbone for all subsequent Claude training.

The April 2022 Series B of $580 million included a $500 million commitment from FTX's Sam Bankman-Fried — a figure that drew significant reputational scrutiny after FTX's collapse in November 2022. Anthropic subsequently sold the FTX stake back to the bankruptcy estate, though the precise settlement terms have not been fully disclosed.¹

    2023: Public launch, Google investment, and Amazon as cloud anchor

Claude and the lighter Claude Instant launched publicly in March 2023, immediately positioning Anthropic as the primary competitor to OpenAI's ChatGPT in the enterprise and API markets. Google made its first investment in Anthropic as part of the Series C in May 2023. Claude 2 followed in July 2023 with a then-unprecedented 100,000-token context window, nearly doubling the practical working memory of any deployed frontier model; Claude 2.1 in November 2023 extended this further to 200,000 tokens.⁶

Amazon announced an initial $1.25 billion investment in September 2023 and was named Anthropic's primary cloud provider, with Anthropic committing to train and serve models on AWS Trainium and Amazon Bedrock. Music publishers filed a copyright lawsuit in October 2023 over alleged lyric infringement in Claude's outputs — the first of several legal actions that would escalate in subsequent years.

    2024: The Claude 3 breakthrough, Computer Use, and Amazon's deepening bet

The Claude 3 family — Opus, Sonnet, and Haiku — launched on 4 March 2024 and reset the competitive conversation: Claude 3 Opus exceeded GPT-4 on a range of key benchmarks at launch, the first time a non-OpenAI model had achieved this across the board.⁶ Mike Krieger, co-founder of Instagram, joined as Chief Product Officer in May 2024. Amazon invested an additional $2.75 billion in March 2024, bringing its committed total to $4 billion.

Claude 3.5 Sonnet launched on 20 June 2024 with a result that surprised even inside Anthropic: it outperformed Opus on most benchmarks at a fraction of the cost and introduced the Artifacts feature — an in-chat canvas for running, editing, and sharing live code, documents, and visualizations. The October 2024 release of Claude 3.5 Haiku and an updated Claude 3.5 Sonnet added Computer Use capability, enabling the model to operate desktop software autonomously via screenshots and simulated keyboard/mouse input. Amazon committed a further $4 billion in November 2024, bringing its cumulative investment to approximately $8 billion.

An authors' class-action filed in August 2024 alleged that Anthropic trained Claude on pirated books obtained through "Project Panama" — a program that court documents alleged involved destructively scanning millions of used books — significantly escalating the legal exposure from copyright suits.

    2025: Reasoning models, Claude 4, and Series E / F financings

Claude 3.7 Sonnet launched on 24 February 2025 with extended thinking — a chain-of-thought deliberation mode allowing the model to reason through complex problems step-by-step before producing a final answer, Anthropic's entry into the reasoning-model category that OpenAI had pioneered with o1.⁶ Claude Code, a command-line agentic coding assistant, launched in February 2025 and quickly became one of Anthropic's fastest-growing products, reaching a reported $2.5 billion+ annualized run-rate by early 2026.

The Claude 4 generation launched on 22 May 2025: Claude Opus 4 became the first model in the industry to receive a Level 3 classification under Anthropic's Responsible Scaling Policy, signaling that it had crossed the capability threshold requiring enhanced safety evaluations and deployment restrictions.⁷ Claude Sonnet 4 released alongside it. A Series E of $3.5 billion at a $61.5 billion post-money valuation closed in March 2025.

Subsequent Claude 4 sub-releases followed rapidly: Claude Opus 4.1 (August 2025) introduced the ability to end conversations that remain persistently harmful; Claude Sonnet 4.5 (September 2025) and Claude Haiku 4.5 (October 2025) refreshed the mid-tier and budget tiers. Claude Opus 4.5 (November 2025) introduced the Infinite Chats context-overflow feature. A Series F of $13 billion at a $183 billion post-money valuation, co-led by Fidelity Management and Lightspeed Venture Partners, closed in September 2025. A $1.5 billion settlement with the class-action authors concluded in September 2025.⁸ A June 2025 court ruling held that Anthropic's purchase and scanning of books for training constituted fair use, but that obtaining pirated copies did not.⁸

Chinese government-sponsored hackers used the Claude API in November 2025 to conduct automated cyberattacks against approximately 30 organizations — an incident that highlighted both the dual-use risks of capable AI systems and the difficulty of preventing API abuse at scale.

    2026: Series G / H, the Pentagon dispute, Mythos class, and IPO filing

The pace of developments in 2026 has been extraordinary. A Series G of $30 billion at $380 billion post-money closed in February 2026, co-led by GIC and Coatue with participation from D.E. Shaw, Dragoneer, Founders Fund, ICONIQ, and MGX.⁹ In the same month Anthropic ran a Super Bowl advertising campaign, donated $20 million to Public First Action for civic research on AI governance, and co-founders collectively pledged 80% of personal wealth (estimated at over $21 billion) to philanthropic efforts addressing AI-driven inequality.¹

Claude Opus 4.6 launched on 5 February 2026 with agent teams orchestration, enabling multiple Claude agents to coordinate autonomously on complex workflows. Claude Cowork — a GUI knowledge-work product competing with Notion, Google Workspace, and Microsoft 365 — launched in January 2026, with Anthropic reporting it shipped over 30 products that month alone. Claude Design (a visual creation tool) launched in April 2026. Claude Opus 4.7 and Claude Opus 4.8 followed in April and May respectively, with Opus 4.8 featuring a Dynamic Workflows research preview for parallel sub-agent orchestration.

The Anthropic Institute, a policy and governance think tank led by co-founder Jack Clark, launched in March 2026. Anthropic Labs — an internal product incubator co-led by Mike Krieger and Ben Mann — was also formalized in 2026 to develop frontier agentic enterprise applications.

The company's most consequential confrontation came in early 2026. After Anthropic refused to allow Claude to be used for autonomous weapons programs or mass surveillance of US citizens, the Pentagon designated Anthropic a "supply chain risk" in February–March 2026. Anthropic filed two federal lawsuits against the Trump administration in March 2026, alleging First Amendment retaliation; a federal judge in San Francisco stated at a March 24 hearing that the government ban appeared punitive, and a preliminary injunction process began.¹⁰ Employees of OpenAI and Google publicly backed Anthropic in the dispute.

Claude Fable 5 and Claude Mythos 5 — the first models in Anthropic's new highest-tier "Mythos class" — launched on 9 June 2026.¹¹ Access was suspended on 12 June 2026 per a US government directive restricting availability for foreign nationals; the precise legal basis and expected duration of the suspension remain unclear.¹¹ A Series H of $65 billion at a $965 billion post-money valuation led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital closed in May 2026, briefly making Anthropic the most valuable AI startup in the world.⁴ Anthropic confidentially filed a draft S-1 with the SEC on 1 June 2026, targeting a public listing in late 2026.³

  Mission, Philosophy & Research Agenda

Anthropic's stated mission is the responsible development and maintenance of advanced AI for the long-term benefit of humanity. The company's philosophical position is openly paradoxical: it acknowledges that building ever-more-capable AI carries grave risks, while arguing that the best chance of a good outcome is for safety-focused organizations to remain at the frontier rather than cede it to developers less committed to alignment.¹ This thesis drives its decision to raise unprecedented capital and push for state-of-the-art capability, even as its research agenda centers on understanding and constraining that capability.

Four pillars define Anthropic's research agenda. First, mechanistic interpretability — Chris Olah's team pursues understanding of what is actually happening inside neural networks at the level of circuits, features, and attention heads, publishing detailed analyses of transformer internals that no other frontier lab matches in depth. Second, Constitutional AI and scalable alignment — the use of AI feedback (RLAIF) rather than purely human labeling to train models according to a written constitution of values, reducing labor costs and enabling more consistent alignment at scale.⁵ Third, AI safety evaluations — Anthropic's Responsible Scaling Policy (RSP) classifies models at AI Safety Levels (ASL-1 through ASL-4), committing the company to specific safety evaluations and deployment restrictions before advancing past each threshold; Claude Opus 4 was the first industry model to reach Level 3 in May 2025.⁷ Fourth, policy and governance, institutionalized via the Anthropic Institute and an active government-engagement program — a domain where Anthropic has been more publicly assertive than any other frontier lab.

  Research & Publications

Anthropic's research output has established several foundational techniques for frontier AI safety and alignment:

• Constitutional AI (December 2022, arXiv) — introduced Reinforcement Learning from AI Feedback (RLAIF), in which a model critiques and revises its own outputs against a written constitution of principles, substantially reducing reliance on human labeling for alignment training.⁵ Now the industry's most widely adopted alternative to pure RLHF.
• Scaling Laws for Neural Language Models — Jared Kaplan co-authored this landmark work (while at OpenAI prior to Anthropic's founding) establishing the power-law relationships between model size, data volume, compute, and performance; this remains foundational to how Anthropic and the wider field plan training runs.
• Mechanistic Interpretability — Chris Olah's team has published landmark work on circuits within transformers: identifying polysemantic neurons, mapping attention heads to interpretable functions, and using sparse autoencoders to decompose model layers into human-readable features. The 2024–2025 Alignment Science Blog series documented these findings across model generations.¹
• Collective Constitutional AI (2024) — extended Constitutional AI to incorporate public deliberation processes for value alignment, using structured citizen input to shape the constitution itself.
• Responsible Scaling Policy (RSP) — Anthropic's internal safety framework, published externally, provides a model for how frontier labs can commit to capability-conditioned safety investments; it is regularly cited in AI governance discussions and has influenced proposed regulatory frameworks.
• Superalignment / Scalable Oversight — led by Jan Leike (who joined from OpenAI in 2024), this research line develops techniques for using capable AI models to evaluate models that exceed human judgment in narrow domains.
• Dynamic Workflows (May 2026 research preview) — parallel sub-agent orchestration within Claude Code, enabling multiple specialized model instances to cooperate on complex engineering tasks.

  Models & Products

Anthropic releases models across three capability tiers — Haiku (fast and lightweight), Sonnet (balanced), and Opus (most capable) — with the new Mythos class above Opus for frontier deployments. All models access the Claude API; consumer users interact through Claude.ai.

Claude 1 family (2023)

• Claude 1 (March 2023) — initial public release.
• Claude Instant (March 2023) — lighter, faster variant for high-throughput applications.

Claude 2 family (2023)

• Claude 2 (July 2023) — 100,000-token context window, substantially improved reasoning.⁶
• Claude 2.1 (November 2023) — extended to 200,000 tokens.

Claude 3 family (2024)

• Claude 3 Opus (March 2024) — exceeded GPT-4 on key benchmarks at launch.
• Claude 3 Sonnet (March 2024) — balanced performance and cost.
• Claude 3 Haiku (March 2024) — fastest and most affordable.

Claude 3.5 family (2024)

• Claude 3.5 Sonnet (June 2024) — outperformed Opus on most benchmarks; introduced Artifacts.
• Claude 3.5 Sonnet v2 (October 2024) — added Computer Use capability.
• Claude 3.5 Haiku (October 2024) — refreshed budget tier.

Claude 3.7 (2025)

• Claude 3.7 Sonnet (February 2025) — introduced extended thinking / chain-of-thought reasoning mode.

Claude 4 family (2025–2026)

• Claude Opus 4 (May 2025) — first model to reach RSP Level 3 classification.⁷
• Claude Sonnet 4 (May 2025).
• Claude Opus 4.1 (August 2025) — can terminate persistently harmful conversations.
• Claude Sonnet 4.5 (September 2025).
• Claude Haiku 4.5 (October 2025).
• Claude Opus 4.5 (November 2025) — Infinite Chats context-overflow.
• Claude Opus 4.6 (February 2026) — agent teams orchestration.
• Claude Sonnet 4.6 (February 2026).
• Claude Opus 4.7 (April 2026).
• Claude Opus 4.8 (May 2026) — Dynamic Workflows research preview; current standard flagship.

Mythos class (2026)

• Claude Fable 5 (9 June 2026, access suspended 12 June 2026) — first Mythos-class model available to the public; priced at $10/M input tokens, $50/M output tokens; includes safety guardrails.¹¹
• Claude Mythos 5 (9 June 2026, access suspended 12 June 2026) — Fable 5 with guardrails removed for approved users under Project Glasswing; subject to the same government-directed suspension.¹¹

Products and surfaces

• Claude.ai — consumer web and mobile chat interface, with iOS, Android, Mac, and Windows apps. Subscription tiers: Free, Pro, Max, Team, Enterprise.
• Claude API — developer and enterprise API with pay-per-token pricing and volume discounts.
• Claude Code — command-line agentic coding assistant (February 2025); web GUI version (October 2025); reported $2.5B+ annualized run-rate by early 2026; holds approximately 42% of the AI coding-tools market.
• Claude Cowork — GUI knowledge-work product for document creation and collaboration (January 2026).
• Claude Design — visual creation and design tool (April 2026).
• Anthropic Institute — external policy and governance think tank (March 2026), led by Jack Clark.
• Anthropic Labs — internal product incubator for frontier agentic applications (2026), co-led by Mike Krieger and Ben Mann.

➡️ See individual model cards for specifications, pricing, and benchmark results.

  People

Anthropic's leadership spans its nine co-founders — all of whom remain active in the company — plus a set of senior hires brought in to scale commercial operations. Dario Amodei sets technical vision and external strategy; Daniela Amodei oversees commercial, go-to-market, and people functions. Jared Kaplan as CSO directs the scientific program; Chris Olah leads interpretability research; Sam McCandlish as Chief Architect leads pre-training. Jack Clark is head of policy and leads the Anthropic Institute. Ben Mann co-leads Anthropic Labs, the internal product incubator.

Significant recent hires include Rahul Patil (CTO, joined October 2025, formerly CTO at Stripe), Ami Vora (CPO, 2026, formerly a senior executive at Meta), and Krishna Rao (CFO, joined May 2024). Mike Krieger, co-founder of Instagram, joined as CPO in May 2024 and transitioned to co-lead Anthropic Labs in 2026 alongside Ben Mann. Jan Leike joined in 2024 as alignment science co-lead, having departed OpenAI's Superalignment team citing concerns about safety culture there.¹² Board members include Reed Hastings (Netflix co-founder) and Chris Liddell (former Microsoft and GM executive).

Anthropic is a significant destination for AI safety researchers who prioritize interpretability and alignment over pure capability maximization, and has drawn talent from OpenAI, Google DeepMind, and academia. The company does not publish a headcount figure but is estimated to employ over 3,000 people as of mid-2026.

  Funding, Ownership & Business

Anthropic's public benefit corporation structure is central to its identity — the PBC charter legally binds the company to weigh public benefit alongside shareholder returns, and the long-term benefit trust was designed to preserve mission alignment even as outside capital scaled into the tens of billions. In practice, capital has scaled at extraordinary speed: the company has raised approximately $132 billion across multiple rounds as of mid-2026 (reported).⁴

Key financing milestones:

Revenue is generated primarily through API consumption (pay-per-token pricing), consumer and enterprise subscriptions (Pro, Max, Team, Enterprise seat-based), and large enterprise contracts. Reported annualized revenue run-rate stood at approximately $47 billion as of May 2026, up from an estimated $9–10 billion at end of 2025 and approximately $4 billion in July 2025 — representing one of the fastest revenue ramps in enterprise software history (all figures reported and unaudited; the S-1 has not been made public).² Over 1,000 customers reportedly spend more than $1 million per year with Anthropic; as of February 2026, eight of the Fortune 10 companies are Claude customers.²

An Amazon decision in April 2026 to commit to up to an additional $25 billion in Anthropic further cements the relationship — Amazon is simultaneously Anthropic's largest investor, primary cloud infrastructure provider, and a major distribution channel through Amazon Bedrock.¹³ The $100 billion ten-year Anthropic commitment to AWS is a stated intent rather than a legally confirmed contract figure and should be treated as such.

  Partnerships & Ecosystem

Anthropic's infrastructure and commercial partnerships are as consequential as its financing relationships. Amazon Web Services is the anchor: Anthropic trains on AWS Trainium2 and Trainium3 chips, serves inference through Bedrock, and in April 2026 Amazon announced a commitment to invest up to an additional $25 billion in a compute infrastructure deal targeting up to 5 gigawatts of capacity.¹³ Over 100,000 customers run Claude on Amazon Bedrock.

Google Cloud is a strategic partner and investor, with a multi-billion-dollar TPU training partnership and a major cloud deal announced in October 2025. A joint Google–Broadcom deal for next-generation TPU capacity at gigawatt scale was announced in 2026, with capacity expected online from 2027.¹⁴ Claude Fable 5 is available on Google Vertex AI. Claude Fable 5 is also available through Microsoft Foundry, signaling meaningful Azure integration despite Microsoft's deep ties to OpenAI.

Beyond hyperscalers, Anthropic signed a $200 million multi-year data partnership with Snowflake in December 2025. A $1.8 billion cloud computing deal with Akamai was announced in May 2026. Anthropic provides Claude to the US intelligence and defense community via Palantir on AWS classified infrastructure. In May 2026 Anthropic secured access to xAI's Colossus supercomputer for additional training capacity. Anthropic simultaneously trains on AWS Trainium, Google TPUs, and NVIDIA GPUs — a deliberate multi-cloud strategy designed to avoid single-vendor dependency and create competition for its compute supply.

  Compute & Infrastructure

Compute is Anthropic's largest operational constraint and central strategic investment. The company's primary training infrastructure runs on AWS Trainium2 chips; Trainium2 capacity came online in the first half of 2026, with approximately one gigawatt of additional Trainium2/3 capacity expected by end of 2026.¹³ Google's TPUs form the secondary training substrate, with the Google–Broadcom gigawatt-scale TPU deal creating a longer-dated supply line through 2027 and beyond.¹⁴

Anthropic's multi-cloud training posture — AWS Trainium, Google TPUs, and NVIDIA GPUs running concurrently — is unusual at this scale and reflects both practical supply constraints and strategic optionality. The xAI Colossus supercomputer access (May 2026) and the Akamai cloud deal add further capacity outside the two hyperscaler relationships. Specific hardware configurations, cluster sizes, and training FLOPs for Mythos-class models have not been publicly disclosed.

  Notable Events & Controversies

Anthropic's rapid ascent has been accompanied by significant legal, political, and reputational episodes.

FTX investment (2022): $500 million of the April 2022 Series B came from FTX and Sam Bankman-Fried. When FTX collapsed in November 2022, Anthropic faced reputational questions about having taken the funds. The company subsequently sold the stake back to the FTX bankruptcy estate; exact settlement terms are not fully disclosed.¹

Copyright litigation: Music publishers sued in October 2023 over alleged lyric reproduction in Claude outputs. An authors' class-action filed in August 2024 alleged that Anthropic trained on pirated books through "Project Panama," described in court documents as involving the destructive scanning of millions of purchased used books. A June 2025 court ruling found that purchasing and scanning books for training constituted fair use, but that obtaining pirated copies did not. Anthropic reached a $1.5 billion settlement with the class-action authors in September 2025.⁸

Reddit lawsuit (June 2025): Reddit filed a suit alleging unauthorized training on user-generated content. The status and outcome of this case as of June 2026 remain unconfirmed.

Chinese cyberattack via Claude API (November 2025): Chinese government-linked hackers reportedly used the Claude API to conduct automated cyberattacks against approximately 30 organizations, highlighting the challenge of preventing adversarial API use at scale.

Pentagon supply-chain designation and federal litigation (2026): After Anthropic declined to permit Claude for autonomous weapons programs or mass surveillance of US citizens, the US Department of Defense designated Anthropic a "supply chain risk" in early 2026. Anthropic filed two federal lawsuits against the Trump administration in March 2026, alleging the designation constituted First Amendment retaliation for the company's public policy statements.¹⁰ A federal judge in San Francisco stated at a March 24, 2026 hearing that the government ban appeared punitive, and a preliminary injunction process was underway. Employees of OpenAI and Google publicly backed Anthropic in the dispute.

Fable 5 / Mythos 5 access suspension (June 2026): Claude Fable 5 and Claude Mythos 5 were released on June 9, 2026 and access was suspended three days later on June 12, 2026 per a US government directive restricting access for foreign nationals. The precise legal basis, the identity of the issuing authority, and the expected duration of the suspension have not been made public.¹¹

Co-founders philanthropy pledge (January 2026): All seven co-founders publicly pledged 80% of their personal wealth — estimated collectively at over $21 billion — to philanthropic efforts addressing economic inequality driven by AI automation.¹

  Competitive Position

Anthropic is one of three recognized frontier AI labs alongside OpenAI and Google DeepMind, and as of mid-2026 has staked a significant claim to commercial leadership. By annualized revenue run-rate, Anthropic reportedly surpassed OpenAI in early-to-mid 2026 (~$47B vs a reported ~$20B for OpenAI), making it the largest pure-play AI company by revenue — though these are unaudited, company-reported figures and the gap should be treated with caution.²

Consumer market share tells a different story: ChatGPT holds an estimated 57–68% of the US AI chatbot market, Gemini approximately 18–25%, and Claude approximately 3–4%. However, in enterprise and developer segments Anthropic has moved substantially ahead: Claude holds approximately 42% of the AI coding-tools market, driven by Claude Code, and the share of US businesses paying for Anthropic reportedly rose from approximately 4% to approximately 25% in a single year.²

Anthropic's principal differentiators are its safety-first positioning — which resonates with regulated industries requiring compliance-grade AI — its Constitutional AI and RSP framework (credible with enterprise legal and risk teams), its long context windows, and Claude's consistently strong coding and reasoning performance. Google DeepMind leads on benchmark totals (Gemini 3.1 Pro tops the majority of major benchmarks as of mid-2026) but trails in enterprise revenue. OpenAI retains consumer brand dominance and the Microsoft distribution advantage. The open-weight model ecosystem (DeepSeek, Mistral, Meta Llama) exerts pricing pressure on the API business.

At the reported $965 billion Series H valuation Anthropic briefly became the most valuable AI startup globally, overtaking OpenAI.⁴ Whether that valuation is sustained through the public listing remains the defining near-term question.

  Outlook & Roadmap

Anthropic filed a confidential draft S-1 with the SEC on 1 June 2026, targeting a public listing in late 2026 at approximately the $965 billion reported valuation.³ The IPO timeline and exact pricing remain unconfirmed; regulatory scrutiny over the Pentagon dispute and the Fable/Mythos suspension creates meaningful uncertainty.

Revenue run-rate is approximately $47 billion as of May 2026 and is expected — by the company's own reported projections — to exceed $50 billion by end of June 2026 (unaudited; treat as reported estimates).² Infrastructure buildout through H2 2026 adds Trainium2/3 capacity on AWS, with the Google–Broadcom gigawatt-scale TPU deal providing a longer supply runway from 2027 onward.

On the model roadmap, the Mythos class represents the new highest frontier tier, though Fable 5 and Mythos 5 remain suspended pending resolution of the government access directive. Anthropic Labs, co-led by Mike Krieger and Ben Mann, is incubating the next generation of agentic enterprise products following the early traction of Claude Code and Cowork. The company is building toward ASL-4 safety evaluations as it approaches the capability thresholds that its Responsible Scaling Policy defines as requiring the most stringent pre-deployment review — a threshold no lab has yet reached.

Key strategic risks include the unresolved Pentagon dispute and potential government restrictions on model deployment, regulatory and investor scrutiny ahead of the IPO, customer concentration within the Amazon and Google relationships (which are simultaneously investor, infrastructure, and distribution), and the possibility that rival labs — or open-weight models — close the capability gap before Anthropic reaches public markets.

---

  References

1. Anthropic — Wikipedia
2. Anthropic surpasses OpenAI in revenue — CNBC, May 2026
3. Anthropic confidential S-1 filing, June 2026
4. Anthropic Series H and valuation — CNBC / Bloomberg, May 2026
5. Constitutional AI: Harmlessness from AI Feedback — Anthropic Research
6. Claude (language model) — Wikipedia
7. Responsible Scaling Policy and ASL-3 classification — Anthropic
8. Authors class-action settlement — Wikipedia: Anthropic
9. Anthropic Series G $30B announcement
10. Anthropic sues Pentagon over supply-chain designation — NPR, March 2026
11. Claude Fable 5 and Mythos 5 announcement — TechCrunch / Anthropic
12. Jan Leike joins Anthropic — OpenAI departures, 2024
13. Amazon commits up to $25B additional investment — CNBC, April 2026
14. Anthropic Google and Broadcom compute partnership

  References

1. Founding, PBC structure, FTX handling, and co-founder pledge — Wikipedia: Anthropic. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

2. Revenue run-rate and enterprise market share figures are company-reported estimates, unaudited — CNBC, May 2026. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

3. Confidential S-1 filing targeting late 2026 IPO — Wikipedia: Anthropic. ↩ ↩² ↩³

4. Series H $65B at $965B post-money valuation — CNBC, May 2026. ↩ ↩² ↩³ ↩⁴ ↩⁵

5. Constitutional AI paper (December 2022) — Anthropic Research. ↩ ↩² ↩³

6. Claude model generations and context window milestones — Wikipedia: Claude. ↩ ↩² ↩³ ↩⁴

7. RSP Level 3 classification for Claude Opus 4 — Anthropic RSP. ↩ ↩² ↩³

8. Copyright litigation settlement ($1.5B, September 2025) and June 2025 fair-use ruling — Wikipedia: Anthropic. ↩ ↩² ↩³

9. Series G $30B at $380B — Anthropic Series G announcement. ↩

10. Pentagon designation and federal lawsuits — NPR, March 2026. ↩ ↩²

11. Claude Fable 5 and Mythos 5 release and suspension — TechCrunch, June 2026; Anthropic platform docs. ↩ ↩² ↩³ ↩⁴ ↩⁵

12. Jan Leike departure from OpenAI citing safety culture, joins Anthropic — Wikipedia: Anthropic. ↩

13. Amazon up to $25B additional investment and compute deal — CNBC, April 2026; Anthropic AWS compute announcement. ↩ ↩² ↩³

14. Google and Broadcom gigawatt-scale TPU deal — Anthropic announcement. ↩ ↩²